Previously, unless unsupported platforms were allowed, the following
would fail to evaluate (from an "x86_64-linux" system):
pkgsCross.x86_64-freebsd.__splicedPackages.docutils.__spliced.buildHost
It shouldn't have, because the buildHost package ends up being for
Linux. This broke evaluation of e.g. pkgsCross.x86_64-freebsd.libdrm,
because it has docutils in nativeBuildInputs. mkDerivation would try
to go through __spliced.buildHost on docutils to get to the Linux
version, but the check in ensurePythonModules would kick in first,
triggering the meta check because of the equality check in the
implementation of hasPythonModule, which would fail because Python is
not marked as supported on FreeBSD in Nixpkgs at the moment. Thus,
even though they're not supposed to be, the meta checks would be
triggered even though the only attribute being accessed on the
unsupported derivation was __spliced.
We can fix this by using the same mechanism used to implement the meta
checks themselves: lib.extendDerivation. Now, attempting to access
drvPath or outPath on an attribute that fails the validity check will
produce the same error as before, but other accesses will be allowed
through, fixing splicing.
I've tested evaluation of packages that pass and fail the validity
check, and confirmed that the behaviour is still correct.
It wasn't clear before, but it turns out that this is just four
separate conditions, and an attribute is valid if any of the
conditions is true, so we don't need a sea of parentheses to express
it.
checkInputs used to be added to nativeBuildInputs. Now we have
nativeCheckInputs to do that instead. Doing this treewide change allows
to keep hashes identical to before the introduction of
nativeCheckInputs.
fetchPypi doesn't use python under the hood and doesn't need to be tied
to a specific version of python. Moving it to top level makes it more
consistent with other fetchers and makes code generation easier.
- Remove code related to i686 archs (not supported anymore since #209100)
- Split non-install steps in installPhase in installCheckPhase/preFixup
- Add missing pre/post hooks
Also build it with pythonInterpreters.pypy27_prebuilt instead of
python27, to avoid it being broken after #201859 is merged.
Keep in mind that the above change will make this derivation only work
with x86_64-linux for now (since pypy27_prebuilt is only packaged for
that platform). However support for newer platforms can be added later,
and a working package for one platform is better than a broken package
in 4 platforms.
The package set hasn't been working for a long time now, due to infinite
recursions, that nobody was going to fix.
The release is going to go EOL in 2023/06 and we don't want to ship it
in NixOS 23.05 anyway.
Include full $name of the derivation into docdir, so documentation for
python package "foo" is installed into $out/share/doc/python3.10-foo-1.2.3
instead of just $out/share/doc/foo, where it may conflict with some
other package named "foo" and not-coinstallable with different
versions of the same python package.
Change from $name to $pname was introduced in [1ee5fca], probably as
unindented side-effect of adding support for formats other than html.
Without `--with-system-{ffi,expat}` flags, Python will use its own
embedded libraries that are out-of-date. Thanks to it, they can be a
security issue. So let's use our own libraries instead.
This is already what Python 3.x does, so should be safe.
This PR strips down the modified `python27` derivation used by `resholve`. The
idea is to reduce the possible security issues, and also to make it easier to
bootstrap.
ActiveState is a company that is maintaining a fork of Python 2 to fixes
its security issues. Their support is paid, however the code is
open-source. See the details here:
https://www.activestate.com/products/python/python-2-end-of-life-security-updates/
This enable us to drop a bunch of CVE's patches for Python 2.7 and also
it should be easier to maintain, since we can just bump the version once
ActiveState tags a new version.
Prior to this commit, pythonRelaxDeps would only support removing
version constraints from "Requires-Dist" lines formatted in a particular
way ("foo (>= 1.2.3)"). This way is deprecated as per PyPA Core Metadata
Specs v2.1 [1]:
> Tools parsing the format should accept optional parentheses around
> this, but tools generating it should not use parentheses.
Additionally, a "Requires-Dist" dependency specification can contain
other metadata than just package name and version (extra names,
environment marker). These were being silently dropped by the prior
version of pythonRelaxDeps, or the version could not be relaxed.
The actual grammar is defined in PEP 508 [2]. Our tool of choice here is
sed extended regexps, so there's only so much we can do to be correct
with this parser. The regexp implemented in this commit makes an attempt
at supporting [extra] names, ; env_markers, as well as version specs
without parentheses. There are still unsupported features (URL specs) as
well as unhandled edge cases, but at some point trying to make the
regexp better is bound to awake ZALGO [3].
[1] https://packaging.python.org/en/latest/specifications/core-metadata/#requires-dist-multiple-use
[2] https://peps.python.org/pep-0508/#grammar
[3] https://stackoverflow.com/a/1732454/179806
stdenv is included in the package set. It does not provide a Python
module however and was therefore catched by the test. Because we do need
it we provide an exception for it.
This adds a test to ensure no new uses of `buildPythonApplication` can
be added to `python-packages.nix`.
Python packages can be grouped into two groups: 1) applications and 2)
packages providing importable modules. In `python-packages.nix` we only
want to have 2). 1) should be in the top-level package set.
To achieve this, all setup hooks need to be marked as being a setup hook.
For the setup hooks in the Python packages set this is done by creating
a new builder, `makePythonHook`.
Because there were issues with splicing, the file importing all the hooks
is converted to an extension. All non-packages were moved out of `python-packages.nix`
into `python-packages-base.nix`. The `keep` argument to `makeScopeWithSplicing
was cleaned up as well; there is no need to keep this one manually in sync
reducing the risk of breaking cross-compilation.
Before packages did not had to set `format = "other";` if had a custom
installPhase and would not produce a dist folder.
The current error message is hard to understand by new users:
> Executing pythonOutputDistPhase
> mv: cannot stat 'dist': No such file or directory
This commit improves the error message.
BEFORE:
the python derivation did not contain .nativeDrv and .crossDrv because
it was not from the __splicedPackages set
the python used in mk-python-derivation.nix was for host
```
nix-repl> pkgsCross.aarch64-multiplatform.python3Packages.xpybutil.nativeBuildInputs
[ «derivation /nix/store/bhz39ds4v02hn6x4py4mzjyilw4a589h-python3-aarch64-unknown-linux-gnu-3.10.7.drv» «derivation /nix/store/v880cnh4ml7czmivfbk3cdh93hz9yvbn-hook.drv» «derivation /nix/store/f243ab7wv92gqsmc9h7gr0qcnj0xcgdb-hook.drv» «derivation /nix/store/880lf8895bzn8d94lrr2y7ilgkxq0lc4-python-remove-tests-dir-hook.drv» «derivation /nix/store/fzjnhawfs1wpw58hcd1vxd9y750dc08y-python-remove-bin-bytecode-hook.drv» «derivation /nix/store/n8l59iparx98yfw8g5ydqmzmk3fdic75-setuptools-setup-hook.drv» «deri→
```
AFTER:
the python derivation does contain .nativeDrv and .crossDrv because
it is from the __splicedPackages set
those 2 are what makes nativeBuildInputs and buildInputs function
properly
the python used in mk-python-derivation.nix is for build
```
nix-repl> pkgsCross.aarch64-multiplatform.python3Packages.xpybutil.nativeBuildInputs
[ «derivation /nix/store/hvb9yxgv1133cfhxxd869sibldvv2vdx-python3-3.10.7.drv» «derivation /nix/store/v880cnh4ml7czmivfbk3cdh93hz9yvbn-hook.drv» «derivation /nix/store/f243ab7wv92gqsmc9h7gr0qcnj0xcgdb-hook.drv» «derivation /nix/store/880lf8895bzn8d94lrr2y7ilgkxq0lc4-python-remove-tests-dir-hook.drv» «derivation /nix/store/fzjnhawfs1wpw58hcd1vxd9y750dc08y-python-remove-bin-bytecode-hook.drv» «derivation /nix/store/n8l59iparx98yfw8g5ydqmzmk3fdic75-setuptools-setup-hook.drv» «derivation /nix/store/7vyhynla→
```
Since the Nix command used is an implementation detail of the script,
it's the script that will have to be updated if the command changes,
it's the script that should be opting into this, rather than making
the user do it globally.
Since c0972c16dc ("update-python-libraries: add missing dependency nix"),
Nix will be provided as a dependency, so we know we'll always be using
a modern version and no longer need the fallback.
*Flags implies a list
slightly relevant:
> stdenv: start deprecating non-list configureFlags https://github.com/NixOS/nixpkgs/pull/173172
the makeInstalledTests function in `nixos/tests/installed-tests/default.nix` isn't available outside of nixpkgs so
it's not a breaking change
By default buildPython* runs a hook for detecting conflicting packages.
This hook needs pkg_resources which is part of setuptools.
Before this commit, setuptools was simply added to the build. This meant
that when setuptools was forgotten to be added to the build, the build
and installation would still succeed because of this package from the
hook. During runtime (and cross-compilation) one would notice the
missing package.
Store the intermediate artifacts. In time, we should build, install and
test in separate derivations as that reduces circular dependencies,
avoids rebuilds when fixing tests, and makes it possible to use the
wheels for creating say virtualenv's.
Removes the up until now unused option to specify a `sphinxOutdir` in
favor of allowing to specify multiple builders, which is for example
useful for generating both documentation and manpages using the hook.
Since the output path cannot be determined from within the package we
automatically generate it and add a diversion for manpages, so they land
in the correct output and path.
Before the change separate-debug-info.sh did the stripping itself.
This scheme has a few problems:
1. Stripping happens only on ELF files. *.a and *.o files are skipped.
Derivations have to do it manually. Usually incorrectly
as they don't run $RANLIB (true for `glibc` and `musl`).
2. Stripping happens on all paths. Ideally only `stripDebugList` paths
should be considered.
3. Host strip is called on Target files.
This change offloads stripping logic to strip hook. This strips more
files for `glibc` and `musl`. Now we can remove most $STRIP calls
from individual derivations.
Co-authored-by: Sandro <sandro.jaeckel@gmail.com>
Python package sets can be overridden by overriding an interpreter
and passing in `packageOverrides = self: super: {...};`. This is fine
in case you need a single interpreter, however, it does not help you
when you want to override all sets.
With this change it is possible to override all sets at once by
appending a list of "extensions" to `pythonPackagesExtensions`.
From reading the implementation you might wonder why a list is used, and
not
`lib.composeExtensions`? The reason is the latter requires knowledge of
the library function. This approach should be easier for most users
as it is similar to how we append to lists of e.g. inputs or patches
when overriding a derivation.
The PEP600 standard gives Python's naming scheme for various
architectures; it follows the convention which was in use by Fedora in
2014. According to PEP600, the architecture name for Power PC is
`ppc64le`, not `powerpc64le`. This is also how python3 declares its
"supported wheels" under Debian on PowerPC, as checked with `pip debug
--verbose`
$ pip debug --verbose | grep powerpc
$ pip debug --verbose | grep ppc | head
cp39-cp39-manylinux_2_31_ppc64le
cp39-cp39-manylinux_2_30_ppc64le
cp39-cp39-manylinux_2_29_ppc64le
cp39-cp39-manylinux_2_28_ppc64le
cp39-cp39-manylinux_2_27_ppc64le
cp39-cp39-manylinux_2_26_ppc64le
cp39-cp39-manylinux_2_25_ppc64le
cp39-cp39-manylinux_2_24_ppc64le
cp39-cp39-manylinux_2_23_ppc64le
Let's adjust the `pythonHostPlatform` expression in
cpython/default.nix to pass the architecture using the naming scheme
Python expects.
Verified on a Raptor Computing Systems Talos II. Without this commit,
PyQt5 fails to build, failing with "unsupported wheel". With this
commit, it builds successfully.
Every package should have a maintainer and we should not throw it away if we arleady have one.
Also in reality the maintainer of python3 is also the one of python3-minimal
GCC does not come with a `libgcc_eh.a` for the target platform if
it was built without `--enable-shared`. That flag was removed with
c6dd11ca39, meaning we should no longer
attempt to link against that lib.
We have a common pattern here in nixpkgs for Python applications: when a
Python package ships with either a requirements.txt or setup.py file, we
generally end up having to modify its version restriction, otherwise we have
build failures since we package only one specific version of each package
normally.
However, this end up being done in a completely ad-hoc way: some people
use substituteInPlace, some others use sed, others uses patches, etc.
In many cases, the code ends up being buggy, so it may work in one
version and breaks on the next one. We can instead implement one
standard way of doing this, and trying to be a correct as possible.
So this is what this commit does: it implements a new build hook, that
when called will automatically patch the wheel file. This is one of the
most generic ways to patch Python dependencies, and should work in
multiple cases.
Noticed option globbing when tried to enable parallelism by default
locally for most packages by default. python3Packages.yt-dlp failed as:
python3.9-yt-dlp> /nix/store/5mywvxdjkk1q6srwwwgdkzc37ibla801-python3.9-setuptools-61.2.0/lib/python3.9/site-packages/setuptools/dist.py:516: UserWarning: Normalizing '2022.04.08' to '2022.4.8'
python3.9-yt-dlp> warnings.warn(tmpl.format(**locals()))
python3.9-yt-dlp> invalid command name 'build_lazy_extractors--parallel'
The change adds leading whitespace everywhere where options might
already be present.
Overriding the interpreters did not work correctly. When overriding
packages would end up twice in the build time closure: one corresponding
to the overridden interpreter and one corresponding to the original
interpreter. The reason is that the override was not applied to the
interpreters in the spliced package sets.
This simplifies usages and makes the default value consistent.
In a few cases, the default value was interpreted to be `false`,
but this is useless, because virtually nobody will explicitly
set `allowAliases = true;`.
It seems the additional linker flags were added in 9d3b0a2 (May 2008) as
a workaround for readline support. They were kept since then, but it
also means that `python3-config` outputs these flags
```console
$ python3-config --libs
-lpthread -ldl -lcrypt -lncurses -lutil -lm -lm
```
while other GNU/Linux distros do not. For example, Debian 11:
```console
$ sudo apt install python3-dev
$ python3-config --libs
-lcrypt -lpthread -ldl -lutil -lm -lm
```
This change removes the `-lncurses` flag and aligns Nix/NixOS with other
distros.
