Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
// Copyright 2018-2019 - Omar Sandoval
|
|
|
|
// SPDX-License-Identifier: GPL-3.0+
|
|
|
|
|
|
|
|
#include <ctype.h>
|
|
|
|
#include <float.h>
|
|
|
|
#include <inttypes.h>
|
|
|
|
#include <limits.h>
|
|
|
|
#include <math.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
|
|
|
|
#include "internal.h"
|
|
|
|
#include "hash_table.h"
|
|
|
|
#include "language.h"
|
|
|
|
#include "lexer.h"
|
|
|
|
#include "memory_reader.h"
|
|
|
|
#include "object.h"
|
|
|
|
#include "program.h"
|
|
|
|
#include "string_builder.h"
|
|
|
|
#include "type.h"
|
|
|
|
#include "type_index.h"
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_variable(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb);
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_define_type(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb);
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
static bool append_tabs(int n, struct string_builder *sb)
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
{
|
|
|
|
while (n-- > 0) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '\t'))
|
|
|
|
return false;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
return true;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_variable_name(struct string_callback *name,
|
|
|
|
void *arg, struct string_builder *sb)
|
|
|
|
{
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, arg))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_append_qualifiers(enum drgn_qualifiers qualifiers,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
static const char *qualifier_names[] = {
|
|
|
|
"const", "volatile", "restrict", "_Atomic",
|
|
|
|
};
|
|
|
|
bool first = true;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
static_assert((1 << ARRAY_SIZE(qualifier_names)) - 1 ==
|
|
|
|
DRGN_ALL_QUALIFIERS, "missing C qualifier name");
|
|
|
|
|
|
|
|
for (i = 0; (1U << i) & DRGN_ALL_QUALIFIERS; i++) {
|
|
|
|
if (!(qualifiers & (1U << i)))
|
|
|
|
continue;
|
|
|
|
if (!first) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, qualifier_names[i]))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
first = false;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_basic(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent, sb))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (qualified_type.qualifiers) {
|
|
|
|
err = c_append_qualifiers(qualified_type.qualifiers, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb,
|
|
|
|
drgn_type_kind(qualified_type.type) == DRGN_TYPE_VOID ?
|
|
|
|
"void" : drgn_type_name(qualified_type.type)))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (name) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
err = string_callback_call(name, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_append_tagged_name(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
const char *keyword, *tag;
|
|
|
|
|
|
|
|
switch (drgn_type_kind(qualified_type.type)) {
|
|
|
|
case DRGN_TYPE_STRUCT:
|
|
|
|
keyword = "struct";
|
|
|
|
break;
|
|
|
|
case DRGN_TYPE_UNION:
|
|
|
|
keyword = "union";
|
|
|
|
break;
|
2019-11-15 01:12:47 +00:00
|
|
|
case DRGN_TYPE_CLASS:
|
|
|
|
keyword = "class";
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case DRGN_TYPE_ENUM:
|
|
|
|
keyword = "enum";
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent, sb))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (qualified_type.qualifiers) {
|
|
|
|
err = c_append_qualifiers(qualified_type.qualifiers, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, keyword))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
tag = drgn_type_tag(qualified_type.type);
|
|
|
|
if (tag) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' ') ||
|
|
|
|
!string_builder_append(sb, tag))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_tagged(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
|
|
|
if (drgn_type_is_anonymous(qualified_type.type))
|
|
|
|
err = c_define_type(qualified_type, indent, sb);
|
|
|
|
else
|
|
|
|
err = c_append_tagged_name(qualified_type, indent, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (name) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
err = string_callback_call(name, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_pointer_name(struct string_callback *name,
|
|
|
|
void *arg, struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_qualified_type *qualified_type = arg;
|
|
|
|
struct drgn_qualified_type referenced_type;
|
|
|
|
enum drgn_type_kind referenced_kind;
|
|
|
|
bool parenthesize;
|
|
|
|
|
|
|
|
referenced_type = drgn_type_type(qualified_type->type);
|
|
|
|
referenced_kind = drgn_type_kind(referenced_type.type);
|
|
|
|
parenthesize = (referenced_kind == DRGN_TYPE_ARRAY ||
|
|
|
|
referenced_kind == DRGN_TYPE_FUNCTION);
|
2019-05-14 18:07:50 +01:00
|
|
|
if (parenthesize && !string_builder_appendc(sb, '('))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '*'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (qualified_type->qualifiers) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
err = c_append_qualifiers(qualified_type->qualifiers, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (name) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
err = string_callback_call(name, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (parenthesize && !string_builder_appendc(sb, ')'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_pointer(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct string_callback pointer_name = {
|
|
|
|
.fn = c_pointer_name,
|
|
|
|
.str = name,
|
|
|
|
.arg = &qualified_type,
|
|
|
|
};
|
|
|
|
struct drgn_qualified_type referenced_type;
|
|
|
|
|
|
|
|
referenced_type = drgn_type_type(qualified_type.type);
|
|
|
|
return c_declare_variable(referenced_type, &pointer_name, indent, sb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_array_name(struct string_callback *name, void *arg,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_qualified_type *qualified_type = arg;
|
|
|
|
|
|
|
|
err = string_callback_call(name, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (drgn_type_is_complete(qualified_type->type)) {
|
|
|
|
uint64_t length = drgn_type_length(qualified_type->type);
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "[%" PRIu64 "]", length))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, "[]"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_array(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct string_callback array_name = {
|
|
|
|
.fn = c_array_name,
|
|
|
|
.str = name,
|
|
|
|
.arg = &qualified_type,
|
|
|
|
};
|
|
|
|
struct drgn_qualified_type element_type;
|
|
|
|
|
|
|
|
element_type = drgn_type_type(qualified_type.type);
|
|
|
|
return c_declare_variable(element_type, &array_name, indent, sb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_function(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_type_parameter *parameters;
|
|
|
|
size_t num_parameters, i;
|
|
|
|
struct drgn_qualified_type return_type;
|
|
|
|
|
|
|
|
if (!name) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"function must have name");
|
|
|
|
}
|
|
|
|
|
|
|
|
parameters = drgn_type_parameters(qualified_type.type);
|
|
|
|
num_parameters = drgn_type_num_parameters(qualified_type.type);
|
|
|
|
|
|
|
|
return_type = drgn_type_type(qualified_type.type);
|
|
|
|
err = c_declare_variable(return_type, name, indent, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '('))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
for (i = 0; i < num_parameters; i++) {
|
|
|
|
const char *parameter_name = parameters[i].name;
|
|
|
|
struct drgn_qualified_type parameter_type;
|
|
|
|
struct string_callback name_cb = {
|
|
|
|
.fn = c_variable_name,
|
|
|
|
.arg = (void *)parameter_name,
|
|
|
|
};
|
|
|
|
|
|
|
|
err = drgn_parameter_type(¶meters[i], ¶meter_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (i > 0) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, ", "))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
err = c_declare_variable(parameter_type,
|
|
|
|
parameter_name && parameter_name[0] ?
|
|
|
|
&name_cb : NULL, 0, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
if (num_parameters && drgn_type_is_variadic(qualified_type.type)) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, ", ..."))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else if (!num_parameters &&
|
|
|
|
!drgn_type_is_variadic(qualified_type.type)) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, "void"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ')'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_declare_variable(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_callback *name, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
switch (drgn_type_kind(qualified_type.type)) {
|
|
|
|
case DRGN_TYPE_VOID:
|
|
|
|
case DRGN_TYPE_INT:
|
|
|
|
case DRGN_TYPE_BOOL:
|
|
|
|
case DRGN_TYPE_FLOAT:
|
|
|
|
case DRGN_TYPE_COMPLEX:
|
|
|
|
case DRGN_TYPE_TYPEDEF:
|
|
|
|
return c_declare_basic(qualified_type, name, indent, sb);
|
|
|
|
case DRGN_TYPE_STRUCT:
|
|
|
|
case DRGN_TYPE_UNION:
|
2019-11-15 01:12:47 +00:00
|
|
|
case DRGN_TYPE_CLASS:
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case DRGN_TYPE_ENUM:
|
|
|
|
return c_declare_tagged(qualified_type, name, indent, sb);
|
|
|
|
case DRGN_TYPE_POINTER:
|
|
|
|
return c_declare_pointer(qualified_type, name, indent, sb);
|
|
|
|
case DRGN_TYPE_ARRAY:
|
|
|
|
return c_declare_array(qualified_type, name, indent, sb);
|
|
|
|
case DRGN_TYPE_FUNCTION:
|
|
|
|
return c_declare_function(qualified_type, name, indent, sb);
|
|
|
|
}
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_define_compound(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_type_member *members;
|
|
|
|
size_t num_members, i;
|
|
|
|
|
|
|
|
if (!drgn_type_is_complete(qualified_type.type)) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"cannot get definition of incomplete compound type");
|
|
|
|
}
|
|
|
|
|
|
|
|
members = drgn_type_members(qualified_type.type);
|
|
|
|
num_members = drgn_type_num_members(qualified_type.type);
|
|
|
|
|
|
|
|
err = c_append_tagged_name(qualified_type, indent, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, " {\n"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
for (i = 0; i < num_members; i++) {
|
|
|
|
const char *member_name = members[i].name;
|
|
|
|
struct drgn_qualified_type member_type;
|
|
|
|
struct string_callback name_cb = {
|
|
|
|
.fn = c_variable_name,
|
|
|
|
.arg = (void *)member_name,
|
|
|
|
};
|
|
|
|
|
|
|
|
err = drgn_member_type(&members[i], &member_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = c_declare_variable(member_type,
|
|
|
|
member_name && member_name[0] ?
|
|
|
|
&name_cb : NULL, indent + 1, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (members[i].bit_field_size) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, " : %" PRIu64,
|
|
|
|
members[i].bit_field_size))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, ";\n"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent, sb) || !string_builder_appendc(sb, '}'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_define_enum(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
const struct drgn_type_enumerator *enumerators;
|
|
|
|
size_t num_enumerators, i;
|
|
|
|
bool is_signed;
|
|
|
|
|
|
|
|
if (!drgn_type_is_complete(qualified_type.type)) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"cannot get definition of incomplete enum type");
|
|
|
|
}
|
|
|
|
|
|
|
|
enumerators = drgn_type_enumerators(qualified_type.type);
|
|
|
|
num_enumerators = drgn_type_num_enumerators(qualified_type.type);
|
|
|
|
|
|
|
|
err = c_append_tagged_name(qualified_type, indent, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, " {\n"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
is_signed = drgn_enum_type_is_signed(qualified_type.type);
|
|
|
|
for (i = 0; i < num_enumerators; i++) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent + 1, sb) ||
|
|
|
|
!string_builder_append(sb, enumerators[i].name) ||
|
|
|
|
!string_builder_append(sb, " = "))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (is_signed) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRId64 ",\n",
|
|
|
|
enumerators[i].svalue))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRIu64 ",\n",
|
|
|
|
enumerators[i].uvalue))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent, sb) || !string_builder_appendc(sb, '}'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_define_typedef(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct string_callback typedef_name = {
|
|
|
|
.fn = c_variable_name,
|
|
|
|
.arg = (char *)drgn_type_name(qualified_type.type),
|
|
|
|
};
|
|
|
|
struct drgn_qualified_type aliased_type;
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!append_tabs(indent, sb))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (qualified_type.qualifiers) {
|
|
|
|
err = c_append_qualifiers(qualified_type.qualifiers, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ' '))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, "typedef "))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
aliased_type = drgn_type_type(qualified_type.type);
|
|
|
|
return c_declare_variable(aliased_type, &typedef_name, 0, sb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_define_type(struct drgn_qualified_type qualified_type, size_t indent,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
switch (drgn_type_kind(qualified_type.type)) {
|
|
|
|
case DRGN_TYPE_VOID:
|
|
|
|
case DRGN_TYPE_INT:
|
|
|
|
case DRGN_TYPE_BOOL:
|
|
|
|
case DRGN_TYPE_FLOAT:
|
|
|
|
case DRGN_TYPE_COMPLEX:
|
|
|
|
return c_declare_basic(qualified_type, NULL, indent, sb);
|
|
|
|
case DRGN_TYPE_STRUCT:
|
|
|
|
case DRGN_TYPE_UNION:
|
2019-11-15 01:12:47 +00:00
|
|
|
case DRGN_TYPE_CLASS:
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return c_define_compound(qualified_type, indent, sb);
|
|
|
|
case DRGN_TYPE_ENUM:
|
|
|
|
return c_define_enum(qualified_type, indent, sb);
|
|
|
|
case DRGN_TYPE_TYPEDEF:
|
|
|
|
return c_define_typedef(qualified_type, indent, sb);
|
|
|
|
case DRGN_TYPE_POINTER:
|
|
|
|
return c_declare_pointer(qualified_type, NULL, indent, sb);
|
|
|
|
case DRGN_TYPE_ARRAY:
|
|
|
|
return c_declare_array(qualified_type, NULL, indent, sb);
|
|
|
|
case DRGN_TYPE_FUNCTION:
|
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"function type cannot be pretty-printed");
|
|
|
|
}
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_anonymous_type_name(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
|
|
|
err = c_append_tagged_name(qualified_type, 0, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, " <anonymous>"))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_type_name_impl(struct drgn_qualified_type qualified_type,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
if (drgn_type_is_anonymous(qualified_type.type)) {
|
|
|
|
return c_anonymous_type_name(qualified_type, sb);
|
|
|
|
} else if (drgn_type_kind(qualified_type.type) == DRGN_TYPE_FUNCTION) {
|
|
|
|
struct string_callback name_cb = {
|
|
|
|
.fn = c_variable_name,
|
|
|
|
.arg = (void *)"",
|
|
|
|
};
|
|
|
|
|
|
|
|
return c_declare_function(qualified_type, &name_cb, 0, sb);
|
|
|
|
} else {
|
|
|
|
return c_declare_variable(qualified_type, NULL, 0, sb);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_type_name(struct drgn_qualified_type qualified_type, char **ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
2019-05-09 23:25:58 +01:00
|
|
|
struct string_builder sb = {};
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
err = c_pretty_print_type_name_impl(qualified_type, &sb);
|
|
|
|
if (err) {
|
|
|
|
free(sb.str);
|
|
|
|
return err;
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_finalize(&sb, ret))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_type(struct drgn_qualified_type qualified_type, char **ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
2019-05-09 23:25:58 +01:00
|
|
|
struct string_builder sb = {};
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (drgn_type_is_complete(qualified_type.type))
|
|
|
|
err = c_define_type(qualified_type, 0, &sb);
|
|
|
|
else
|
|
|
|
err = c_pretty_print_type_name_impl(qualified_type, &sb);
|
|
|
|
if (err) {
|
|
|
|
free(sb.str);
|
|
|
|
return err;
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_finalize(&sb, ret))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_object_impl(const struct drgn_object *obj, bool cast,
|
|
|
|
bool dereference, size_t indent,
|
|
|
|
size_t one_line_columns, size_t multi_line_columns,
|
|
|
|
struct string_builder *sb);
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_int_object(const struct drgn_object *obj,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
|
|
|
switch (obj->kind) {
|
|
|
|
case DRGN_OBJECT_SIGNED: {
|
|
|
|
int64_t svalue;
|
|
|
|
|
|
|
|
err = drgn_object_read_signed(obj, &svalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRId64, svalue))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
case DRGN_OBJECT_UNSIGNED: {
|
|
|
|
uint64_t uvalue;
|
|
|
|
|
|
|
|
err = drgn_object_read_unsigned(obj, &uvalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRIu64, uvalue))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_float_object(const struct drgn_object *obj,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
double fvalue;
|
|
|
|
|
|
|
|
err = drgn_object_read_float(obj, &fvalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (rint(fvalue) == fvalue) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%.1f", fvalue))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%.*g", DBL_DECIMAL_DIG,
|
|
|
|
fvalue))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_pretty_print_members(const struct drgn_object *obj,
|
|
|
|
struct drgn_object *member,
|
|
|
|
struct drgn_type *type,
|
|
|
|
uint64_t bit_offset,
|
|
|
|
size_t indent,
|
|
|
|
size_t multi_line_columns,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_type_member *members;
|
|
|
|
size_t num_members, i;
|
|
|
|
|
|
|
|
if (!drgn_type_has_members(type))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
members = drgn_type_members(type);
|
|
|
|
num_members = drgn_type_num_members(type);
|
|
|
|
for (i = 0; i < num_members; i++) {
|
|
|
|
struct drgn_qualified_type member_type;
|
|
|
|
|
|
|
|
err = drgn_member_type(&members[i], &member_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (members[i].name) {
|
|
|
|
size_t member_start, remaining_columns;
|
|
|
|
|
|
|
|
if (multi_line_columns == 0)
|
|
|
|
return &drgn_stop;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '\n') ||
|
|
|
|
!append_tabs(indent + 1, sb))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
member_start = sb->len;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, ".%s = ",
|
|
|
|
members[i].name))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (__builtin_sub_overflow(multi_line_columns,
|
|
|
|
8 * (indent + 1) +
|
|
|
|
sb->len - member_start + 1,
|
|
|
|
&remaining_columns))
|
|
|
|
remaining_columns = 0;
|
|
|
|
|
|
|
|
err = drgn_object_slice(member, obj, member_type,
|
|
|
|
bit_offset + members[i].bit_offset,
|
|
|
|
members[i].bit_field_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = c_pretty_print_object_impl(member, true, false,
|
|
|
|
indent + 1,
|
|
|
|
remaining_columns,
|
|
|
|
multi_line_columns,
|
|
|
|
sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ','))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
|
|
|
err = c_pretty_print_members(obj, member,
|
|
|
|
member_type.type,
|
|
|
|
bit_offset + members[i].bit_offset,
|
|
|
|
indent, multi_line_columns,
|
|
|
|
sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_compound_object(const struct drgn_object *obj,
|
|
|
|
struct drgn_type *underlying_type,
|
|
|
|
size_t indent, size_t multi_line_columns,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
size_t old_len;
|
|
|
|
struct drgn_object member;
|
|
|
|
|
|
|
|
if (!drgn_type_is_complete(underlying_type)) {
|
2019-11-15 01:12:47 +00:00
|
|
|
const char *keyword;
|
|
|
|
|
|
|
|
switch (drgn_type_kind(underlying_type)) {
|
|
|
|
case DRGN_TYPE_STRUCT:
|
|
|
|
keyword = "struct";
|
|
|
|
break;
|
|
|
|
case DRGN_TYPE_UNION:
|
|
|
|
keyword = "union";
|
|
|
|
break;
|
|
|
|
case DRGN_TYPE_CLASS:
|
|
|
|
keyword = "class";
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return drgn_error_format(DRGN_ERROR_TYPE,
|
|
|
|
"cannot format incomplete %s object",
|
2019-11-15 01:12:47 +00:00
|
|
|
keyword);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '{'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
old_len = sb->len;
|
|
|
|
|
|
|
|
drgn_object_init(&member, obj->prog);
|
|
|
|
err = c_pretty_print_members(obj, &member, underlying_type, 0, indent,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
drgn_object_deinit(&member);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (sb->len != old_len) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '\n') ||
|
|
|
|
!append_tabs(indent, sb))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '}'))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_enum_object(const struct drgn_object *obj,
|
|
|
|
struct drgn_type *underlying_type,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_type_enumerator *enumerators;
|
|
|
|
size_t num_enumerators, i;
|
|
|
|
|
|
|
|
if (!drgn_type_is_complete(underlying_type)) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_TYPE,
|
|
|
|
"cannot format incomplete enum object");
|
|
|
|
}
|
|
|
|
|
|
|
|
enumerators = drgn_type_enumerators(underlying_type);
|
|
|
|
num_enumerators = drgn_type_num_enumerators(underlying_type);
|
|
|
|
if (drgn_enum_type_is_signed(underlying_type)) {
|
|
|
|
int64_t svalue;
|
|
|
|
|
|
|
|
err = drgn_object_read_signed(obj, &svalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
for (i = 0; i < num_enumerators; i++) {
|
|
|
|
if (enumerators[i].svalue == svalue) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb,
|
|
|
|
enumerators[i].name))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRId64, svalue))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
|
|
|
uint64_t uvalue;
|
|
|
|
|
|
|
|
err = drgn_object_read_unsigned(obj, &uvalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
for (i = 0; i < num_enumerators; i++) {
|
|
|
|
if (enumerators[i].uvalue == uvalue) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb,
|
|
|
|
enumerators[i].name))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "%" PRIu64, uvalue))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool is_character_type(struct drgn_type *type)
|
|
|
|
{
|
2019-04-20 00:05:19 +01:00
|
|
|
switch (drgn_type_primitive(type)) {
|
|
|
|
case DRGN_C_TYPE_CHAR:
|
|
|
|
case DRGN_C_TYPE_SIGNED_CHAR:
|
|
|
|
case DRGN_C_TYPE_UNSIGNED_CHAR:
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_character(unsigned char c, struct string_builder *sb)
|
|
|
|
{
|
2019-05-14 18:07:50 +01:00
|
|
|
bool ret;
|
|
|
|
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
switch (c) {
|
|
|
|
case '\a':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\a");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\b':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\b");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\t':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\t");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\n':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\n");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\v':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\v");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\f':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\f");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\r':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\r");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '"':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\\"");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
case '\\':
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_append(sb, "\\\\");
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
default:
|
|
|
|
if (c <= '\x1f' || c >= '\x7f')
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_appendf(sb, "\\x%02x", c);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
else
|
2019-05-14 18:07:50 +01:00
|
|
|
ret = string_builder_appendc(sb, c);
|
|
|
|
break;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
return ret ? NULL : &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_string(struct drgn_memory_reader *reader, uint64_t address,
|
|
|
|
uint64_t length, struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '"'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
while (length) {
|
|
|
|
unsigned char c;
|
|
|
|
|
|
|
|
err = drgn_memory_reader_read(reader, &c, address++, 1, false);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (c == '\0') {
|
|
|
|
break;
|
|
|
|
} else {
|
|
|
|
err = c_pretty_print_character(c, sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
length--;
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '"'))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_pointer_object(const struct drgn_object *obj,
|
|
|
|
struct drgn_type *underlying_type, bool cast,
|
|
|
|
bool dereference, size_t indent,
|
|
|
|
size_t one_line_columns,
|
|
|
|
size_t multi_line_columns,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
bool is_c_string;
|
|
|
|
uint64_t uvalue;
|
|
|
|
size_t old_len, address_end;
|
|
|
|
|
|
|
|
is_c_string = is_character_type(drgn_type_type(underlying_type).type);
|
|
|
|
/* Always dereference strings. */
|
|
|
|
if (is_c_string)
|
|
|
|
dereference = true;
|
|
|
|
|
|
|
|
old_len = sb->len;
|
|
|
|
if (dereference && !is_c_string) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '*'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
if (cast) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '('))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
err = c_pretty_print_type_name_impl(drgn_object_qualified_type(obj),
|
|
|
|
sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ')'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
err = drgn_object_read_unsigned(obj, &uvalue);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "0x%" PRIx64, uvalue))
|
|
|
|
return &drgn_enomem;
|
|
|
|
if (!dereference)
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
address_end = sb->len;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, " = "))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (__builtin_sub_overflow(one_line_columns, sb->len - old_len,
|
|
|
|
&one_line_columns))
|
|
|
|
one_line_columns = 0;
|
|
|
|
|
|
|
|
if (is_c_string) {
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_pretty_print_string(&obj->prog->reader, uvalue,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
UINT64_MAX, sb);
|
|
|
|
} else {
|
|
|
|
struct drgn_object dereferenced;
|
|
|
|
|
|
|
|
drgn_object_init(&dereferenced, obj->prog);
|
|
|
|
err = drgn_object_dereference(&dereferenced, obj);
|
|
|
|
if (err) {
|
|
|
|
drgn_object_deinit(&dereferenced);
|
|
|
|
if (err->code == DRGN_ERROR_TYPE)
|
|
|
|
goto no_dereference;
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
err = c_pretty_print_object_impl(&dereferenced, false, false,
|
|
|
|
indent, one_line_columns,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
drgn_object_deinit(&dereferenced);
|
|
|
|
}
|
|
|
|
if (!err || err->code != DRGN_ERROR_FAULT) {
|
|
|
|
/* We either succeeded or hit a fatal error. */
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
no_dereference:
|
|
|
|
/*
|
|
|
|
* We hit a non-fatal error. Delete the asterisk and truncate everything
|
|
|
|
* after the address.
|
|
|
|
*/
|
|
|
|
drgn_error_destroy(err);
|
|
|
|
sb->len = address_end;
|
|
|
|
if (!is_c_string) {
|
|
|
|
sb->len--;
|
|
|
|
memmove(&sb->str[old_len], &sb->str[old_len + 1],
|
|
|
|
sb->len - old_len);
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_array_object(const struct drgn_object *obj,
|
|
|
|
struct drgn_type *underlying_type, size_t indent,
|
|
|
|
size_t one_line_columns, size_t multi_line_columns,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_qualified_type element_type;
|
|
|
|
uint64_t element_bit_size;
|
|
|
|
struct drgn_object element;
|
|
|
|
uint64_t length, i;
|
|
|
|
size_t old_len, remaining_columns, start_columns;
|
|
|
|
|
|
|
|
length = drgn_type_length(underlying_type);
|
|
|
|
element_type = drgn_type_type(underlying_type);
|
|
|
|
|
|
|
|
if (length && is_character_type(element_type.type)) {
|
|
|
|
if (obj->is_reference) {
|
2019-05-01 19:22:59 +01:00
|
|
|
return c_pretty_print_string(&obj->prog->reader,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
obj->reference.address,
|
|
|
|
length, sb);
|
|
|
|
} else {
|
|
|
|
const unsigned char *buf;
|
|
|
|
uint64_t size;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '"'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
buf = (const unsigned char *)drgn_object_buffer(obj);
|
2019-10-28 18:28:38 +00:00
|
|
|
size = drgn_buffer_object_size(obj);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
for (i = 0; i < size; i++) {
|
|
|
|
if (buf[i] == '\0')
|
|
|
|
break;
|
|
|
|
err = c_pretty_print_character(buf[i], sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '"'))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
err = drgn_type_bit_size(element_type.type, &element_bit_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
drgn_object_init(&element, obj->prog);
|
|
|
|
while (length) {
|
|
|
|
bool truthy;
|
|
|
|
|
|
|
|
err = drgn_object_slice(&element, obj, element_type,
|
|
|
|
(length - 1) * element_bit_size, 0);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
err = drgn_object_truthiness(&element, &truthy);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
if (!truthy)
|
|
|
|
length--;
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (!length) {
|
2019-05-14 18:07:50 +01:00
|
|
|
if (string_builder_append(sb, "{}"))
|
|
|
|
err = NULL;
|
|
|
|
else
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* First, try to fit everything on one line. */
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, "{ ")) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
old_len = sb->len - 1; /* Minus one for the space. */
|
|
|
|
if (__builtin_sub_overflow(one_line_columns, 2, &remaining_columns))
|
|
|
|
remaining_columns = 0;
|
|
|
|
/* Stop if we can't fit the comma, space, and closing brace. */
|
|
|
|
for (i = 0; i < length && remaining_columns >= 3; i++) {
|
|
|
|
size_t element_start;
|
|
|
|
|
|
|
|
err = drgn_object_slice(&element, obj, element_type,
|
|
|
|
i * element_bit_size, 0);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
element_start = sb->len;
|
|
|
|
err = c_pretty_print_object_impl(&element, false, false,
|
|
|
|
indent + 1,
|
|
|
|
remaining_columns - 3, 0, sb);
|
|
|
|
if (err && err->code == DRGN_ERROR_STOP)
|
|
|
|
break;
|
|
|
|
else if (err)
|
|
|
|
goto out;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_append(sb, ", ")) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (__builtin_sub_overflow(remaining_columns,
|
|
|
|
sb->len - element_start,
|
|
|
|
&remaining_columns))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (i >= length && remaining_columns >= 1) {
|
|
|
|
/* Everything fit. */
|
2019-05-14 18:07:50 +01:00
|
|
|
if (string_builder_appendc(sb, '}'))
|
|
|
|
err = NULL;
|
|
|
|
else
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (multi_line_columns == 0) {
|
|
|
|
/* We were asked to fit on one line and we couldn't. */
|
|
|
|
return &drgn_stop;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Start over (truncate the string) and use multiple lines. */
|
|
|
|
sb->len = old_len;
|
|
|
|
if (__builtin_sub_overflow(multi_line_columns, 8 * (indent + 1),
|
|
|
|
&start_columns))
|
|
|
|
start_columns = 0;
|
|
|
|
remaining_columns = 0;
|
|
|
|
for (i = 0; i < length; i++) {
|
|
|
|
size_t newline;
|
|
|
|
|
|
|
|
err = drgn_object_slice(&element, obj, element_type,
|
|
|
|
i * element_bit_size, 0);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
newline = sb->len;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '\n') ||
|
|
|
|
!append_tabs(indent + 1, sb)) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (start_columns > 1) {
|
|
|
|
size_t element_start = sb->len;
|
|
|
|
|
|
|
|
err = c_pretty_print_object_impl(&element, false, false,
|
|
|
|
0, start_columns - 1,
|
|
|
|
0, sb);
|
|
|
|
if (!err) {
|
|
|
|
size_t element_len = sb->len - element_start;
|
|
|
|
|
|
|
|
if (element_len +
|
|
|
|
(remaining_columns == start_columns ? 1 : 2)
|
|
|
|
<= remaining_columns) {
|
|
|
|
/*
|
|
|
|
* It would've fit on the previous line.
|
|
|
|
* Move it over.
|
|
|
|
*/
|
|
|
|
if (remaining_columns != start_columns) {
|
|
|
|
sb->str[newline++] = ' ';
|
|
|
|
remaining_columns--;
|
|
|
|
}
|
|
|
|
memmove(&sb->str[newline],
|
|
|
|
&sb->str[element_start],
|
|
|
|
element_len);
|
|
|
|
sb->len = newline + element_len;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ',')) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
remaining_columns -= element_len + 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (element_len < start_columns) {
|
|
|
|
/* It fit on the new line. */
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ',')) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
remaining_columns = (start_columns -
|
|
|
|
element_len - 1);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
} else if (err->code != DRGN_ERROR_STOP) {
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/* It didn't fit. */
|
|
|
|
sb->len = element_start;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = c_pretty_print_object_impl(&element, false, false,
|
|
|
|
indent + 1, 0,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ',')) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
remaining_columns = 0;
|
|
|
|
}
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '\n') || !append_tabs(indent, sb) ||
|
|
|
|
!string_builder_appendc(sb, '}')) {
|
|
|
|
err = &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
2019-05-14 18:07:50 +01:00
|
|
|
}
|
|
|
|
err = NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
out:
|
|
|
|
drgn_object_deinit(&element);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_function_object(const struct drgn_object *obj,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
/* Function values currently aren't possible anyways. */
|
|
|
|
if (!obj->is_reference) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_TYPE,
|
|
|
|
"cannot format function value");
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendf(sb, "0x%" PRIx64, obj->reference.address))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_pretty_print_object_impl(const struct drgn_object *obj, bool cast,
|
|
|
|
bool dereference, size_t indent,
|
|
|
|
size_t one_line_columns, size_t multi_line_columns,
|
|
|
|
struct string_builder *sb)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_type *underlying_type = drgn_underlying_type(obj->type);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Pointers are special because they can have an asterisk prefix if
|
|
|
|
* we're dereferencing them.
|
|
|
|
*/
|
|
|
|
if (drgn_type_kind(underlying_type) == DRGN_TYPE_POINTER) {
|
|
|
|
return c_pretty_print_pointer_object(obj, underlying_type, cast,
|
|
|
|
dereference, indent,
|
|
|
|
one_line_columns,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cast) {
|
|
|
|
size_t old_len = sb->len;
|
|
|
|
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, '('))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
err = c_pretty_print_type_name_impl(drgn_object_qualified_type(obj),
|
|
|
|
sb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_appendc(sb, ')'))
|
|
|
|
return &drgn_enomem;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
if (__builtin_sub_overflow(one_line_columns, sb->len - old_len,
|
|
|
|
&one_line_columns))
|
|
|
|
one_line_columns = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (drgn_type_kind(underlying_type)) {
|
|
|
|
case DRGN_TYPE_VOID:
|
|
|
|
return drgn_error_create(DRGN_ERROR_TYPE,
|
|
|
|
"cannot format void object");
|
|
|
|
case DRGN_TYPE_INT:
|
|
|
|
case DRGN_TYPE_BOOL:
|
|
|
|
return c_pretty_print_int_object(obj, sb);
|
|
|
|
case DRGN_TYPE_FLOAT:
|
|
|
|
return c_pretty_print_float_object(obj, sb);
|
|
|
|
case DRGN_TYPE_COMPLEX:
|
|
|
|
return drgn_error_format(DRGN_ERROR_TYPE,
|
|
|
|
"complex object formatting is not implemented");
|
|
|
|
case DRGN_TYPE_STRUCT:
|
|
|
|
case DRGN_TYPE_UNION:
|
2019-11-15 01:12:47 +00:00
|
|
|
case DRGN_TYPE_CLASS:
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return c_pretty_print_compound_object(obj, underlying_type,
|
|
|
|
indent,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
case DRGN_TYPE_ENUM:
|
|
|
|
return c_pretty_print_enum_object(obj, underlying_type, sb);
|
|
|
|
case DRGN_TYPE_ARRAY:
|
|
|
|
return c_pretty_print_array_object(obj, underlying_type, indent,
|
|
|
|
one_line_columns,
|
|
|
|
multi_line_columns, sb);
|
|
|
|
case DRGN_TYPE_FUNCTION:
|
|
|
|
return c_pretty_print_function_object(obj, sb);
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_pretty_print_object(const struct drgn_object *obj,
|
|
|
|
size_t columns, char **ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
2019-05-09 23:25:58 +01:00
|
|
|
struct string_builder sb = {};
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
err = c_pretty_print_object_impl(obj, true, true, 0, columns,
|
|
|
|
max(columns, (size_t)1), &sb);
|
|
|
|
if (err) {
|
|
|
|
free(sb.str);
|
|
|
|
return err;
|
|
|
|
}
|
2019-05-14 18:07:50 +01:00
|
|
|
if (!string_builder_finalize(&sb, ret))
|
|
|
|
return &drgn_enomem;
|
|
|
|
return NULL;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* This obviously incomplete since we only handle the tokens we care about. */
|
|
|
|
enum {
|
|
|
|
C_TOKEN_EOF = -1,
|
|
|
|
MIN_KEYWORD_TOKEN,
|
|
|
|
MIN_SPECIFIER_TOKEN = MIN_KEYWORD_TOKEN,
|
|
|
|
C_TOKEN_VOID = MIN_SPECIFIER_TOKEN,
|
|
|
|
C_TOKEN_CHAR,
|
|
|
|
C_TOKEN_SHORT,
|
|
|
|
C_TOKEN_INT,
|
|
|
|
C_TOKEN_LONG,
|
|
|
|
C_TOKEN_SIGNED,
|
|
|
|
C_TOKEN_UNSIGNED,
|
|
|
|
C_TOKEN_BOOL,
|
|
|
|
C_TOKEN_FLOAT,
|
|
|
|
C_TOKEN_DOUBLE,
|
|
|
|
C_TOKEN_COMPLEX,
|
|
|
|
MAX_SPECIFIER_TOKEN = C_TOKEN_COMPLEX,
|
|
|
|
MIN_QUALIFIER_TOKEN,
|
|
|
|
C_TOKEN_CONST = MIN_QUALIFIER_TOKEN,
|
|
|
|
C_TOKEN_RESTRICT,
|
|
|
|
C_TOKEN_VOLATILE,
|
|
|
|
C_TOKEN_ATOMIC,
|
|
|
|
MAX_QUALIFIER_TOKEN = C_TOKEN_ATOMIC,
|
|
|
|
C_TOKEN_STRUCT,
|
|
|
|
C_TOKEN_UNION,
|
|
|
|
C_TOKEN_ENUM,
|
|
|
|
MAX_KEYWORD_TOKEN = C_TOKEN_ENUM,
|
|
|
|
C_TOKEN_LPAREN,
|
|
|
|
C_TOKEN_RPAREN,
|
|
|
|
C_TOKEN_LBRACKET,
|
|
|
|
C_TOKEN_RBRACKET,
|
|
|
|
C_TOKEN_ASTERISK,
|
|
|
|
C_TOKEN_DOT,
|
|
|
|
C_TOKEN_NUMBER,
|
|
|
|
C_TOKEN_IDENTIFIER,
|
|
|
|
};
|
|
|
|
|
|
|
|
static const char *token_spelling[] = {
|
|
|
|
[C_TOKEN_VOID] = "void",
|
|
|
|
[C_TOKEN_CHAR] = "char",
|
|
|
|
[C_TOKEN_SHORT] = "short",
|
|
|
|
[C_TOKEN_INT] = "int",
|
|
|
|
[C_TOKEN_LONG] = "long",
|
|
|
|
[C_TOKEN_SIGNED] = "signed",
|
|
|
|
[C_TOKEN_UNSIGNED] = "unsigned",
|
|
|
|
[C_TOKEN_BOOL] = "_Bool",
|
|
|
|
[C_TOKEN_FLOAT] = "float",
|
|
|
|
[C_TOKEN_DOUBLE] = "double",
|
|
|
|
[C_TOKEN_COMPLEX] = "_Complex",
|
|
|
|
[C_TOKEN_CONST] = "const",
|
|
|
|
[C_TOKEN_RESTRICT] = "restrict",
|
|
|
|
[C_TOKEN_VOLATILE] = "volatile",
|
|
|
|
[C_TOKEN_ATOMIC] = "_Atomic",
|
|
|
|
[C_TOKEN_STRUCT] = "struct",
|
|
|
|
[C_TOKEN_UNION] = "union",
|
|
|
|
[C_TOKEN_ENUM] = "enum",
|
|
|
|
};
|
|
|
|
|
|
|
|
DEFINE_HASH_MAP(c_keyword_map, struct string, int, string_hash, string_eq);
|
|
|
|
|
|
|
|
static struct c_keyword_map c_keywords;
|
|
|
|
|
|
|
|
__attribute__((constructor(101)))
|
|
|
|
static void c_keywords_init(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
c_keyword_map_init(&c_keywords);
|
|
|
|
for (i = MIN_KEYWORD_TOKEN; i <= MAX_KEYWORD_TOKEN; i++) {
|
2019-05-16 09:25:43 +01:00
|
|
|
struct c_keyword_map_entry entry = {
|
|
|
|
.key = {
|
|
|
|
.str = token_spelling[i],
|
|
|
|
.len = strlen(token_spelling[i]),
|
|
|
|
},
|
|
|
|
.value = i,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
};
|
|
|
|
|
2019-05-16 09:25:43 +01:00
|
|
|
if (c_keyword_map_insert(&c_keywords, &entry, NULL) != 1)
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
abort();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
__attribute__((destructor(101)))
|
|
|
|
static void c_keywords_deinit(void)
|
|
|
|
{
|
|
|
|
c_keyword_map_deinit(&c_keywords);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct drgn_error *drgn_lexer_c(struct drgn_lexer *lexer,
|
|
|
|
struct drgn_token *token) {
|
|
|
|
const char *p = lexer->p;
|
|
|
|
|
|
|
|
while (isspace(*p))
|
|
|
|
p++;
|
|
|
|
|
|
|
|
token->value = p;
|
|
|
|
switch (*p) {
|
|
|
|
case '\0':
|
|
|
|
token->kind = C_TOKEN_EOF;
|
|
|
|
break;
|
|
|
|
case '(':
|
|
|
|
token->kind = C_TOKEN_LPAREN;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
case ')':
|
|
|
|
token->kind = C_TOKEN_RPAREN;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
case '[':
|
|
|
|
token->kind = C_TOKEN_LBRACKET;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
case ']':
|
|
|
|
token->kind = C_TOKEN_RBRACKET;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
case '*':
|
|
|
|
token->kind = C_TOKEN_ASTERISK;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
case '.':
|
|
|
|
token->kind = C_TOKEN_DOT;
|
|
|
|
p++;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
if (isalpha(*p) || *p == '_') {
|
|
|
|
struct string key;
|
2019-05-16 09:25:43 +01:00
|
|
|
struct c_keyword_map_iterator it;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
do {
|
|
|
|
p++;
|
|
|
|
} while (isalnum(*p) || *p == '_');
|
|
|
|
|
|
|
|
key.str = token->value;
|
|
|
|
key.len = p - token->value;
|
2019-05-16 09:25:43 +01:00
|
|
|
it = c_keyword_map_search(&c_keywords, &key);
|
|
|
|
token->kind = (it.entry ? it.entry->value :
|
|
|
|
C_TOKEN_IDENTIFIER);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else if ('0' <= *p && *p <= '9') {
|
|
|
|
token->kind = C_TOKEN_NUMBER;
|
|
|
|
if (*p++ == '0' && *p == 'x') {
|
|
|
|
p++;
|
|
|
|
while (('0' <= *p && *p <= '9') ||
|
|
|
|
('a' <= *p && *p <= 'f') ||
|
|
|
|
('A' <= *p && *p <= 'F')) {
|
|
|
|
p++;
|
|
|
|
}
|
|
|
|
if (p - token->value <= 2) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"invalid number");
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
while ('0' <= *p && *p <= '9')
|
|
|
|
p++;
|
|
|
|
}
|
|
|
|
if (isalpha(*p) || *p == '_') {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"invalid number");
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"invalid character \\x%02x", (unsigned char)*p);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
token->len = p - token->value;
|
|
|
|
lexer->p = p;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_token_to_u64(const struct drgn_token *token,
|
|
|
|
uint64_t *ret)
|
|
|
|
{
|
|
|
|
uint64_t x = 0;
|
|
|
|
size_t i;
|
|
|
|
|
|
|
|
assert(token->kind == C_TOKEN_NUMBER);
|
|
|
|
if (token->len > 2 && token->value[0] == '0' &&
|
|
|
|
token->value[1] == 'x') {
|
|
|
|
for (i = 2; i < token->len; i++) {
|
|
|
|
char c = token->value[i];
|
|
|
|
int digit;
|
|
|
|
|
|
|
|
if ('0' <= c && c <= '9')
|
|
|
|
digit = c - '0';
|
|
|
|
else if ('a' <= c && c <= 'f')
|
|
|
|
digit = c - 'a';
|
|
|
|
else /* ('A' <= c && c <= 'F') */
|
|
|
|
digit = c - 'A';
|
|
|
|
if (x > UINT64_MAX / 16)
|
|
|
|
goto overflow;
|
|
|
|
x *= 16;
|
|
|
|
if (x > UINT64_MAX - digit)
|
|
|
|
goto overflow;
|
|
|
|
x += digit;
|
|
|
|
}
|
|
|
|
} else if (token->value[0] == '0') {
|
|
|
|
for (i = 1; i < token->len; i++) {
|
|
|
|
int digit;
|
|
|
|
|
|
|
|
digit = token->value[i] - '0';
|
|
|
|
if (x > UINT64_MAX / 8)
|
|
|
|
goto overflow;
|
|
|
|
x *= 8;
|
|
|
|
if (x > UINT64_MAX - digit)
|
|
|
|
goto overflow;
|
|
|
|
x += digit;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
for (i = 0; i < token->len; i++) {
|
|
|
|
int digit;
|
|
|
|
|
|
|
|
digit = token->value[i] - '0';
|
|
|
|
if (x > UINT64_MAX / 10)
|
|
|
|
goto overflow;
|
|
|
|
x *= 10;
|
|
|
|
if (x > UINT64_MAX - digit)
|
|
|
|
goto overflow;
|
|
|
|
x += digit;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
*ret = x;
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
overflow:
|
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"number is too large");
|
|
|
|
}
|
|
|
|
|
|
|
|
enum c_type_specifier {
|
|
|
|
SPECIFIER_ERROR,
|
|
|
|
SPECIFIER_VOID,
|
|
|
|
SPECIFIER_CHAR,
|
|
|
|
SPECIFIER_SIGNED_CHAR,
|
|
|
|
SPECIFIER_UNSIGNED_CHAR,
|
|
|
|
SPECIFIER_SHORT,
|
|
|
|
SPECIFIER_SHORT_INT,
|
|
|
|
SPECIFIER_SIGNED_SHORT_INT,
|
|
|
|
SPECIFIER_UNSIGNED_SHORT_INT,
|
|
|
|
SPECIFIER_SIGNED_SHORT,
|
|
|
|
SPECIFIER_UNSIGNED_SHORT,
|
|
|
|
SPECIFIER_INT,
|
|
|
|
SPECIFIER_SIGNED_INT,
|
|
|
|
SPECIFIER_UNSIGNED_INT,
|
|
|
|
SPECIFIER_LONG,
|
|
|
|
SPECIFIER_LONG_INT,
|
|
|
|
SPECIFIER_SIGNED_LONG,
|
|
|
|
SPECIFIER_UNSIGNED_LONG,
|
|
|
|
SPECIFIER_SIGNED_LONG_INT,
|
|
|
|
SPECIFIER_UNSIGNED_LONG_INT,
|
|
|
|
SPECIFIER_LONG_LONG,
|
|
|
|
SPECIFIER_LONG_LONG_INT,
|
|
|
|
SPECIFIER_SIGNED_LONG_LONG_INT,
|
|
|
|
SPECIFIER_UNSIGNED_LONG_LONG_INT,
|
|
|
|
SPECIFIER_SIGNED_LONG_LONG,
|
|
|
|
SPECIFIER_UNSIGNED_LONG_LONG,
|
|
|
|
SPECIFIER_SIGNED,
|
|
|
|
SPECIFIER_UNSIGNED,
|
|
|
|
SPECIFIER_BOOL,
|
|
|
|
SPECIFIER_FLOAT,
|
|
|
|
SPECIFIER_DOUBLE,
|
|
|
|
SPECIFIER_LONG_DOUBLE,
|
|
|
|
SPECIFIER_NONE,
|
|
|
|
NUM_SPECIFIER_STATES,
|
|
|
|
};
|
|
|
|
|
|
|
|
static const char *specifier_spelling[NUM_SPECIFIER_STATES] = {
|
|
|
|
[SPECIFIER_VOID] = "void",
|
|
|
|
[SPECIFIER_CHAR] = "char",
|
|
|
|
[SPECIFIER_SIGNED_CHAR] = "signed char",
|
|
|
|
[SPECIFIER_UNSIGNED_CHAR] = "unsigned char",
|
|
|
|
[SPECIFIER_SHORT] = "short",
|
|
|
|
[SPECIFIER_SHORT_INT] = "short int",
|
|
|
|
[SPECIFIER_SIGNED_SHORT_INT] = "signed short int",
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT_INT] = "unsigned short int",
|
|
|
|
[SPECIFIER_SIGNED_SHORT] = "signed short",
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT] = "unsigned short",
|
|
|
|
[SPECIFIER_INT] = "int",
|
|
|
|
[SPECIFIER_SIGNED_INT] = "signed int",
|
|
|
|
[SPECIFIER_UNSIGNED_INT] = "unsigned int",
|
|
|
|
[SPECIFIER_LONG] = "long",
|
|
|
|
[SPECIFIER_LONG_INT] = "long int",
|
|
|
|
[SPECIFIER_SIGNED_LONG] = "signed long",
|
|
|
|
[SPECIFIER_UNSIGNED_LONG] = "unsigned long",
|
|
|
|
[SPECIFIER_SIGNED_LONG_INT] = "signed long int",
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_INT] = "unsigned long int",
|
|
|
|
[SPECIFIER_LONG_LONG] = "long long",
|
|
|
|
[SPECIFIER_LONG_LONG_INT] = "long long int",
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG_INT] = "signed long long int",
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG_INT] = "unsigned long long int",
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG] = "signed long long",
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG] = "unsigned long long",
|
|
|
|
[SPECIFIER_SIGNED] = "signed",
|
|
|
|
[SPECIFIER_UNSIGNED] = "unsigned",
|
|
|
|
[SPECIFIER_BOOL] = "_Bool",
|
|
|
|
[SPECIFIER_FLOAT] = "float",
|
|
|
|
[SPECIFIER_DOUBLE] = "double",
|
|
|
|
[SPECIFIER_LONG_DOUBLE] = "long double",
|
|
|
|
};
|
|
|
|
|
|
|
|
static const enum drgn_qualifiers qualifier_from_token[MAX_QUALIFIER_TOKEN + 1] = {
|
|
|
|
[C_TOKEN_CONST] = DRGN_QUALIFIER_CONST,
|
|
|
|
[C_TOKEN_RESTRICT] = DRGN_QUALIFIER_RESTRICT,
|
|
|
|
[C_TOKEN_VOLATILE] = DRGN_QUALIFIER_VOLATILE,
|
|
|
|
[C_TOKEN_ATOMIC] = DRGN_QUALIFIER_ATOMIC,
|
|
|
|
};
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
static const enum c_type_specifier
|
|
|
|
specifier_transition[NUM_SPECIFIER_STATES][MAX_SPECIFIER_TOKEN + 1] = {
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
[SPECIFIER_NONE] = {
|
|
|
|
[C_TOKEN_VOID] = SPECIFIER_VOID,
|
|
|
|
[C_TOKEN_CHAR] = SPECIFIER_CHAR,
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_SHORT,
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_LONG,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED,
|
|
|
|
[C_TOKEN_BOOL] = SPECIFIER_BOOL,
|
|
|
|
[C_TOKEN_FLOAT] = SPECIFIER_FLOAT,
|
|
|
|
[C_TOKEN_DOUBLE] = SPECIFIER_DOUBLE,
|
|
|
|
},
|
|
|
|
[SPECIFIER_VOID] = {},
|
|
|
|
[SPECIFIER_CHAR] = {
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_CHAR,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_CHAR,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_CHAR] = {},
|
|
|
|
[SPECIFIER_UNSIGNED_CHAR] = {},
|
|
|
|
[SPECIFIER_SHORT] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_SHORT_INT,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_SHORT,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_SHORT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SHORT_INT] = {
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_SHORT_INT,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_SHORT_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_SHORT_INT] = {},
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT_INT] = {},
|
|
|
|
[SPECIFIER_SIGNED_SHORT] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_SIGNED_SHORT_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_UNSIGNED_SHORT_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_INT] = {
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_SHORT_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_LONG_INT,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_INT,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_INT] = {
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_SIGNED_SHORT_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_SIGNED_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED_INT] = {
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_UNSIGNED_SHORT_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_UNSIGNED_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_LONG] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_LONG_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_LONG_LONG,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_LONG,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_LONG,
|
|
|
|
[C_TOKEN_DOUBLE] = SPECIFIER_LONG_DOUBLE,
|
|
|
|
},
|
|
|
|
[SPECIFIER_LONG_INT] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_LONG_LONG_INT,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_LONG_INT,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_LONG] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_SIGNED_LONG_LONG,
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_SIGNED_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED_LONG] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_UNSIGNED_LONG_LONG,
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_UNSIGNED_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_LONG_INT] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_SIGNED_LONG_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_INT] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_UNSIGNED_LONG_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_LONG_LONG] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_LONG_LONG_INT,
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_LONG_LONG,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_LONG_LONG,
|
|
|
|
},
|
|
|
|
[SPECIFIER_LONG_LONG_INT] = {
|
|
|
|
[C_TOKEN_SIGNED] = SPECIFIER_SIGNED_LONG_LONG_INT,
|
|
|
|
[C_TOKEN_UNSIGNED] = SPECIFIER_UNSIGNED_LONG_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG_INT] = {},
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG_INT] = {},
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_SIGNED_LONG_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG] = {
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_UNSIGNED_LONG_LONG_INT,
|
|
|
|
},
|
|
|
|
[SPECIFIER_SIGNED] = {
|
|
|
|
[C_TOKEN_CHAR] = SPECIFIER_SIGNED_CHAR,
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_SIGNED_SHORT,
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_SIGNED_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_SIGNED_LONG,
|
|
|
|
},
|
|
|
|
[SPECIFIER_UNSIGNED] = {
|
|
|
|
[C_TOKEN_CHAR] = SPECIFIER_UNSIGNED_CHAR,
|
|
|
|
[C_TOKEN_SHORT] = SPECIFIER_UNSIGNED_SHORT,
|
|
|
|
[C_TOKEN_INT] = SPECIFIER_UNSIGNED_INT,
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_UNSIGNED_LONG,
|
|
|
|
},
|
|
|
|
[SPECIFIER_BOOL] = {},
|
|
|
|
[SPECIFIER_FLOAT] = {},
|
|
|
|
[SPECIFIER_DOUBLE] = {
|
|
|
|
[C_TOKEN_LONG] = SPECIFIER_LONG_DOUBLE,
|
|
|
|
},
|
|
|
|
[SPECIFIER_LONG_DOUBLE] = {},
|
|
|
|
};
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
static const enum drgn_primitive_type specifier_kind[NUM_SPECIFIER_STATES] = {
|
|
|
|
[SPECIFIER_VOID] = DRGN_C_TYPE_VOID,
|
|
|
|
[SPECIFIER_CHAR] = DRGN_C_TYPE_CHAR,
|
|
|
|
[SPECIFIER_SIGNED_CHAR] = DRGN_C_TYPE_SIGNED_CHAR,
|
|
|
|
[SPECIFIER_UNSIGNED_CHAR] = DRGN_C_TYPE_UNSIGNED_CHAR,
|
|
|
|
[SPECIFIER_SHORT] = DRGN_C_TYPE_SHORT,
|
|
|
|
[SPECIFIER_SHORT_INT] = DRGN_C_TYPE_SHORT,
|
|
|
|
[SPECIFIER_SIGNED_SHORT_INT] = DRGN_C_TYPE_SHORT,
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT_INT] = DRGN_C_TYPE_UNSIGNED_SHORT,
|
|
|
|
[SPECIFIER_SIGNED_SHORT] = DRGN_C_TYPE_SHORT,
|
|
|
|
[SPECIFIER_UNSIGNED_SHORT] = DRGN_C_TYPE_UNSIGNED_SHORT,
|
|
|
|
[SPECIFIER_INT] = DRGN_C_TYPE_INT,
|
|
|
|
[SPECIFIER_SIGNED_INT] = DRGN_C_TYPE_INT,
|
|
|
|
[SPECIFIER_UNSIGNED_INT] = DRGN_C_TYPE_UNSIGNED_INT,
|
|
|
|
[SPECIFIER_LONG] = DRGN_C_TYPE_LONG,
|
|
|
|
[SPECIFIER_LONG_INT] = DRGN_C_TYPE_LONG,
|
|
|
|
[SPECIFIER_SIGNED_LONG] = DRGN_C_TYPE_LONG,
|
|
|
|
[SPECIFIER_UNSIGNED_LONG] = DRGN_C_TYPE_UNSIGNED_LONG,
|
|
|
|
[SPECIFIER_SIGNED_LONG_INT] = DRGN_C_TYPE_LONG,
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_INT] = DRGN_C_TYPE_UNSIGNED_LONG,
|
|
|
|
[SPECIFIER_LONG_LONG] = DRGN_C_TYPE_LONG_LONG,
|
|
|
|
[SPECIFIER_LONG_LONG_INT] = DRGN_C_TYPE_LONG_LONG,
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG_INT] = DRGN_C_TYPE_LONG_LONG,
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG_INT] = DRGN_C_TYPE_UNSIGNED_LONG_LONG,
|
|
|
|
[SPECIFIER_SIGNED_LONG_LONG] = DRGN_C_TYPE_LONG_LONG,
|
|
|
|
[SPECIFIER_UNSIGNED_LONG_LONG] = DRGN_C_TYPE_UNSIGNED_LONG_LONG,
|
|
|
|
[SPECIFIER_SIGNED] = DRGN_C_TYPE_INT,
|
|
|
|
[SPECIFIER_UNSIGNED] = DRGN_C_TYPE_UNSIGNED_INT,
|
|
|
|
[SPECIFIER_BOOL] = DRGN_C_TYPE_BOOL,
|
|
|
|
[SPECIFIER_FLOAT] = DRGN_C_TYPE_FLOAT,
|
|
|
|
[SPECIFIER_DOUBLE] = DRGN_C_TYPE_DOUBLE,
|
|
|
|
[SPECIFIER_LONG_DOUBLE] = DRGN_C_TYPE_LONG_DOUBLE,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
};
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
enum drgn_primitive_type c_parse_specifier_list(const char *s)
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_lexer lexer;
|
|
|
|
enum c_type_specifier specifier = SPECIFIER_NONE;
|
2019-04-20 00:05:19 +01:00
|
|
|
enum drgn_primitive_type primitive = DRGN_NOT_PRIMITIVE_TYPE;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
drgn_lexer_init(&lexer, drgn_lexer_c, s);
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(&lexer, &token);
|
|
|
|
if (err) {
|
|
|
|
drgn_error_destroy(err);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (MIN_SPECIFIER_TOKEN <= token.kind &&
|
|
|
|
token.kind <= MAX_SPECIFIER_TOKEN)
|
|
|
|
specifier = specifier_transition[specifier][token.kind];
|
|
|
|
else if (token.kind == C_TOKEN_EOF)
|
|
|
|
break;
|
|
|
|
else
|
|
|
|
specifier = SPECIFIER_ERROR;
|
|
|
|
if (specifier == SPECIFIER_ERROR)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
primitive = specifier_kind[specifier];
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
out:
|
|
|
|
drgn_lexer_deinit(&lexer);
|
2019-04-20 00:05:19 +01:00
|
|
|
return primitive;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_specifier_qualifier_list(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_lexer *lexer, const char *filename,
|
|
|
|
struct drgn_qualified_type *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
enum c_type_specifier specifier = SPECIFIER_NONE;
|
|
|
|
enum drgn_qualifiers qualifiers = 0;
|
|
|
|
const char *identifier = NULL;
|
|
|
|
size_t identifier_len = 0;
|
|
|
|
int tag_token = C_TOKEN_EOF;
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
/* type-qualifier */
|
|
|
|
if (MIN_QUALIFIER_TOKEN <= token.kind &&
|
|
|
|
token.kind <= MAX_QUALIFIER_TOKEN) {
|
|
|
|
qualifiers |= qualifier_from_token[token.kind];
|
|
|
|
/* type-specifier */
|
|
|
|
} else if (MIN_SPECIFIER_TOKEN <= token.kind &&
|
|
|
|
token.kind <= MAX_SPECIFIER_TOKEN) {
|
|
|
|
enum c_type_specifier prev_specifier;
|
|
|
|
|
|
|
|
if (tag_token != C_TOKEN_EOF) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"cannot combine '%s' with '%s'",
|
|
|
|
token_spelling[token.kind],
|
|
|
|
token_spelling[tag_token]);
|
|
|
|
}
|
|
|
|
if (identifier) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"cannot combine '%s' with identifier",
|
|
|
|
token_spelling[token.kind]);
|
|
|
|
}
|
|
|
|
prev_specifier = specifier;
|
|
|
|
specifier = specifier_transition[specifier][token.kind];
|
|
|
|
if (specifier == SPECIFIER_ERROR) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"cannot combine '%s' with '%s'",
|
|
|
|
token_spelling[token.kind],
|
|
|
|
specifier_spelling[prev_specifier]);
|
|
|
|
}
|
|
|
|
} else if (token.kind == C_TOKEN_IDENTIFIER &&
|
|
|
|
specifier == SPECIFIER_NONE && !identifier) {
|
|
|
|
identifier = token.value;
|
|
|
|
identifier_len = token.len;
|
|
|
|
} else if (token.kind == C_TOKEN_STRUCT ||
|
|
|
|
token.kind == C_TOKEN_UNION ||
|
|
|
|
token.kind == C_TOKEN_ENUM) {
|
|
|
|
if (identifier) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"cannot combine '%s' with identifier",
|
|
|
|
token_spelling[token.kind]);
|
|
|
|
}
|
|
|
|
if (specifier != SPECIFIER_NONE) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"cannot combine '%s' with '%s'",
|
|
|
|
token_spelling[token.kind],
|
|
|
|
specifier_spelling[specifier]);
|
|
|
|
}
|
|
|
|
tag_token = token.kind;
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token.kind != C_TOKEN_IDENTIFIER) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected identifier after '%s'",
|
|
|
|
token_spelling[token.kind]);
|
|
|
|
|
|
|
|
}
|
|
|
|
identifier = token.value;
|
|
|
|
identifier_len = token.len;
|
|
|
|
} else {
|
|
|
|
err = drgn_lexer_push(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (specifier == SPECIFIER_NONE) {
|
|
|
|
enum drgn_type_kind kind;
|
|
|
|
|
|
|
|
if (tag_token == C_TOKEN_STRUCT) {
|
|
|
|
kind = DRGN_TYPE_STRUCT;
|
|
|
|
} else if (tag_token == C_TOKEN_UNION) {
|
|
|
|
kind = DRGN_TYPE_UNION;
|
|
|
|
} else if (tag_token == C_TOKEN_ENUM) {
|
|
|
|
kind = DRGN_TYPE_ENUM;
|
|
|
|
} else if (identifier) {
|
2019-12-12 21:25:27 +00:00
|
|
|
if (strstartswith(identifier, "size_t")) {
|
2019-04-22 20:30:15 +01:00
|
|
|
err = drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_SIZE_T,
|
|
|
|
&ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-04-20 00:05:19 +01:00
|
|
|
ret->qualifiers = 0;
|
|
|
|
goto out;
|
2019-12-12 21:25:27 +00:00
|
|
|
} else if (strstartswith(identifier, "ptrdiff_t")) {
|
2019-04-22 20:30:15 +01:00
|
|
|
err = drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_PTRDIFF_T,
|
|
|
|
&ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-04-20 00:05:19 +01:00
|
|
|
ret->qualifiers = 0;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto out;
|
|
|
|
} else {
|
|
|
|
kind = DRGN_TYPE_TYPEDEF;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected type specifier");
|
|
|
|
}
|
|
|
|
|
2019-04-28 16:42:28 +01:00
|
|
|
err = drgn_type_index_find_parsed(tindex, kind, identifier,
|
|
|
|
identifier_len, filename,
|
|
|
|
ret);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
} else {
|
2019-04-22 20:30:15 +01:00
|
|
|
err = drgn_type_index_find_primitive(tindex,
|
|
|
|
specifier_kind[specifier],
|
|
|
|
&ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2019-04-20 00:05:19 +01:00
|
|
|
ret->qualifiers = 0;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
out:
|
|
|
|
ret->qualifiers |= qualifiers;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct c_declarator {
|
|
|
|
/* C_TOKEN_ASTERISK or C_TOKEN_LBRACKET. */
|
|
|
|
int kind;
|
|
|
|
enum drgn_qualifiers qualifiers;
|
|
|
|
/* Only for C_TOKEN_LBRACKET. */
|
|
|
|
bool is_complete;
|
|
|
|
uint64_t length;
|
|
|
|
struct c_declarator *next;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* These functions don't free the declarator list on error. */
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_abstract_declarator(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_lexer *lexer,
|
|
|
|
struct c_declarator **outer,
|
|
|
|
struct c_declarator **inner);
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_optional_type_qualifier_list(struct drgn_lexer *lexer,
|
|
|
|
enum drgn_qualifiers *qualifiers)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
*qualifiers = 0;
|
|
|
|
for (;;) {
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (token.kind < MIN_QUALIFIER_TOKEN ||
|
|
|
|
token.kind > MAX_QUALIFIER_TOKEN) {
|
|
|
|
err = drgn_lexer_push(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
*qualifiers |= qualifier_from_token[token.kind];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_pointer(struct drgn_type_index *tindex, struct drgn_lexer *lexer,
|
|
|
|
struct c_declarator **outer, struct c_declarator **inner)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token.kind != C_TOKEN_ASTERISK)
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX, "expected '*'");
|
|
|
|
|
|
|
|
*inner = NULL;
|
|
|
|
for (;;) {
|
|
|
|
struct c_declarator *tmp;
|
|
|
|
|
|
|
|
tmp = malloc(sizeof(*tmp));
|
|
|
|
if (!tmp)
|
|
|
|
return &drgn_enomem;
|
|
|
|
|
|
|
|
tmp->kind = C_TOKEN_ASTERISK;
|
|
|
|
tmp->next = *outer;
|
|
|
|
*outer = tmp;
|
|
|
|
|
|
|
|
err = c_parse_optional_type_qualifier_list(lexer,
|
|
|
|
&(*outer)->qualifiers);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (!*inner)
|
|
|
|
*inner = *outer;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token.kind != C_TOKEN_ASTERISK)
|
|
|
|
return drgn_lexer_push(lexer, &token);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_direct_abstract_declarator(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_lexer *lexer,
|
|
|
|
struct c_declarator **outer,
|
|
|
|
struct c_declarator **inner)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
*inner = NULL;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token.kind == C_TOKEN_LPAREN) {
|
|
|
|
struct drgn_token token2;
|
|
|
|
|
|
|
|
err = drgn_lexer_peek(lexer, &token2);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token2.kind == C_TOKEN_ASTERISK ||
|
|
|
|
token2.kind == C_TOKEN_LPAREN ||
|
|
|
|
token2.kind == C_TOKEN_LBRACKET) {
|
|
|
|
err = c_parse_abstract_declarator(tindex, lexer, outer,
|
|
|
|
inner);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
err = drgn_lexer_pop(lexer, &token2);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token2.kind != C_TOKEN_RPAREN) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected ')'");
|
|
|
|
}
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
if (token.kind == C_TOKEN_LBRACKET) {
|
|
|
|
struct c_declarator *tmp;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
tmp = malloc(sizeof(*tmp));
|
|
|
|
if (!tmp)
|
|
|
|
return &drgn_enomem;
|
|
|
|
|
|
|
|
tmp->kind = C_TOKEN_LBRACKET;
|
|
|
|
tmp->qualifiers = 0;
|
|
|
|
if (token.kind == C_TOKEN_NUMBER) {
|
|
|
|
tmp->is_complete = true;
|
|
|
|
err = c_token_to_u64(&token, &tmp->length);
|
|
|
|
if (err) {
|
|
|
|
free(tmp);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err) {
|
|
|
|
free(tmp);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
tmp->is_complete = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (*inner) {
|
|
|
|
tmp->next = (*inner)->next;
|
|
|
|
*inner = (*inner)->next = tmp;
|
|
|
|
} else {
|
|
|
|
tmp->next = *outer;
|
|
|
|
*outer = *inner = tmp;
|
|
|
|
}
|
|
|
|
if (token.kind != C_TOKEN_RBRACKET) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected ']'");
|
|
|
|
}
|
|
|
|
} else if (token.kind == C_TOKEN_LPAREN) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"function pointer types are not implemented");
|
|
|
|
} else {
|
|
|
|
err = drgn_lexer_push(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (!*inner) {
|
|
|
|
return drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected abstract declarator");
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *
|
|
|
|
c_parse_abstract_declarator(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_lexer *lexer,
|
|
|
|
struct c_declarator **outer,
|
|
|
|
struct c_declarator **inner)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
err = drgn_lexer_peek(lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (token.kind == C_TOKEN_ASTERISK) {
|
|
|
|
err = c_parse_pointer(tindex, lexer, outer, inner);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = drgn_lexer_peek(lexer, &token);
|
|
|
|
if (token.kind == C_TOKEN_LPAREN ||
|
|
|
|
token.kind == C_TOKEN_LBRACKET) {
|
|
|
|
struct c_declarator *tmp;
|
|
|
|
|
|
|
|
err = c_parse_direct_abstract_declarator(tindex, lexer,
|
|
|
|
outer, &tmp);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
} else {
|
|
|
|
return c_parse_direct_abstract_declarator(tindex, lexer, outer,
|
|
|
|
inner);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* This always frees the declarator list regardless of success or failure. */
|
|
|
|
static struct drgn_error *
|
|
|
|
c_type_from_declarator(struct drgn_type_index *tindex,
|
|
|
|
struct c_declarator *declarator,
|
|
|
|
struct drgn_qualified_type *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
|
|
|
if (!declarator)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
err = c_type_from_declarator(tindex, declarator->next, ret);
|
|
|
|
if (err) {
|
|
|
|
free(declarator);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (declarator->kind == C_TOKEN_ASTERISK) {
|
2019-04-21 10:10:33 +01:00
|
|
|
err = drgn_type_index_pointer_type(tindex, *ret, &ret->type);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else if (declarator->is_complete) {
|
|
|
|
err = drgn_type_index_array_type(tindex, declarator->length,
|
|
|
|
*ret, &ret->type);
|
|
|
|
} else {
|
|
|
|
err = drgn_type_index_incomplete_array_type(tindex, *ret,
|
|
|
|
&ret->type);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!err)
|
|
|
|
ret->qualifiers = declarator->qualifiers;
|
|
|
|
free(declarator);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_find_type(struct drgn_type_index *tindex,
|
|
|
|
const char *name, const char *filename,
|
|
|
|
struct drgn_qualified_type *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_lexer lexer;
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
drgn_lexer_init(&lexer, drgn_lexer_c, name);
|
|
|
|
|
|
|
|
err = c_parse_specifier_qualifier_list(tindex, &lexer, filename, ret);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(&lexer, &token);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
if (token.kind != C_TOKEN_EOF) {
|
|
|
|
struct c_declarator *outer = NULL, *inner;
|
|
|
|
|
|
|
|
err = drgn_lexer_push(&lexer, &token);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = c_parse_abstract_declarator(tindex, &lexer, &outer,
|
|
|
|
&inner);
|
|
|
|
if (err) {
|
|
|
|
while (outer) {
|
|
|
|
struct c_declarator *next;
|
|
|
|
|
|
|
|
next = outer->next;
|
|
|
|
free(outer);
|
|
|
|
outer = next;
|
|
|
|
}
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = c_type_from_declarator(tindex, outer, ret);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(&lexer, &token);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
if (token.kind != C_TOKEN_EOF) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"extra tokens after type name");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
err = NULL;
|
|
|
|
out:
|
|
|
|
drgn_lexer_deinit(&lexer);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_bit_offset(struct drgn_program *prog,
|
|
|
|
struct drgn_type *type,
|
|
|
|
const char *member_designator,
|
|
|
|
uint64_t *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_lexer lexer;
|
|
|
|
int state = INT_MIN;
|
|
|
|
uint64_t bit_offset = 0;
|
|
|
|
|
|
|
|
drgn_lexer_init(&lexer, drgn_lexer_c, member_designator);
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
struct drgn_token token;
|
|
|
|
|
|
|
|
err = drgn_lexer_pop(&lexer, &token);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
switch (state) {
|
|
|
|
case INT_MIN:
|
|
|
|
case C_TOKEN_DOT:
|
|
|
|
if (token.kind == C_TOKEN_IDENTIFIER) {
|
|
|
|
struct drgn_member_value *member;
|
|
|
|
struct drgn_qualified_type member_type;
|
|
|
|
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_find_member(&prog->tindex,
|
2019-04-22 22:21:25 +01:00
|
|
|
type,
|
|
|
|
token.value,
|
|
|
|
token.len,
|
|
|
|
&member);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
if (__builtin_add_overflow(bit_offset,
|
|
|
|
member->bit_offset,
|
|
|
|
&bit_offset)) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_OVERFLOW,
|
|
|
|
"offset is too large");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
err = drgn_lazy_type_evaluate(member->type,
|
|
|
|
&member_type);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
type = member_type.type;
|
|
|
|
} else if (state == C_TOKEN_DOT) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected identifier after '.'");
|
|
|
|
goto out;
|
|
|
|
} else {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected identifier");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case C_TOKEN_IDENTIFIER:
|
|
|
|
case C_TOKEN_RBRACKET:
|
|
|
|
switch (token.kind) {
|
|
|
|
case C_TOKEN_EOF:
|
|
|
|
*ret = bit_offset;
|
|
|
|
err = NULL;
|
|
|
|
goto out;
|
|
|
|
case C_TOKEN_DOT:
|
|
|
|
case C_TOKEN_LBRACKET:
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
if (state == C_TOKEN_IDENTIFIER) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected '.' or '[' after identifier");
|
|
|
|
goto out;
|
|
|
|
} else {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected '.' or '[' after ']'");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case C_TOKEN_LBRACKET:
|
|
|
|
if (token.kind == C_TOKEN_NUMBER) {
|
|
|
|
struct drgn_type *underlying_type;
|
|
|
|
struct drgn_type *element_type;
|
|
|
|
uint64_t index, bit_size, element_offset;
|
|
|
|
|
|
|
|
err = c_token_to_u64(&token, &index);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
underlying_type = drgn_underlying_type(type);
|
|
|
|
if (drgn_type_kind(underlying_type) != DRGN_TYPE_ARRAY) {
|
|
|
|
err = drgn_type_error("'%s' is not an array",
|
|
|
|
type);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
element_type =
|
|
|
|
drgn_type_type(underlying_type).type;
|
|
|
|
err = drgn_type_bit_size(element_type,
|
|
|
|
&bit_size);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
if (__builtin_mul_overflow(index, bit_size,
|
|
|
|
&element_offset) ||
|
|
|
|
__builtin_add_overflow(bit_offset,
|
|
|
|
element_offset,
|
|
|
|
&bit_offset)) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_OVERFLOW,
|
|
|
|
"offset is too large");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
type = element_type;
|
|
|
|
} else {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected number after '['");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case C_TOKEN_NUMBER:
|
|
|
|
if (token.kind != C_TOKEN_RBRACKET) {
|
|
|
|
err = drgn_error_create(DRGN_ERROR_SYNTAX,
|
|
|
|
"expected ']' after number");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
state = token.kind;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
drgn_lexer_deinit(&lexer);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_integer_literal(struct drgn_object *res,
|
|
|
|
uint64_t uvalue)
|
|
|
|
{
|
2019-04-22 20:30:15 +01:00
|
|
|
static const enum drgn_primitive_type types[] = {
|
|
|
|
DRGN_C_TYPE_INT,
|
|
|
|
DRGN_C_TYPE_LONG,
|
|
|
|
DRGN_C_TYPE_LONG_LONG,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_LONG_LONG,
|
|
|
|
};
|
|
|
|
struct drgn_error *err;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
unsigned int bits;
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_qualified_type qualified_type;
|
|
|
|
size_t i;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
_Static_assert(sizeof(unsigned long long) == 8,
|
|
|
|
"unsigned long long is not 64 bits");
|
|
|
|
bits = uvalue ? 64 - __builtin_clzll(uvalue) : 0;
|
|
|
|
|
|
|
|
qualified_type.qualifiers = 0;
|
2019-04-22 20:30:15 +01:00
|
|
|
for (i = 0; i < ARRAY_SIZE(types); i++) {
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_find_primitive(&res->prog->tindex,
|
2019-04-22 20:30:15 +01:00
|
|
|
types[i],
|
|
|
|
&qualified_type.type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (drgn_type_is_signed(qualified_type.type) &&
|
|
|
|
bits < 8 * drgn_type_size(qualified_type.type)) {
|
|
|
|
return drgn_object_set_signed(res,
|
|
|
|
qualified_type,
|
|
|
|
uvalue, 0);
|
|
|
|
} else if (!drgn_type_is_signed(qualified_type.type) &&
|
|
|
|
bits <= 8 * drgn_type_size(qualified_type.type)) {
|
|
|
|
return drgn_object_set_unsigned(res, qualified_type,
|
|
|
|
uvalue, 0);
|
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
2019-04-22 20:30:15 +01:00
|
|
|
return drgn_error_create(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"integer literal is too large");
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_bool_literal(struct drgn_object *res, bool bvalue)
|
|
|
|
{
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_qualified_type qualified_type;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_find_primitive(&res->prog->tindex,
|
2019-04-22 20:30:15 +01:00
|
|
|
DRGN_C_TYPE_INT,
|
|
|
|
&qualified_type.type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
qualified_type.qualifiers = 0;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return drgn_object_set_signed(res, qualified_type, bvalue, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_float_literal(struct drgn_object *res,
|
|
|
|
double fvalue)
|
|
|
|
{
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_qualified_type qualified_type;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_find_primitive(&res->prog->tindex,
|
2019-04-22 20:30:15 +01:00
|
|
|
DRGN_C_TYPE_DOUBLE,
|
|
|
|
&qualified_type.type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
qualified_type.qualifiers = 0;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return drgn_object_set_float(res, qualified_type, fvalue);
|
|
|
|
}
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
static const int c_integer_conversion_rank[] = {
|
|
|
|
[DRGN_C_TYPE_BOOL] = 0,
|
|
|
|
[DRGN_C_TYPE_CHAR] = 1,
|
|
|
|
[DRGN_C_TYPE_SIGNED_CHAR] = 1,
|
|
|
|
[DRGN_C_TYPE_UNSIGNED_CHAR] = 1,
|
|
|
|
[DRGN_C_TYPE_SHORT] = 2,
|
|
|
|
[DRGN_C_TYPE_UNSIGNED_SHORT] = 2,
|
|
|
|
[DRGN_C_TYPE_INT] = 3,
|
|
|
|
[DRGN_C_TYPE_UNSIGNED_INT] = 3,
|
|
|
|
[DRGN_C_TYPE_LONG] = 4,
|
|
|
|
[DRGN_C_TYPE_UNSIGNED_LONG] = 4,
|
|
|
|
[DRGN_C_TYPE_LONG_LONG] = 5,
|
|
|
|
[DRGN_C_TYPE_UNSIGNED_LONG_LONG] = 5,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
static bool c_can_represent_all_values(struct drgn_type *type1,
|
|
|
|
uint64_t bit_field_size1,
|
|
|
|
struct drgn_type *type2,
|
|
|
|
uint64_t bit_field_size2)
|
|
|
|
{
|
|
|
|
uint64_t width1, width2;
|
|
|
|
bool is_signed1, is_signed2;
|
|
|
|
|
|
|
|
if (drgn_type_kind(type1) == DRGN_TYPE_BOOL) {
|
|
|
|
width1 = 1;
|
|
|
|
is_signed1 = false;
|
|
|
|
} else {
|
|
|
|
width1 = (bit_field_size1 ? bit_field_size1 :
|
|
|
|
8 * drgn_type_size(type1));
|
|
|
|
is_signed1 = drgn_type_is_signed(type1);
|
|
|
|
}
|
|
|
|
if (drgn_type_kind(type2) == DRGN_TYPE_BOOL) {
|
|
|
|
width2 = 1;
|
|
|
|
is_signed2 = false;
|
|
|
|
} else {
|
|
|
|
width2 = (bit_field_size2 ? bit_field_size2 :
|
|
|
|
8 * drgn_type_size(type2));
|
|
|
|
is_signed2 = drgn_type_is_signed(type2);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_signed1 == is_signed2)
|
|
|
|
return width1 >= width2;
|
|
|
|
else if (is_signed1 && !is_signed2)
|
|
|
|
return width1 > width2;
|
|
|
|
else
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_integer_promotions(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_object_type *type)
|
|
|
|
{
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_error *err;
|
2019-04-20 00:05:19 +01:00
|
|
|
enum drgn_primitive_type primitive;
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_type *int_type;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
|
|
|
switch (drgn_type_kind(type->underlying_type)) {
|
|
|
|
case DRGN_TYPE_ENUM:
|
|
|
|
/* Convert the enum to its compatible type. */
|
|
|
|
type->type = type->underlying_type =
|
|
|
|
drgn_type_type(type->underlying_type).type;
|
|
|
|
if (!type->type) {
|
|
|
|
return drgn_error_format(DRGN_ERROR_INVALID_ARGUMENT,
|
|
|
|
"operand cannot have incomplete enum type");
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case DRGN_TYPE_INT:
|
|
|
|
case DRGN_TYPE_BOOL:
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
primitive = drgn_type_primitive(type->underlying_type);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
/*
|
|
|
|
* Integer promotions are performed on types whose integer conversion
|
|
|
|
* rank is less than or equal to the rank of int and unsigned int.
|
|
|
|
*
|
|
|
|
* If this isn't a standard integer type, then we don't know the rank,
|
|
|
|
* so we may need to promote it. According to the C standard, "the rank
|
|
|
|
* of a signed integer type shall be greater than the rank of any signed
|
|
|
|
* integer type with less precision", and "the rank of any standard
|
|
|
|
* integer type shall be greater than the rank of any extended integer
|
|
|
|
* type with the same width". If an extended signed integer type has
|
|
|
|
* less precision than int, or the same width as int, then all of its
|
|
|
|
* values can be represented by int (and likewise for an extended
|
|
|
|
* unsigned integer type and unsigned int). Therefore, an extended
|
|
|
|
* integer type should be promoted iff all of its values can be
|
|
|
|
* represented by int or unsigned int.
|
|
|
|
*
|
|
|
|
* Integer promotions are also performed on bit fields. The C standard
|
|
|
|
* only requires that bit fields of type _Bool, int, or unsigned int are
|
|
|
|
* supported, so it does not define how integer promotions should affect
|
|
|
|
* a bit field which cannot be represented by int or unsigned int. Clang
|
|
|
|
* promotes it to the full width, but GCC does not. We implement the GCC
|
|
|
|
* behavior of preserving the width.
|
|
|
|
*/
|
2019-04-22 20:30:15 +01:00
|
|
|
if (primitive >= ARRAY_SIZE(c_integer_conversion_rank) ||
|
|
|
|
type->bit_field_size) {
|
|
|
|
err = drgn_type_index_find_primitive(tindex, DRGN_C_TYPE_INT,
|
|
|
|
&int_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (c_can_represent_all_values(int_type, 0,
|
|
|
|
type->underlying_type,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
type->bit_field_size)) {
|
2019-04-22 20:30:15 +01:00
|
|
|
type->type = type->underlying_type = int_type;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
type->bit_field_size = 0;
|
2019-04-22 20:30:15 +01:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_INT,
|
|
|
|
&int_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (c_can_represent_all_values(int_type, 0,
|
|
|
|
type->underlying_type,
|
|
|
|
type->bit_field_size)) {
|
|
|
|
type->type = type->underlying_type = int_type;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
type->bit_field_size = 0;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
if (primitive == DRGN_C_TYPE_INT ||
|
|
|
|
primitive == DRGN_C_TYPE_UNSIGNED_INT ||
|
|
|
|
c_integer_conversion_rank[primitive] >
|
|
|
|
c_integer_conversion_rank[DRGN_C_TYPE_INT])
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If int can represent all values of the original type, then the result
|
|
|
|
* is int. Otherwise, the result is unsigned int.
|
|
|
|
*/
|
2019-04-22 20:30:15 +01:00
|
|
|
err = drgn_type_index_find_primitive(tindex, DRGN_C_TYPE_INT,
|
|
|
|
&int_type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (c_can_represent_all_values(int_type, 0, type->underlying_type, 0)) {
|
|
|
|
type->type = int_type;
|
|
|
|
} else {
|
|
|
|
err = drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_INT,
|
|
|
|
&type->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
type->underlying_type = type->type;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2019-04-22 20:30:15 +01:00
|
|
|
static struct drgn_error *
|
|
|
|
c_corresponding_unsigned_type(struct drgn_type_index *tindex,
|
|
|
|
enum drgn_primitive_type type,
|
|
|
|
struct drgn_type **ret)
|
|
|
|
{
|
|
|
|
switch (type) {
|
|
|
|
/*
|
|
|
|
* char, signed char, and short are promoted to int, so we don't need to
|
|
|
|
* handle them here.
|
|
|
|
*/
|
|
|
|
case DRGN_C_TYPE_INT:
|
|
|
|
return drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_INT,
|
|
|
|
ret);
|
|
|
|
case DRGN_C_TYPE_LONG:
|
|
|
|
return drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_LONG,
|
|
|
|
ret);
|
|
|
|
case DRGN_C_TYPE_LONG_LONG:
|
|
|
|
return drgn_type_index_find_primitive(tindex,
|
|
|
|
DRGN_C_TYPE_UNSIGNED_LONG_LONG,
|
|
|
|
ret);
|
|
|
|
default:
|
|
|
|
DRGN_UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
static struct drgn_error *c_common_real_type(struct drgn_type_index *tindex,
|
|
|
|
struct drgn_object_type *type1,
|
|
|
|
struct drgn_object_type *type2,
|
|
|
|
struct drgn_object_type *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
2019-04-20 00:05:19 +01:00
|
|
|
enum drgn_primitive_type primitive1, primitive2;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
bool is_float1, is_float2;
|
|
|
|
bool is_signed1, is_signed2;
|
|
|
|
int rank_cmp;
|
|
|
|
|
|
|
|
ret->qualifiers = 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Strictly, the rules are:
|
|
|
|
*
|
|
|
|
* If either operand is long double, then the result is long double.
|
|
|
|
* Otherwise, if either operand is double, then the result is double.
|
|
|
|
* Otherwise, if either operand is float, then the result is float.
|
|
|
|
*
|
|
|
|
* However, we also have to handle other floating types not in the
|
|
|
|
* standard. Thus, the result is always the larger type, with ties
|
|
|
|
* broken in the order unknown > long double > double > float.
|
|
|
|
*/
|
|
|
|
is_float1 = drgn_type_kind(type1->underlying_type) == DRGN_TYPE_FLOAT;
|
|
|
|
is_float2 = drgn_type_kind(type2->underlying_type) == DRGN_TYPE_FLOAT;
|
|
|
|
if (is_float1 && is_float2) {
|
|
|
|
uint64_t size1, size2;
|
|
|
|
|
|
|
|
size1 = drgn_type_size(type1->underlying_type);
|
|
|
|
size2 = drgn_type_size(type2->underlying_type);
|
|
|
|
if (size1 > size2)
|
|
|
|
goto ret1;
|
|
|
|
else if (size2 > size1)
|
|
|
|
goto ret2;
|
2019-04-20 00:05:19 +01:00
|
|
|
else if (drgn_type_primitive(type1->underlying_type) >
|
|
|
|
drgn_type_primitive(type2->underlying_type))
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto ret1;
|
|
|
|
else
|
|
|
|
goto ret2;
|
|
|
|
} else if (is_float1) {
|
|
|
|
goto ret1;
|
|
|
|
} else if (is_float2) {
|
|
|
|
goto ret2;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, the integer promotions are performed before applying the
|
|
|
|
* following rules.
|
|
|
|
*/
|
|
|
|
err = c_integer_promotions(tindex, type1);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
err = c_integer_promotions(tindex, type2);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
is_signed1 = drgn_type_is_signed(type1->underlying_type);
|
|
|
|
is_signed2 = drgn_type_is_signed(type2->underlying_type);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The C standard only requires that bit fields of type _Bool, int, or
|
|
|
|
* unsigned int are supported, which are always promoted to int or
|
|
|
|
* unsigned int, so it does not define how to find the common real type
|
|
|
|
* when one or both of the operands are bit fields. GCC seems to use the
|
|
|
|
* wider operand, or the unsigned operand if they have equal width. As
|
|
|
|
* usual, we pick type2 if the two types are equivalent.
|
|
|
|
*/
|
|
|
|
if (type1->bit_field_size || type2->bit_field_size) {
|
|
|
|
uint64_t width1, width2;
|
|
|
|
|
|
|
|
width1 = (type1->bit_field_size ? type1->bit_field_size :
|
|
|
|
8 * drgn_type_size(type1->type));
|
|
|
|
width2 = (type2->bit_field_size ? type2->bit_field_size :
|
|
|
|
8 * drgn_type_size(type2->type));
|
|
|
|
if (width1 < width2 ||
|
|
|
|
(width1 == width2 && (!is_signed2 || is_signed1)))
|
|
|
|
goto ret2;
|
|
|
|
else
|
|
|
|
goto ret1;
|
|
|
|
}
|
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
primitive1 = drgn_type_primitive(type1->underlying_type);
|
|
|
|
primitive2 = drgn_type_primitive(type2->underlying_type);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
2019-04-20 00:05:19 +01:00
|
|
|
if (primitive1 != DRGN_NOT_PRIMITIVE_TYPE &&
|
|
|
|
primitive2 != DRGN_NOT_PRIMITIVE_TYPE) {
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
/*
|
|
|
|
* If both operands have the same type, then no further
|
|
|
|
* conversions are needed.
|
|
|
|
*
|
|
|
|
* We can return either type1 or type2 here; it only makes a
|
|
|
|
* difference for typedefs. Arbitrarily pick type2 because
|
|
|
|
* that's what GCC seems to do (Clang always throws away the
|
|
|
|
* typedef).
|
|
|
|
*/
|
2019-04-20 00:05:19 +01:00
|
|
|
if (primitive1 == primitive2)
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
goto ret2;
|
|
|
|
|
|
|
|
/* Ranks are small, so this won't overflow. */
|
2019-04-20 00:05:19 +01:00
|
|
|
rank_cmp = (c_integer_conversion_rank[primitive1] -
|
|
|
|
c_integer_conversion_rank[primitive2]);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* We don't know the rank of non-standard integer types.
|
|
|
|
* However, we can usually compare their ranks, because
|
|
|
|
* according to the C standard, "the rank of a signed integer
|
|
|
|
* type shall be greater than the rank of any signed integer
|
|
|
|
* type with less precision", "the rank of any unsigned integer
|
|
|
|
* type shall equal the rank of the corresponding signed integer
|
|
|
|
* type", and "the rank of any standard integer type shall be
|
|
|
|
* greater than the rank of any extended integer type with the
|
|
|
|
* same width". The only case where we can't is if both types
|
|
|
|
* are non-standard and have the same size; we treat them as
|
|
|
|
* having equal rank in this case.
|
|
|
|
*/
|
|
|
|
uint64_t size1, size2;
|
|
|
|
|
|
|
|
size1 = drgn_type_size(type1->underlying_type);
|
|
|
|
size2 = drgn_type_size(type2->underlying_type);
|
2019-04-20 00:05:19 +01:00
|
|
|
if (size1 == size2 && primitive1 == DRGN_NOT_PRIMITIVE_TYPE &&
|
|
|
|
primitive2 == DRGN_NOT_PRIMITIVE_TYPE)
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
rank_cmp = 0;
|
2019-04-20 00:05:19 +01:00
|
|
|
else if ((size1 == size2 && primitive2 != DRGN_NOT_PRIMITIVE_TYPE) ||
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
size1 < size2)
|
|
|
|
rank_cmp = -1;
|
|
|
|
else
|
|
|
|
rank_cmp = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, if both operands have signed integer types or both have
|
|
|
|
* unsigned integer types, then the result is the type of the operand
|
|
|
|
* with the greater rank.
|
|
|
|
*/
|
|
|
|
if (is_signed1 == is_signed2) {
|
|
|
|
if (rank_cmp > 0)
|
|
|
|
goto ret1;
|
|
|
|
else
|
|
|
|
goto ret2;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, if the operand that has unsigned integer type has rank
|
|
|
|
* greater or equal to the rank of the type of the other operand, then
|
|
|
|
* the result is the unsigned integer type.
|
|
|
|
*/
|
|
|
|
if (!is_signed1 && rank_cmp >= 0)
|
|
|
|
goto ret1;
|
|
|
|
else if (!is_signed2 && rank_cmp <= 0)
|
|
|
|
goto ret2;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, if the type of the operand with signed integer type can
|
|
|
|
* represent all of the values of the type of the operand with unsigned
|
|
|
|
* integer type, then the result is the signed integer type.
|
|
|
|
*/
|
|
|
|
if (is_signed1 && c_can_represent_all_values(type1->underlying_type, 0,
|
|
|
|
type2->underlying_type, 0))
|
|
|
|
goto ret1;
|
|
|
|
if (is_signed2 && c_can_represent_all_values(type2->underlying_type, 0,
|
|
|
|
type1->underlying_type, 0))
|
|
|
|
goto ret2;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, the result is the unsigned integer type corresponding to
|
|
|
|
* the type of the operand with signed integer type.
|
|
|
|
*
|
|
|
|
* Note that this case is not reached for non-standard types: if the
|
|
|
|
* types have different signs and the signed integer type has greater
|
|
|
|
* rank, then it must have greater size and thus be able to represent
|
|
|
|
* all values of the unsigned integer type.
|
|
|
|
*/
|
2019-04-22 20:30:15 +01:00
|
|
|
err = c_corresponding_unsigned_type(tindex,
|
|
|
|
is_signed1 ? primitive1 : primitive2,
|
|
|
|
&ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
ret->underlying_type = ret->type;
|
|
|
|
ret->bit_field_size = 0;
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
ret1:
|
|
|
|
*ret = *type1;
|
|
|
|
return NULL;
|
|
|
|
ret2:
|
|
|
|
*ret = *type2;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_operand_type(const struct drgn_object *obj,
|
|
|
|
struct drgn_object_type *type_ret,
|
|
|
|
bool *is_pointer_ret,
|
|
|
|
uint64_t *referenced_size_ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
|
|
|
|
*type_ret = drgn_object_type(obj);
|
|
|
|
switch (drgn_type_kind(type_ret->underlying_type)) {
|
|
|
|
case DRGN_TYPE_ARRAY:
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_pointer_type(&obj->prog->tindex,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
drgn_type_type(type_ret->underlying_type),
|
|
|
|
&type_ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
type_ret->underlying_type = type_ret->type;
|
|
|
|
break;
|
|
|
|
case DRGN_TYPE_FUNCTION: {
|
|
|
|
struct drgn_qualified_type function_type = {
|
|
|
|
.type = type_ret->underlying_type,
|
|
|
|
.qualifiers = type_ret->qualifiers,
|
|
|
|
};
|
|
|
|
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_pointer_type(&obj->prog->tindex,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
function_type,
|
|
|
|
&type_ret->type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
type_ret->underlying_type = type_ret->type;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
type_ret->qualifiers = 0;
|
|
|
|
|
|
|
|
if (is_pointer_ret) {
|
|
|
|
struct drgn_type *type;
|
|
|
|
|
|
|
|
type = type_ret->underlying_type;
|
|
|
|
*is_pointer_ret = drgn_type_kind(type) == DRGN_TYPE_POINTER;
|
2019-04-12 23:25:21 +01:00
|
|
|
if (*is_pointer_ret && referenced_size_ret) {
|
2019-04-13 00:06:24 +01:00
|
|
|
struct drgn_type *referenced_type;
|
|
|
|
|
|
|
|
referenced_type =
|
|
|
|
drgn_underlying_type(drgn_type_type(type).type);
|
|
|
|
if (drgn_type_kind(referenced_type) == DRGN_TYPE_VOID) {
|
|
|
|
*referenced_size_ret = 1;
|
|
|
|
} else {
|
|
|
|
err = drgn_type_sizeof(referenced_type,
|
|
|
|
referenced_size_ret);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2019-04-13 00:37:28 +01:00
|
|
|
static struct drgn_error *c_op_cast(struct drgn_object *res,
|
|
|
|
struct drgn_qualified_type qualified_type,
|
|
|
|
const struct drgn_object *obj)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_object_type type;
|
|
|
|
|
|
|
|
err = c_operand_type(obj, &type, NULL, NULL);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
return drgn_op_cast(res, qualified_type, obj, &type);
|
|
|
|
}
|
|
|
|
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
/*
|
|
|
|
* It's too expensive to check that two pointer types are compatible, so we just
|
|
|
|
* check that they refer to the same kind of type with equal size.
|
|
|
|
*/
|
|
|
|
static bool c_pointers_similar(const struct drgn_object_type *lhs_type,
|
|
|
|
const struct drgn_object_type *rhs_type,
|
|
|
|
uint64_t lhs_size, uint64_t rhs_size)
|
|
|
|
{
|
|
|
|
struct drgn_type *lhs_referenced_type, *rhs_referenced_type;
|
|
|
|
|
|
|
|
lhs_referenced_type = drgn_type_type(lhs_type->underlying_type).type;
|
|
|
|
rhs_referenced_type = drgn_type_type(rhs_type->underlying_type).type;
|
|
|
|
return (drgn_type_kind(lhs_referenced_type) ==
|
|
|
|
drgn_type_kind(rhs_referenced_type) && lhs_size == rhs_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_op_bool(const struct drgn_object *obj, bool *ret)
|
|
|
|
{
|
|
|
|
struct drgn_type *underlying_type;
|
|
|
|
|
|
|
|
underlying_type = drgn_underlying_type(obj->type);
|
|
|
|
if (drgn_type_kind(underlying_type) == DRGN_TYPE_ARRAY) {
|
|
|
|
*ret = true;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!drgn_type_is_scalar(underlying_type)) {
|
|
|
|
return drgn_qualified_type_error("cannot convert '%s' to bool",
|
|
|
|
drgn_object_qualified_type(obj));
|
|
|
|
}
|
|
|
|
|
|
|
|
return drgn_object_truthiness(obj, ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_op_cmp(const struct drgn_object *lhs,
|
|
|
|
const struct drgn_object *rhs, int *ret)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_object_type lhs_type, rhs_type;
|
|
|
|
bool lhs_pointer, rhs_pointer;
|
|
|
|
|
2019-04-12 23:25:21 +01:00
|
|
|
err = c_operand_type(lhs, &lhs_type, &lhs_pointer, NULL);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err)
|
|
|
|
return err;
|
2019-04-12 23:25:21 +01:00
|
|
|
err = c_operand_type(rhs, &rhs_type, &rhs_pointer, NULL);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (lhs_pointer && rhs_pointer) {
|
|
|
|
return drgn_op_cmp_pointers(lhs, rhs, ret);
|
|
|
|
} else if (lhs_pointer || rhs_pointer) {
|
|
|
|
goto type_error;
|
|
|
|
} else {
|
|
|
|
struct drgn_object_type type;
|
|
|
|
|
|
|
|
if (!drgn_type_is_arithmetic(lhs_type.underlying_type) ||
|
|
|
|
!drgn_type_is_arithmetic(rhs_type.underlying_type))
|
|
|
|
goto type_error;
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_common_real_type(&lhs->prog->tindex, &lhs_type,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
&rhs_type, &type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return drgn_op_cmp_impl(lhs, rhs, &type, ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
type_error:
|
|
|
|
return drgn_error_binary_op("comparison", &lhs_type, &rhs_type);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_op_add(struct drgn_object *res,
|
|
|
|
const struct drgn_object *lhs,
|
|
|
|
const struct drgn_object *rhs)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_object_type lhs_type, rhs_type;
|
|
|
|
bool lhs_pointer, rhs_pointer;
|
|
|
|
uint64_t lhs_size, rhs_size;
|
|
|
|
|
|
|
|
err = c_operand_type(lhs, &lhs_type, &lhs_pointer, &lhs_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
err = c_operand_type(rhs, &rhs_type, &rhs_pointer, &rhs_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (lhs_pointer) {
|
|
|
|
if (!drgn_type_is_integer(rhs_type.underlying_type))
|
|
|
|
goto type_error;
|
|
|
|
return drgn_op_add_to_pointer(res, &lhs_type, lhs_size, false, lhs, rhs);
|
|
|
|
} else if (rhs_pointer) {
|
|
|
|
if (!drgn_type_is_integer(lhs_type.underlying_type))
|
|
|
|
goto type_error;
|
|
|
|
return drgn_op_add_to_pointer(res, &rhs_type, rhs_size, false, rhs, lhs);
|
|
|
|
} else {
|
|
|
|
struct drgn_object_type type;
|
|
|
|
|
|
|
|
if (!drgn_type_is_arithmetic(lhs_type.underlying_type) ||
|
|
|
|
!drgn_type_is_arithmetic(rhs_type.underlying_type))
|
|
|
|
goto type_error;
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_common_real_type(&lhs->prog->tindex, &lhs_type,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
&rhs_type, &type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return drgn_op_add_impl(res, &type, lhs, rhs);
|
|
|
|
}
|
|
|
|
|
|
|
|
type_error:
|
|
|
|
return drgn_error_binary_op("binary +", &lhs_type, &rhs_type);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct drgn_error *c_op_sub(struct drgn_object *res,
|
|
|
|
const struct drgn_object *lhs,
|
|
|
|
const struct drgn_object *rhs)
|
|
|
|
{
|
|
|
|
struct drgn_error *err;
|
|
|
|
struct drgn_object_type lhs_type, rhs_type;
|
|
|
|
bool lhs_pointer, rhs_pointer;
|
|
|
|
uint64_t lhs_size, rhs_size;
|
|
|
|
|
|
|
|
err = c_operand_type(lhs, &lhs_type, &lhs_pointer, &lhs_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
err = c_operand_type(rhs, &rhs_type, &rhs_pointer, &rhs_size);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (lhs_pointer && rhs_pointer) {
|
2019-04-22 20:30:15 +01:00
|
|
|
struct drgn_object_type type = {};
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
|
2019-05-01 19:22:59 +01:00
|
|
|
err = drgn_type_index_find_primitive(&lhs->prog->tindex,
|
2019-04-22 20:30:15 +01:00
|
|
|
DRGN_C_TYPE_PTRDIFF_T,
|
|
|
|
&type.type);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
type.underlying_type = drgn_underlying_type(type.type);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (!c_pointers_similar(&lhs_type, &rhs_type, lhs_size,
|
|
|
|
rhs_size))
|
|
|
|
goto type_error;
|
|
|
|
return drgn_op_sub_pointers(res, &type, lhs_size, lhs, rhs);
|
|
|
|
} else if (lhs_pointer) {
|
|
|
|
if (!drgn_type_is_integer(rhs_type.underlying_type))
|
|
|
|
goto type_error;
|
|
|
|
return drgn_op_add_to_pointer(res, &lhs_type, lhs_size, true,
|
|
|
|
lhs, rhs);
|
|
|
|
} else {
|
|
|
|
struct drgn_object_type type;
|
|
|
|
|
|
|
|
if (!drgn_type_is_arithmetic(lhs_type.underlying_type) ||
|
|
|
|
!drgn_type_is_arithmetic(rhs_type.underlying_type))
|
|
|
|
goto type_error;
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_common_real_type(&lhs->prog->tindex, &lhs_type,
|
|
|
|
&rhs_type, &type);
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return drgn_op_sub_impl(res, &type, lhs, rhs);
|
|
|
|
}
|
|
|
|
|
|
|
|
type_error:
|
|
|
|
return drgn_error_binary_op("binary -", &lhs_type, &rhs_type);
|
|
|
|
}
|
|
|
|
|
|
|
|
#define BINARY_OP(op_name, op, check) \
|
|
|
|
static struct drgn_error *c_op_##op_name(struct drgn_object *res, \
|
|
|
|
const struct drgn_object *lhs, \
|
|
|
|
const struct drgn_object *rhs) \
|
|
|
|
{ \
|
|
|
|
struct drgn_error *err; \
|
|
|
|
struct drgn_object_type lhs_type, rhs_type, type; \
|
|
|
|
\
|
|
|
|
err = c_operand_type(lhs, &lhs_type, NULL, NULL); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
err = c_operand_type(rhs, &rhs_type, NULL, NULL); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
if (!drgn_type_is_##check(lhs_type.underlying_type) || \
|
|
|
|
!drgn_type_is_##check(rhs_type.underlying_type)) \
|
|
|
|
return drgn_error_binary_op("binary "#op, &lhs_type, \
|
|
|
|
&rhs_type); \
|
|
|
|
\
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_common_real_type(&lhs->prog->tindex, &lhs_type, &rhs_type, \
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
&type); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
\
|
|
|
|
return drgn_op_##op_name##_impl(res, &type, lhs, rhs); \
|
|
|
|
}
|
|
|
|
BINARY_OP(mul, *, arithmetic)
|
|
|
|
BINARY_OP(div, /, arithmetic)
|
|
|
|
BINARY_OP(mod, %, integer)
|
|
|
|
BINARY_OP(and, &, integer)
|
|
|
|
BINARY_OP(or, |, integer)
|
|
|
|
BINARY_OP(xor, ^, integer)
|
|
|
|
#undef BINARY_OP
|
|
|
|
|
|
|
|
#define SHIFT_OP(op_name, op) \
|
|
|
|
static struct drgn_error *c_op_##op_name(struct drgn_object *res, \
|
|
|
|
const struct drgn_object *lhs, \
|
|
|
|
const struct drgn_object *rhs) \
|
|
|
|
{ \
|
|
|
|
struct drgn_error *err; \
|
|
|
|
struct drgn_object_type lhs_type, rhs_type; \
|
|
|
|
\
|
|
|
|
err = c_operand_type(lhs, &lhs_type, NULL, NULL); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
err = c_operand_type(rhs, &rhs_type, NULL, NULL); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
if (!drgn_type_is_integer(lhs_type.underlying_type) || \
|
|
|
|
!drgn_type_is_integer(rhs_type.underlying_type)) \
|
|
|
|
return drgn_error_binary_op("binary " #op, &lhs_type, \
|
|
|
|
&rhs_type); \
|
|
|
|
\
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_integer_promotions(&lhs->prog->tindex, &lhs_type); \
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err) \
|
|
|
|
return err; \
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_integer_promotions(&lhs->prog->tindex, &rhs_type); \
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
\
|
|
|
|
return drgn_op_##op_name##_impl(res, lhs, &lhs_type, rhs, &rhs_type); \
|
|
|
|
}
|
|
|
|
SHIFT_OP(lshift, <<)
|
|
|
|
SHIFT_OP(rshift, >>)
|
|
|
|
#undef SHIFT_OP
|
|
|
|
|
|
|
|
#define UNARY_OP(op_name, op, check) \
|
|
|
|
static struct drgn_error *c_op_##op_name(struct drgn_object *res, \
|
|
|
|
const struct drgn_object *obj) \
|
|
|
|
{ \
|
|
|
|
struct drgn_error *err; \
|
|
|
|
struct drgn_object_type type; \
|
|
|
|
\
|
|
|
|
err = c_operand_type(obj, &type, NULL, NULL); \
|
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
if (!drgn_type_is_##check(type.underlying_type)) \
|
|
|
|
return drgn_error_unary_op("unary " #op, &type); \
|
|
|
|
\
|
2019-05-01 19:22:59 +01:00
|
|
|
err = c_integer_promotions(&obj->prog->tindex, &type); \
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
if (err) \
|
|
|
|
return err; \
|
|
|
|
\
|
|
|
|
return drgn_op_##op_name##_impl(res, &type, obj); \
|
|
|
|
}
|
|
|
|
UNARY_OP(pos, +, arithmetic)
|
|
|
|
UNARY_OP(neg, -, arithmetic)
|
|
|
|
UNARY_OP(not, ~, integer)
|
|
|
|
#undef UNARY_OP
|
|
|
|
|
|
|
|
const struct drgn_language drgn_language_c = {
|
|
|
|
.name = "C",
|
|
|
|
.pretty_print_type = c_pretty_print_type,
|
|
|
|
.pretty_print_type_name = c_pretty_print_type_name,
|
|
|
|
.pretty_print_object = c_pretty_print_object,
|
|
|
|
.find_type = c_find_type,
|
|
|
|
.bit_offset = c_bit_offset,
|
|
|
|
.integer_literal = c_integer_literal,
|
|
|
|
.bool_literal = c_bool_literal,
|
|
|
|
.float_literal = c_float_literal,
|
2019-04-13 00:37:28 +01:00
|
|
|
.op_cast = c_op_cast,
|
Rewrite drgn core in C
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.
2019-03-22 23:27:46 +00:00
|
|
|
.op_bool = c_op_bool,
|
|
|
|
.op_cmp = c_op_cmp,
|
|
|
|
.op_add = c_op_add,
|
|
|
|
.op_sub = c_op_sub,
|
|
|
|
.op_mul = c_op_mul,
|
|
|
|
.op_div = c_op_div,
|
|
|
|
.op_mod = c_op_mod,
|
|
|
|
.op_lshift = c_op_lshift,
|
|
|
|
.op_rshift = c_op_rshift,
|
|
|
|
.op_and = c_op_and,
|
|
|
|
.op_or = c_op_or,
|
|
|
|
.op_xor = c_op_xor,
|
|
|
|
.op_pos = c_op_pos,
|
|
|
|
.op_neg = c_op_neg,
|
|
|
|
.op_not = c_op_not,
|
|
|
|
};
|