- add " -std=gnu17" to `env.NIX_CFLAGS_COMPILE`
Fixes build failure with gcc15:
```
In file included from Include/Python-ast.h:3,
from Python/bltinmodule.c:4:
Include/asdl.h:9:15: error: cannot use keyword 'false' as enumeration constant
9 | typedef enum {false, true} bool;
| ^~~~~
Include/asdl.h:9:15: note: 'false' is a keyword with '-std=c23' onwards
Include/asdl.h:9:28: error: expected ';', identifier or '(' before 'bool'
9 | typedef enum {false, true} bool;
| ^~~~
Include/asdl.h:9:28: warning: useless type name in empty declaration
In file included from Include/Python-ast.h:3,
from Python/compile.c:26:
Include/asdl.h:9:15: error: cannot use keyword 'false' as enumeration constant
9 | typedef enum {false, true} bool;
| ^~~~~
Include/asdl.h:9:15: note: 'false' is a keyword with '-std=c23' onwards
Include/asdl.h:9:28: error: expected ';', identifier or '(' before 'bool'
9 | typedef enum {false, true} bool;
| ^~~~
```
As documented in https://packaging.python.org/en/latest/specifications/platform-compatibility-tags/#abi-tag.
This is useful to check whether a wheel is compatible with a certain interpreter. [Pyproject.nix](https://github.com/pyproject-nix/pyproject.nix) has [functions to perform wheel compatibility checking](https://pyproject-nix.github.io/pyproject.nix/lib/pypa.html#function-library-lib.pypa.isWheelFileCompatible) against a Python interpreter, and has computed interpreter ABI tags itself.
The recent addition of free threading (`python313FreeThreading`) complicates this by not being introspectable:
A GIL Python (non free-threaded) has an ABI tag `cp313` while the free-threaded Python has `cp313t`, but the package doesn't communicate whether `enableGIL` is true or false, leaving no way to compute the tag.
The same goes for if debugging support was added to the derivation: A `d` suffix would need to be added.
Additionally ABI tags has no defined format and can really only be accurately computed by having insight into how the ABI tags are used by a specific interpreter, meaning that the only correct place to compute ABI tags is within the context of a particular Python derivation.
While this has no immediate use within nixpkgs it could be used as a basis to provide compatibility assertions regarding wheel compat at eval time.
The intermediate pythonMinimal build that uses the glibc from bootstrap
files cannot currently be built with PIE, because the Scrt.o startup
script is missing in there.
The 3.13 release dropped support for the crypt module, so this dependency
could have been dropped much earlier.
Drops the `withLibxcrypt` argument, because manging the default has
become quite a bit more complicated.
Removes the build time dependency on pkg-config if python3 is build with the `withMinimal` flag enabled
This change is driven by the motivation to use python3 earlier in stdenv for hooks.
Make python3 minimal truly minimal by disabling some more dependencies.
This makes it easy to bootstrap python.
After this, the only remaining runtime deps for python3Minimal are:
- bash (for launching subprocesses with shell=True)
- libc
- libgcc
- libffi
All remaining deps seem to be fundamental and cannot be further reduced, except bash which could potentially be removed and replaced with `$SHELL` at runtime.
Done:
- add some more withXXX switches to the cpython package
- use new switches in python3Minimal to disable some deps
- set some other deps to null in python3Minimal
- Set `allowedReferences` to guarantee that the closure remains minimal in the future.
They are not doing anything right now. This is in preparation for their
complete removal from the tree.
Note: several changes that affect the derivation inputs (e.g. removal of
references to stub paths in build instructions) were left out. They will
be cleaned up the next iteration and will require special care.
Note: this PR is a result of a mix of ugly regex (not AST) based
automation and some manual labor. For reference, the regex automation
part was hacked in: https://github.com/booxter/nix-clean-apple_sdk
Signed-off-by: Ihar Hrachyshka <ihar.hrachyshka@gmail.com>
This reverts commit 65a333600d.
This wasn't tested for correctness with something like fodwatch [0],
and should not have been (self-)merged so quickly, especially without
further review.
It also resulted in the breakage of at least one package [1] (and that's
the one we know of and was caught).
A few packages that were updated in between this commit and this revert
were not reverted back to using `rev`, but other than that, this is a
1:1 revert.
[0]: https://codeberg.org/raphaelr/fodwatch
[1]: https://github.com/NixOS/nixpkgs/pull/396904 / 758551e458
An attempt to address #338231. The current release of nixpkgs takes over
20GB of memory to evaluate python when the interpreter is overriden.
This adds memoization to the spliced python packages to avoid
evaluating all the commbinations of overrides.
ghpzin