As we continue iterating on Python infrastructure for Python 3, some
code is starting to diverge for Python 2. If we copy the current state
of mk-python-derivation.nix and freeze it for Python 2, we can iterate
on it for Python 3 with more freedom.
Upstream's recommended "python -m build" way of invoking build fails
when the working directory contains a file named "build.py". This is
common for poetry projects that build C extensions.
Since the wheel file name is based on the package metadata instead of
what we set in the derivation, if you set the version as `unstable-YYYYMMDD`,
for example, the hook would silently fail. We will ignore the version
now and just use a glob instead to match anything that has the package
name in its path.
Fixes#248185.
Python always uses "gnu" prefixed ABI names, and this patch handles this by
doing a musl->gnu string replacement.
With pkgsCross.armv7l-hf-multiplatform.pkgsStatic, the previous name was:
_sysconfigdata__linux_arm-linux-gnu
Now, the corrected name is:
_sysconfigdata__linux_arm-linux-gnueabihf
This reverts commit 2ffca30cde.
`793cc9d982415b71cdba729cf779bfc49e9d2ae7` `python3: splice python within python3Packages.callPackage`
Was reverted because it broke
`pkgsCross.aarch64-multiplatform.python3Packages.cryptography`
But by reverting the revert `pkgsCross.aarch64-multiplatform.python3Packages.cryptography` still builds.
I tried reverting the 2 cryptography update commits, and it still
worked, so I do not know why this would work now.
Deeply-curried functions are pretty error-prone in untyped languages
like Nix. This is a particularly bad case because
`top-level/splice.nix` *also* declares a makeScopeWithSplicing, but
it takes *two fewer arguments*.
Let's switch to attrset-passing form, to provide some minimal level
of sanity-checking.
`lib.extends hooks pythonPackagesFun` includes python which we want to
splice
what this fixes:
```
nix-repl> lib.elemAt pkgsCross.aarch64-multiplatform.python3Packages.xpybutil.nativeBuildInputs 0
«derivation /nix/store/39dkb51rciw6zwg0c2c44gpmpjapddxc-python3-aarch64-unknown-linux-gnu-3.10.12.drv»
```
to
```
nix-repl> lib.elemAt pkgsCross.aarch64-multiplatform.python3Packages.xpybutil.nativeBuildInputs 0
«derivation /nix/store/21ldw2dp26xvv9iyxn9x77a8yh4waqz5-python3-3.10.12.drv»
```
Before https://github.com/NixOS/nixpkgs/pull/194205 The keep python used to work by accident because self was passed from
__splicedPackages https://github.com/NixOS/nixpkgs/pull/196052/files#diff-44ce3495c4f983ce64dd47c86a9d3e77bad210b2709c098a3806998dcd9b000bR213
But now it does not work because overrideAttrs is used on python in
python-packages-base.nix
Preferably we would [ splice the hooks too ](https://github.com/NixOS/nixpkgs/pull/228139) but we cannot do that until [Package sets within derivations (i.e. python3.pkgs) are not spliced](https://github.com/NixOS/nixpkgs/issues/211340) is fixed, because people often use `python3.pkgs.wrapPython` in `nativeBuildInputs` (it's correct but python3.pkgs should be python3Packages to get splicing.
The setuptools-rust requires some environment variables to really
perform cross build, otherwise it just builds for build platform.
This adds setup hook that introduces these environment variables.
There are three variables.
The PYO3_CROSS_LIB_DIR has to point to the target's Python library
directory. This has to be directory for the target not for the build or
host. We have to choose the correct target Python. I am unsure how to do
that simply in nixpkgs and this this implementations just delays this
and waits for the correct Python when package using this hook is build.
The CARGO_BUILD_TARGET triggers cross compilation in setuptools-rust.
This is simply the Rust target specification.
The CARGO_TARGET_*_LINKER variable should not be essentially required
but setuptools-rust probably mangles the Rust build environment somewhat
and that results to the missing linker. By explicitly specifying it
using the environment variable we force the correct linker.
Since python documentation requires Sphinx to build, documentation is built in
separate derivation to avoid hard recursion/bootstrapping questions and plugged
into main derivation. Care is taken that "python3.doc" derivation feels and
looks as close as possible to as if it were main derivation were multi-output
one.
This change does not cause rebuild of python3 itself.
Add a marker file to the python outputs that tells pip and other tooling
following PEP 668 that they should not be installing things system-wide.
This provides better feedback to the user and also potentially avoids issues in
single-user installs where the /nix/store is owned by the user.
For more details, see <https://peps.python.org/pep-0668/>
Here is how it currently looks like:
$ pip install requests
error: externally-managed-environment
× This environment is externally managed
╰─> This command has been disabled as it tries to modify the immutable
`/nix/store` filesystem.
To use Python with Nix and nixpkgs, have a look at the online documentation:
<https://nixos.org/manual/nixpkgs/stable/#python>.
mkPythonDerivation: apply checks to buildInputs as well
fixup: add a suggestion about nativeBuildInputs
Clean up the wording, add reasons for the suggestions
switch to throw
Adjust error to clarify build-time vs. run-time
- Prune unused imports
- Collect imports at the top
- Removed unused exception assignments
- Fill bare except clause with
- Expand overly long check_output lines
Retrieve the current package hash and use it for a more precise match,
which allows us to update a derivation in situation where multiple
hashes are present.
By settings `passthru.skipBulkUpdate = true` in a python package, it
will be excluded from bulk update runs.
This is useful in case of fragile packages, that require special
attention during updates.
Previously when packages that required the git fetcher were updated, we
would wrongly rely on `nix-prefetch-url`, which would reliable break the
hash.
Instead we need to use `nix-prefetch-git` to determine the proper hash,
when the relevant attributes are present.
When PyPy introduced Python 3.8 support with version 7.3.6, they also
migrated to using CPython's directory layout [0]:
> The 3.8 package now uses the same layout as CPython, and many of the
PyPy-specific changes to `sysconfig`, `distutils.sysconfig`, and
`distutils.commands.install.py` have been removed. The stdlib now is
located in `<base>/lib/pypy3.8` on `posix` systems...
When we upgraded past this version and added Python 3.8 support [1], the
`sitePackages` value was never updated, leading `bootstrapped-pip` to fail
to build, because wheel was trying to be located in `$out/site-packages`,
when it was actually installed to `$out/lib/pypy3.8/site-packages`.
[0]: https://www.pypy.org/posts/2021/10/pypy-v736-release.html
[1]: eec28b8cfd
with structuredAttrs lists will be bash arrays which cannot be exported
which will be a issue with some patches and some wrappers like cc-wrapper
this makes it clearer that NIX_CFLAGS_COMPILE must be a string as lists
in env cause a eval failure
The virtualenv mode is trying to be smart and checks whether it can
write to $out, which for some reason fails on darwin.
> virtualenv: error: argument dest: the destination . is not write-able at /nix/store
But what does work is creating it in the pwd and moving it to $out,
because mv just does what it is told.
Allow python3.tests to eval on other platforms than linux, by excluding
conda tests when not on linux, because they include alsa-lib, which is
linux-only.
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
