When a PEP 517 project file is present, pip will not install
prerequisites in `site-packages`:
https://pip.pypa.io/en/stable/reference/pip/#pep-517-and-518-support
For the shell hook, this has the consequence that the generated
temporary directory that is added to PYTHONPATH does not contain
`site.py`. As a result, Python does not discover the Python
module. Thus when a user executes nix-shell in a project, they cannot
import the project's Python module.
This change adds the `--no-build-isolation` option to pip when
creating the editable environment, to correctly generate `site.py`,
even when a `pyproject.toml` is present.
When a PEP 517 project file is present, pip will not install
prerequisites in `site-packages`:
https://pip.pypa.io/en/stable/reference/pip/#pep-517-and-518-support
For the shell hook, this has the consequence that the generated
temporary directory that is added to PYTHONPATH does not contain
`site.py`. As a result, Python does not discover the Python
module. Thus when a user executes nix-shell in a project, they cannot
import the project's Python module.
This change adds the `--no-build-isolation` option to pip when
creating the editable environment, to correctly generate `site.py`,
even when a `pyproject.toml` is present.
I took a close look at how Debian builds the Python interpreter,
because I noticed it ran substantially faster than the one in nixpkgs
and I was curious why.
One thing that I found made a material difference in performance was
this pair of linker flags (passed to the compiler):
-Wl,-O1 -Wl,-Bsymbolic-functions
In other words, effectively the linker gets passed the flags:
-O1 -Bsymbolic-functions
Doing the same thing in nixpkgs turns out to make the interpreter
run about 6% faster, which is quite a big win for such an easy
change. So, let's apply it.
---
I had not known there was a `-O1` flag for the *linker*!
But indeed there is.
These flags are unrelated to "link-time optimization" (LTO), despite
the latter's name. LTO means doing classic compiler optimizations
on the actual code, at the linking step when it becomes possible to
do them with cross-object-file information. These two flags, by
contrast, cause the linker to make certain optimizations within the
scope of its job as the linker.
Documentation is here, though sparse:
https://sourceware.org/binutils/docs-2.31/ld/Options.html
The meaning of -O1 was explained in more detail in this LWN article:
https://lwn.net/Articles/192624/
Apparently it makes the resulting symbol table use a bigger hash
table, so the load factor is smaller and lookups are faster. Cool.
As for -Bsymbolic-functions, the documentation indicates that it's a
way of saving lookups through the symbol table entirely. There can
apparently be situations where it changes the behavior of a program,
specifically if the program relies on linker tricks to provide
customization features:
https://bugs.launchpad.net/ubuntu/+source/xfe/+bug/644645https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=637184#35
But I'm pretty sure CPython doesn't permit that kind of trick: you
don't load a shared object that tries to redefine some symbol found
in the interpreter core.
The stronger reason I'm confident using -Bsymbolic-functions is
safe, though, is empirical. Both Debian and Ubuntu have been
shipping a Python built this way since forever -- it was introduced
for the Python 2.4 and 2.5 in Ubuntu "hardy", and Debian "lenny",
released in 2008 and 2009. In those 12 years they haven't seen a
need to drop this flag; and I've been unable to locate any reports
of trouble related to it, either on the Web in general or on the
Debian bug tracker. (There are reports of a handful of other
programs breaking with it, but not Python/CPython.) So that seems
like about as thorough testing as one could hope for.
---
As for the performance impact: I ran CPython upstream's preferred
benchmark suite, "pyperformance", in the same way as described in
the previous commit. On top of that commit's change, the results
across the 60 benchmarks in the suite are:
The median is 6% faster.
The middle half (aka interquartile range) is from 4% to 8% faster.
Out of 60 benchmarks, 3 come out slower, by 1-4%. At the other end,
5 are at least 10% faster, and one is 17% faster.
So, that's quite a material speedup! I don't know how big the
effect of these flags is for other software; but certainly CPython
tends to do plenty of dynamic linking, as that's how it loads
extension modules, which are ubiquitous in the stdlib as well as
popular third-party libraries. So perhaps that helps explain why
optimizing the dynamic linker has such an impact.
Without this flag, the configure script prints a warning at the end,
like this (reformatted):
If you want a release build with all stable optimizations active
(PGO, etc), please run ./configure --enable-optimizations
We're doing a build to distribute to people for day-to-day use,
doing things other than developing the Python interpreter. So
that's certainly a release build -- we're the target audience for
this recommendation.
---
And, trying it out, upstream isn't kidding! I ran the standard
benchmark suite that the CPython developers use for performance
work, "pyperformance". Following its usage instructions:
https://pyperformance.readthedocs.io/usage.html
I ran the whole suite, like so:
$ nix-shell -p ./result."$variant" --run '
cd $(mktemp -d); python -m venv venv; . venv/bin/activate
pip install pyperformance
pyperformance run -o ~/tmp/result.'"$variant"'.json
'
and then examined the results with commands like:
$ python -m pyperf compare_to --table -G \
~/tmp/result.{$before,$after}.json
Across all the benchmarks in the suite, the median speedup was 16%.
(Meaning 1.16x faster; 14% less time).
The middle half of them ranged from a 13% to a 22% speedup.
Each of the 60 benchmarks in the suite got faster, by speedups
ranging from 3% to 53%.
---
One reason this isn't just the default to begin with is that, until
recently, it made the build a lot slower. What it does is turn on
profile-guided optimization, which means first build for profiling,
then run some task to get a profile, then build again using the
profile. And, short of further customization, the task it would use
would be nearly the full test suite, which includes a lot of
expensive and slow tests, and can easily take half an hour to run.
Happily, in 2019 an upstream developer did the work to carefully
select a more appropriate set of tests to use for the profile:
https://github.com/python/cpython/commit/4e16a4a31https://bugs.python.org/issue36044
This suite takes just 2 minutes to run. And the resulting final
build is actually slightly faster than with the much longer suite,
at least as measured by those standard "pyperformance" benchmarks.
That work went into the 3.8 release, but the same list works great
if used on older releases too.
So, start passing that --enable-optimizations flag; and backport
that good-for-PGO set of tests, so that we use it on all releases.
This should enable (manual) building of RPMs from python projects using
the `python setup.py bdist_rpm` command on systems where `rpmbuild` is
not located in `/usr/bin/`. (e.g. NixOS)
The discovery of the rpmbuild command was fixed upstream in Python 3.8,
so this commit backports the relevant patch to our currently supported
Python 3 versions.
Fixes: #85204
The ./configure script prints a warning when passed this flag,
starting with 3.7:
configure: WARNING: unrecognized options: --with-threads
The reason is that there's no longer such a thing as a build
without threads.
Eliminate the warning, by only passing the flag on the older releases
that accept it.
Upstream change and discussion:
https://github.com/python/cpython/commit/a6a4dc816https://bugs.python.org/issue31370
The prefix will now be correct in case of Nix env.
Note, however, that creating a venv from a Nix env still does not function. This does not seem to be possible
with the current approach either, because venv will copy or symlink our Python wrapper. In case it symlinks
(the default) it won't see a pyvenv.cfg. If it is copied I think it should function but it does not...
This is needed in case of `python.buildEnv` to make sure site.PREFIXES
does not only point to the unwrapped executable prefix.
--------------------------------------------------------------------------------
This PR is a story where your valiant hero sets out on a very simple adventure but ends up having to slay dragons, starts questioning his own sanity and finally manages to gain enough knowledge to slay the evil dragon and finally win the proverbial price.
It all started out on sunny spring day with trying to tackle the Nixops plugin infrastructure and make that nice enough to work with.
Our story begins in the shanty town of [NixOps-AWS](https://github.com/nixos/nixops-aws) where [mypy](http://mypy-lang.org/) type checking has not yet been seen.
As our deuteragonist (@grahamc) has made great strides in the capital city of [NixOps](https://github.com/nixos/nixops) our hero wanted to bring this out into the land and let the people rejoice in reliability and a wonderful development experience.
The plugin work itself was straight forward and our hero quickly slayed the first small dragon, at this point things felt good and our hero thought he was going to reach the town of NixOps-AWS very quickly.
But alas! Mypy did not want to go, it said:
`Cannot find implementation or library stub for module named 'nixops'`
Our hero felt a small sliver of life escape from his body. Things were not going to be so easy.
After some frustration our hero discovered there was a [rule of the land of Python](https://www.python.org/dev/peps/pep-0561/) that governed the import of types into the kingdom, more specificaly a very special document (file) called `py.typed`.
Things were looking good.
But no, what the law said did not seem to match reality. How could things be so?
After some frustrating debugging our valiant hero thought to himself "Hmm, I wonder if this is simply a Nix idiosyncrasy", and it turns out indeed it was.
Things that were working in the blessed way of the land of Python (inside a `virtualenv`) were not working the way they were from his home town of Nix (`nix-shell` + `python.withPackages`).
After even more frustrating attempts at reading the mypy documentation and trying to understand how things were supposed to work our hero started questioning his sanity.
This is where things started to get truly interesting.
Our hero started to use a number of powerful weapons, both forged in the land of Python (pdb) & by the mages of UNIX (printf-style-debugging & strace).
After first trying to slay the dragon simply by `strace` and a keen eye our hero did not spot any weak points.
Time to break out a more powerful sword (`pdb`) which also did not divulge any secrets about what was wrong.
Our hero went back to the `strace` output and after a fair bit of thought and analysis a pattern started to emerge. Mypy was looking in the wrong place (i.e. not in in the environment created by `python.withPackages` but in the interpreter store path) and our princess was in another castle!
Our hero went to the pub full of old grumpy men giving out the inner workings of the open source universe (Github) and acquired a copy of Mypy.
He littered the code with print statements & break points.
After a fierce battle full of blood, sweat & tears he ended up in 20f7f2dd71/mypy/sitepkgs.py and realised that everything came down to the Python `site` module and more specifically https://docs.python.org/3.7/library/site.html#site.getsitepackages which in turn relies on https://docs.python.org/3.7/library/site.html#site.PREFIXES .
Our hero created a copy of the environment created by `python.withPackages` and manually modified it to confirm his findings, and it turned out it was indeed the case.
Our hero had damaged the dragon and it was time for a celebration.
He went out and acquired some mead which he ingested while he typed up his story and waited for the dragon to finally die (the commit caused a mass-rebuild, I had to wait for my repro).
In the end all was good in [NixOps-AWS](https://github.com/nixos/nixops-aws)-town and type checks could run. (PR for that incoming tomorrow).
