Consider dropping bespoke pep517 backend

[rossburton]

Is your feature request related to a problem?

No response

Describe the solution you'd like

As per https://cython.readthedocs.io/en/latest/src/userguide/source_files_and_compilation.html#basic-setup-py, it's now possible to build cython extensions in setuptools. Would you consider dropping the bespoke backend?

Describe alternatives you've considered

.

Additional context

• PEP517 backend is non-deterministic frozenlist#577
• [ENH] normalize build path cython/cython#5949

Code of Conduct

• I agree to follow the aio-libs Code of Conduct

[webknjaz]

Partially, which is why I encouraged upstream to implement that originally.
It doesn't cover every aspect of the wrapper, however. I'd want to see which parts can be migrated to be declarative first.

Keep in mind that a copy of this backend exists in multiple repositories and any changes would have to happen in all the synchronized repos. They must not diverge too much (short of project-specific bits like dist names et al.)
It should remain possible to easily sync changes with something like nvim -d, comparing diff per-file.

Note that since your motivation here is unclear, I can't guide you more at this point.

[rossburton]

Where is the canonical location for the backend? I've a number of issues to file.

My primary motivation is reproducible builds of yarl/frozenlist/proplist which are all not reproducible (see cython/cython#5949).

My final .so has host paths embedded, despite the compiler being told to remap build paths, because temporary directories outside the expected build tree are used.

[webknjaz]

It's these repos you named (s/proplist/propcache/ tho) + ansible/pylibssh. There's no dedicated repo just with the backend. I just happened to be working on projects with similar Cython infras and so this is duplicated. I never got to unifying it better. If you can reduce the about of hacks and submit them in sync into all these places, I'd be in favor. Although, I'd recommend sending small patches as we don't know how this would go.
I've also been wanting to transfer this pattern into aiohttp but I'm yet to establish how to best handle the vendored library with this setup. Making the wrapper thiner would be useful for this too.

[webknjaz]

xref aio-libs/frozenlist#577

[webknjaz]

FTR, in aio-libs/frozenlist#577 (comment) I mentioned wanting to have another config setting similar to https://github.com/aio-libs/frozenlist/blob/8c15ec9/packaging/pep517_backend/_backend.py#L76-L80 for letting the caller override auto-computed in-tree build toggle. Implementing that by example, mimicing other toggles, would be the easiest path forward right now, I think.

[webknjaz]

@rossburton you might want to keep an eye on #1590.

[webknjaz]

With #1591, can we consider this solved?

[rossburton]

I'm a fan of less build backends in general as they're mostly duplicates, but if you've considered using setuptools to build cython and don't want right now then sure, close the bug. I'll file new bugs for any further issues. I'll test that normalisation patch later today, hopefully it solves the problem for us.