Naive concatenation of $LD_LIBRARY_PATH can result in an empty colon-delimited segment; this tells glibc to load libraries from the current directory, which is definitely wrong, and may be a security vulnerability if the current directory is untrusted. (See #67234, for example.) Fix this throughout the tree. Followup to #76804. Fixes #144646. Signed-off-by: Anders Kaseorg <andersk@mit.edu>
9.0 KiB
Overlays
This chapter describes how to extend and change Nixpkgs using overlays. Overlays are used to add layers in the fixed-point used by Nixpkgs to compose the set of all packages.
Nixpkgs can be configured with a list of overlays, which are applied in order. This means that the order of the overlays can be significant if multiple layers override the same package.
Installing overlays
The list of overlays can be set either explicitly in a Nix expression, or through <nixpkgs-overlays>
or user configuration files.
Set overlays in NixOS or Nix expressions
On a NixOS system the value of the nixpkgs.overlays
option, if present, is passed to the system Nixpkgs directly as an argument. Note that this does not affect the overlays for non-NixOS operations (e.g. nix-env
), which are looked up independently.
The list of overlays can be passed explicitly when importing nixpkgs, for example import <nixpkgs> { overlays = [ overlay1 overlay2 ]; }
.
NOTE: DO NOT USE THIS in nixpkgs. Further overlays can be added by calling the pkgs.extend
or pkgs.appendOverlays
, although it is often preferable to avoid these functions, because they recompute the Nixpkgs fixpoint, which is somewhat expensive to do.
Install overlays via configuration lookup
The list of overlays is determined as follows.
-
First, if an
overlays
argument to the Nixpkgs function itself is given, then that is used and no path lookup will be performed. -
Otherwise, if the Nix path entry
<nixpkgs-overlays>
exists, we look for overlays at that path, as described below.See the section on
NIX_PATH
in the Nix manual for more details on how to set a value for<nixpkgs-overlays>.
-
If one of
~/.config/nixpkgs/overlays.nix
and~/.config/nixpkgs/overlays/
exists, then we look for overlays at that path, as described below. It is an error if both exist.
If we are looking for overlays at a path, then there are two cases:
-
If the path is a file, then the file is imported as a Nix expression and used as the list of overlays.
-
If the path is a directory, then we take the content of the directory, order it lexicographically, and attempt to interpret each as an overlay by:
-
Importing the file, if it is a
.nix
file. -
Importing a top-level
default.nix
file, if it is a directory.
-
Because overlays that are set in NixOS configuration do not affect non-NixOS operations such as nix-env
, the overlays.nix
option provides a convenient way to use the same overlays for a NixOS system configuration and user configuration: the same file can be used as overlays.nix
and imported as the value of nixpkgs.overlays
.
Defining overlays
Overlays are Nix functions which accept two arguments, conventionally called self
and super
, and return a set of packages. For example, the following is a valid overlay.
self: super:
{
boost = super.boost.override {
python = self.python3;
};
rr = super.callPackage ./pkgs/rr {
stdenv = self.stdenv_32bit;
};
}
The first argument (self
) corresponds to the final package set. You should use this set for the dependencies of all packages specified in your overlay. For example, all the dependencies of rr
in the example above come from self
, as well as the overridden dependencies used in the boost
override.
The second argument (super
) corresponds to the result of the evaluation of the previous stages of Nixpkgs. It does not contain any of the packages added by the current overlay, nor any of the following overlays. This set should be used either to refer to packages you wish to override, or to access functions defined in Nixpkgs. For example, the original recipe of boost
in the above example, comes from super
, as well as the callPackage
function.
The value returned by this function should be a set similar to pkgs/top-level/all-packages.nix
, containing overridden and/or new packages.
Overlays are similar to other methods for customizing Nixpkgs, in particular the packageOverrides
attribute described in . Indeed, packageOverrides
acts as an overlay with only the super
argument. It is therefore appropriate for basic use, but overlays are more powerful and easier to distribute.
Using overlays to configure alternatives
Certain software packages have different implementations of the same interface. Other distributions have functionality to switch between these. For example, Debian provides DebianAlternatives. Nixpkgs has what we call alternatives
, which are configured through overlays.
BLAS/LAPACK
In Nixpkgs, we have multiple implementations of the BLAS/LAPACK numerical linear algebra interfaces. They are:
-
The Nixpkgs attribute is
openblas
for ILP64 (integer width = 64 bits) andopenblasCompat
for LP64 (integer width = 32 bits).openblasCompat
is the default. -
LAPACK reference (also provides BLAS)
The Nixpkgs attribute is
lapack-reference
. -
Intel MKL (only works on the x86_64 architecture, unfree)
The Nixpkgs attribute is
mkl
. -
BLIS, available through the attribute
blis
, is a framework for linear algebra kernels. In addition, it implements the BLAS interface. -
AMD BLIS/LIBFLAME (optimized for modern AMD x86_64 CPUs)
The AMD fork of the BLIS library, with attribute
amd-blis
, extends BLIS with optimizations for modern AMD CPUs. The changes are usually submitted to the upstream BLIS project after some time. However, AMD BLIS typically provides some performance improvements on AMD Zen CPUs. The complementary AMD LIBFLAME library, with attributeamd-libflame
, provides a LAPACK implementation.
Introduced in PR #83888, we are able to override the blas
and lapack
packages to use different implementations, through the blasProvider
and lapackProvider
argument. This can be used to select a different provider. BLAS providers will have symlinks in $out/lib/libblas.so.3
and $out/lib/libcblas.so.3
to their respective BLAS libraries. Likewise, LAPACK providers will have symlinks in $out/lib/liblapack.so.3
and $out/lib/liblapacke.so.3
to their respective LAPACK libraries. For example, Intel MKL is both a BLAS and LAPACK provider. An overlay can be created to use Intel MKL that looks like:
self: super:
{
blas = super.blas.override {
blasProvider = self.mkl;
};
lapack = super.lapack.override {
lapackProvider = self.mkl;
};
}
This overlay uses Intel's MKL library for both BLAS and LAPACK interfaces. Note that the same can be accomplished at runtime using LD_LIBRARY_PATH
of libblas.so.3
and liblapack.so.3
. For instance:
$ LD_LIBRARY_PATH=$(nix-build -A mkl)/lib${LD_LIBRARY_PATH:+:}$LD_LIBRARY_PATH nix-shell -p octave --run octave
Intel MKL requires an openmp
implementation when running with multiple processors. By default, mkl
will use Intel's iomp
implementation if no other is specified, but this is a runtime-only dependency and binary compatible with the LLVM implementation. To use that one instead, Intel recommends users set it with LD_PRELOAD
. Note that mkl
is only available on x86_64-linux
and x86_64-darwin
. Moreover, Hydra is not building and distributing pre-compiled binaries using it.
For BLAS/LAPACK switching to work correctly, all packages must depend on blas
or lapack
. This ensures that only one BLAS/LAPACK library is used at one time. There are two versions of BLAS/LAPACK currently in the wild, LP64
(integer size = 32 bits) and ILP64
(integer size = 64 bits). Some software needs special flags or patches to work with ILP64
. You can check if ILP64
is used in Nixpkgs with blas.isILP64
and lapack.isILP64
. Some software does NOT work with ILP64
, and derivations need to specify an assertion to prevent this. You can prevent ILP64
from being used with the following:
{ stdenv, blas, lapack, ... }:
assert (!blas.isILP64) && (!lapack.isILP64);
stdenv.mkDerivation {
...
}
Switching the MPI implementation
All programs that are built with MPI support use the generic attribute mpi
as an input. At the moment Nixpkgs natively provides two different MPI implementations:
-
Open MPI (default), attribute name
openmpi
-
MPICH, attribute name
mpich
-
MVAPICH, attribute name
mvapich
To provide MPI enabled applications that use MPICH
, instead of the default Open MPI
, simply use the following overlay:
self: super:
{
mpi = self.mpich;
}