WIP - port to Cython 3, and GAP 4.12

[dimpase]

an attempt to start port to Cython 3 and GAP 4.12 (4.13 hopefully then).

It builds, and import gap works.

Also, in needs work to add stream output, which got broken due to changed functionality in GAP.

[dimpase]

porting to Pyhton 3.12 has to be done. Easiest would be to use the recent code by @tornaria in sage/fpylll/cypari2.

[embray]

Did you see #23 ? It made some fixes that might be relevant here too.

[dimpase]

yeah, some of #23 repeat what I did, some of it makes no sense.

The output thing needs to use the new GAP kernel function, CALL_WITH_STREAM added in 4.12, see gap-system/gap#2839

It's on my todo list.

[embray]

Will definitely have a look soon. Just need to get GAP itself set up again

[dimpase]

by the way, in Sage we are getting bitten by the absence of a stateless way to convert Python objects to GAP, to call GAP functions.
(CallWith3Args etc).
see links in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114872 (was resolved invalid, and for a good reason).

Basically one must declare most if not all the Obj created in Sage volatile, otherwise in Pyhton 3.12
one gets segfaults (I guess in earlier Pythons one gets memory leaks)
from longmps in GAP, due to gcc putting Obj data in registers.

[dimpase]

gap-system/gap#5714

[dimpase]

Hi @saraedum - could you explain what's wrong in

Run conda install -q -c conda-forge gap-defaults=4.13.0
Channels:
 - conda-forge
 - defaults
Platform: linux-64
Collecting package metadata (repodata.json): ...working... done
Solving environment: ...working... failed

PackagesNotFoundError: The following packages are not available from current channels:

  - gap-defaults=4.13.0*

Current channels:

  - https://conda.anaconda.org/conda-forge
  - defaults

As far as I can see, conda-forge does have GAP-4.13.0.

[saraedum]

Note: 'use_only_tar_bz2' is enabled. This might be omitting some
packages from the index. Set this option to 'false' and retry.

These packages are not in the .tar.bz2 format.

[saraedum]

Why is that flag in the workflow file?

[dimpase]

This is the question for whoever put it there

[dimpase]

I'll see what happens if it's removed

[embray]

According to 3 years ago me "It's apparently required for caching to work properly" b2df390

Which makes a certain sense. But I don't know if we really need it.

[dimpase]

now the tests are running (some failing, as expected, due to incomplete porting)

[embray]

Would it still make sense to continue supporting GAP 4.10.x via preprocessor macros? Or has the community mostly moved beyond it?

[dimpase]

Would it still make sense to continue supporting GAP 4.10.x via preprocessor macros? Or has the community mostly moved beyond it?

I don't think any non-obsolete distro has GAP 4.10. See https://repology.org/project/gap/versions
So it can be dropped without any problem for anyone.

[embray]

Ok, just checking. I'm inclined to agree it's not worth the effort at this point, especially if we want to get a working release out.

[dimpase]

I propose to stick to gap 4.12 and 4.13, actually, due to incompatible API changes in libgap, compared to 4.11.

[FayeAlephNil]

@dimpase or @embray do yall have a broad idea of what changes still need to be made (/ areas to focus on) to make this pass tests? Perhaps it's porting everything to GAP 4.12/4.13 APIs.

I was planning on using gappy for a personal research project of mine, so I'd be willing to put in some work on this. But I don't know much about the status :)

[dimpase]

the project isn't active now. I am busy, and @embray even moreso.
I am thinking of trying to use nanobuild instead of cython, starting from scratch

[embray]

Still do really want to get back to this--the more time passes the more it gets harder to keep up. The first prototype of gappy was working quite well actually, and I think GAP has had some API improvements that will make some parts cleaner. One day I hope to have time 🙃

I might just take it up again for a few hours next time I'm feeling listless.

[dimpase]

maybe one should make a radical pivot from Cython to pybind11 or nanobind here.
Less boilerplate.