bpo-40212: Re-enable posix_fadvise and f_allocate on AIX

[isidentical]

https://bugs.python.org/issue40212

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @isidentical for commit 4881d1a 🤖

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

[taleinat]

Would it be possible to check for the version of AIX to only enable this on recent enough versions where this bug is indeed fixed?

[kadler]

@taleinat That would require a new configure check. An easy way to do so would be to add a try_compile and see if posix_fadvise and posix_fallocate are defined as macros to posix_fadvise64 and posix_fallocate64. Something like:

#include <fcntl.h>

#if defined(_LARGE_FILES) && !defined(__64BIT__) && !defined(posix_fadvise)
#error posix_fadvise contains the wrong definition on this AIX
#endif

void foo() {}

[tiran]

Would it be possible to check for the version of AIX to only enable this on recent enough versions where this bug is indeed fixed?

I wouldn't bother. The issue has been fixed in 2014 and the bug was finally closed in 2016.

@isidentical There is a conflict. Please rebase your PR and re-generate argument clinic file.

[isidentical]

I also would like to hear @aixtools 's opinions on this

[aixtools]

reading. attending a (remote) conference today. will be back soon.

[aixtools]

Quick feedback: need to look at which version it is fixed. Maybe the AIX versions where it is buggy are no longer being considered (e.g., AIX 5 and before).

[isidentical]

Maybe the AIX versions where it is buggy are no longer being considered (e.g., AIX 5 and before).

We already removed support for them, so that is one of the points (AFAIK the lowest we support right now is 6.1).

[tiran]

According to https://www.ibm.com/support/pages/apar/IV56170 the bug was fixed in AIX7.1.

[kadler]

Looks like there's a related APAR for 6.1: https://www.ibm.com/support/pages/apar/IV47048

Edit: Looks like it was marked "fixed in next release", though, so 6.1 would seem permanently broken.

[aixtools]

For AIX 7.1 I was able to locate the fix: AIX 7.1 TL4 (py36) aixtools@x064:[/home/aixtools/aixpm]instfix -ciqk IV56170 IV56170:bos.64bit:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA IV56170:bos.adt.include:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA IV56170:bos.rte.control:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA IV56170:bos.rte.libc:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA IV56170:bos.rte.shell:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA IV56170:mcr.rte:7.1.4.0:7.1.4.35:+:32-BIT CALL TO POSIX_FADVISE FAILS WHEN _LARGE_FILES IS ENA I was not able to find it in AIX 6.1 (AIX 7.1 TL3 parallels AIX 6.1 TL9 - AIX 6.1 TL10 was never released, so likely, it was never fixed, read released, in 6.1). IMHO: the special treatment for AIX may be removed. The bots run on AIX 7.1 TL4 or later.

…

On 17/11/2020 16:32, Kevin Adler wrote: Looks like there's a related APAR for 6.1: https://www.ibm.com/support/pages/apar/IV47048 <https://www.ibm.com/support/pages/apar/IV47048> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#19403 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACSZR5PH663IT4JDPCABOZTSQKJSXANCNFSM4MCXZGNQ>.

[isidentical]

I was not able to find it in AIX 6.1 (AIX 7.1 TL3 parallels AIX 6.1 TL9

• AIX 6.1 TL10 was never released, so likely, it was never fixed, read
  released, in 6.1).

I am not too interested in dropping AIX 6.1 support right away

[furkanonder]

@isidentical Could you resolve the conflicts?

[encukou]

Conflicts resolved in #144784.