gh-121249: unconditionally support complex types in struct

[skirpichev]

Each complex type interpreted as an array type containing exactly two elements of the corresponding real type (real and imaginary parts, respectively).

• Issue: Add support for C99 complex type (_Complex) to the struct module #121249

---

📚 Documentation preview 📚: https://cpython-previews--132864.org.readthedocs.build/

[picnixz]

I forgot but what was the decision about ctypes?

[picnixz]

Buildbots didn't get launched apparently but that's good as we're modifying the code. Let's wait until everytihng is stable and we'll launch the buildbots again

[skirpichev]

I forgot but what was the decision about ctypes?

@picnixz, are you about this suggestion: #121249 (comment)?

Patch is here: #121249 (comment). I think it's much less urgent, as for the ctypes module - support can't be available unconditionally.

PS: Not sure why bots aren't started. Perhaps, this label should be added by core dev.

[picnixz]

I think it's much less urgent, as for the ctypes module - support can't be available unconditionally.

Ok! I actually wondered about the fate of ctypes because I wondered whether we still needed the _complex.h header.

PS: Not sure why bots aren't started. Perhaps, this label should be added by core dev.

I don't remember so? but maybe it is.. anyway, I'll add the label once you've confirmed whether the previous blurb needs an update.

[skirpichev]

When all this work settles and everything gets merged, I can try to prove my claim with a patch.

Feel free to do this even right now. But you should be able to prove that my concern is invalid.

[dalcinl]

But you should be able to prove that my concern is invalid.

The thing is, I'm still not sure what's your concern, I don't understand what exactly you meant by "annex G support still required from system". I actually agree with such claim, as long a "system" mean the C compiler used to build FFI, and the C compiler used to build a shared lib that ctypes call into, but NOT the C compiler used to build CPython and ctypes.
For example, I claim that the FFI library and some third-party library that accepts/returns complex numbers can be built with a modern GCC version, thus both sides have annex G support. But then you can build CPython/ctypes with some experimental in-development compiler that does not implement annex G support just yet, and things should still work fine, ctypes is passing complex values back and forth via the FFI library, and not the C compiler used to build CPython/ctypes. If this claim is not correct, then for sure I have a misunderstanding of how the FFI library implementation works, and in such case, I apologize for all the noise.

[skirpichev]

I don't understand what exactly you meant by "annex G support still required from system".

I tried to explain it twice: #121249 (comment) and #121249 (comment). Sorry, it's my best.

ctypes is passing complex values back and forth via the FFI library

The only way to pass it on ctypes side will be a two-element array then. I'm not sure if this will work, as it's not how complexes passed to a function.

[picnixz]

@hugovk Was it correct to amend the Misc/NEWS.d/3.14.0a1.rst for the main branch as the feature that we added there was actually changed? (that way, the online docs will contain only the most recent information and not duplicated entries). The entry will still exist in the tagged release though and will correctly reflect what 3.14.0a1 does.

EDIT: no it doesn't work that way :')

[picnixz]

@skirpichev I'm so sorry... but I didn't see that the entire changelog contains all previous alphas as well.. so, if we were to remove the NEWS we'll have a missing entry in the changelog page... So just restore the 3.14.0a1.rst file to what it has been but keep the new one for the beta section. You can say that previously it was conditioned though.

[skirpichev]

CC @vstinner

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @skirpichev for commit 38fa2a7 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F132864%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[skirpichev]

I see no failures, related to the pr.