gh-123681: handle c99-specific specifiers for strftime when cross-compiling

[blhsing]

• Issue: test_strftime_y2k fails on embedded Linux #123681

[blhsing]

Hmm how do I trigger the JIT checks?

[picnixz]

Hmm how do I trigger the JIT checks?

Change a line in any file matching Python/optimizer*.c (ok that's an ugly hack but I can't run it manually for you I think)

[picnixz]

@blhsing Ok, I think I know how to run the JIT manually:

• Go to your own fork.
• Open the Actions tab.
• Expand the workflows list and locate the JIT one.
• Manually trigger it on the branch you want.

[blhsing]

@blhsing Ok, I think I know how to run the JIT manually:

• Go to your own fork.
• Open the Actions tab.
• Expand the workflows list and locate the JIT one.
• Manually trigger it on the branch you want.

Ah I see. Thanks a lot for the tips. Will definitely do it that way next time!

With the optimizer.c hack though the commit did pass the JIT checks so I'm reverting the hack now.

@serhiy-storchaka would you help review and possibly merge the PR? Thanks!

[zware]

I'm not sure we can make just this change. AC_RUN_IFELSE takes input, action-if-true, action-if-false, and action-if-cross-compiling; this is changing action-if-cross-compiling to unconditionally set ac_cv_strftime_c99_support=yes. We can only do that if we can unconditionally set that in any case, so we might as well remove the whole check. I think we should be able to do that; IIUC we currently require C11.

[zware]

Checking ./configure logs for recent buildbot builds, I see no indication that any platform ever gets a false result from the configure check (other than in cross-compiles), so I really see no reason to keep it.

[blhsing]

We can only do that if we can unconditionally set that in any case, so we might as well remove the whole check. I think we should be able to do that; IIUC we currently require C11.

Very true. There's no need to check for C99 support since CPython now requires C11. Will remove the check entirely then.

[erlend-aasland]

Very true. There's no need to check for C99 support since CPython now requires C11. Will remove the check entirely then.

For inspiration, take a look at how we detect libm C99 features around L6000. FTR, there's a few other places configure.ac where we can tear out C99 stuff.

[zanieb]

@erlend-aasland Hey! I'm following up on this because I ran into the problem in #128104. When you say

For inspiration, take a look at how we detect libm C99 features around L6000.

Are you suggesting we continue to test if strftime behaves in a C99-compliant manner or just drop that check entirely? I don't quite follow how we could do the same thing was is done for libm.

edit: Followed up with a patch in #128106 — happy to chat there

[picnixz]

Considering this PR has been stale for two months and that the other PR decides on a cleaner approach IMO, I'd suggest closing this one (I've also closed the corresponding issue I've opened since it's no more up-to-date).

[erlend-aasland]

I agree with Bénédikt. Closing.