➜
This issue tracker has been migrated to GitHub,
and is currently read-only.
For more information,
see the GitHub FAQs in the Python's Developer Guide.
Created on 2016-09-07 15:21 by sarterm, last changed 2022-04-11 14:58 by admin. This issue is now closed.
| Files | ||||
|---|---|---|---|---|
| File name | Uploaded | Description | Edit | |
| Python-3.5.1-compat.patch | sarterm, 2016-09-07 15:21 | Patch to allow using _LINUX_SOURCE_COMPAT when building on AIX | ||
| Messages (7) | |||
|---|---|---|---|
| msg274819 - (view) | Author: Matthieu S (sarterm) * | Date: 2016-09-07 15:21 | |
Python build fails on AIX when using the _LINUX_SOURCE_COMPAT flag. When using this flag, the gethostbyname_r subroutine has 6 parameters instead of 3, like on Linux, but Modules/socketmodule.c always expects a gethostbyname_r subroutine with only 3 parameters on AIX. The attached patch allows to build with this flag. It has been tested for Python 3.5.1 and 3.5.2 with GCC and XLC compilers. |
|||
| msg274971 - (view) | Author: Martin Panter (martin.panter) *  |
Date: 2016-09-08 05:58 | |
Would it be appropriate to always build with _LINUX_SOURCE_COMPAT enabled or disabled? Or is there a disadvantage (e.g. compatibility with different AIX versions, or less functionality?) Do you know if this would also be applicable to 2.7? |
|||
| msg274972 - (view) | Author: Martin Panter (martin.panter) * |
Date: 2016-09-08 06:06 | |
Also, there is code in configure.ac to detect 3- and 6-arg versions. Perhaps can we use that in all cases for AIX (as for Linux)? The configure logic was overridden a long time ago: revision dad14ffe1c22. But I don’t know exactly why. |
|||
| msg280295 - (view) | Author: Matthieu S (sarterm) * | Date: 2016-11-08 12:08 | |
Sorry for the very late reply... Using _LINUX_SOURCE_COMPAT is sometimes preferable, as it greatly improves compatibility with code intended for Linux and improves XPG specs compliance. But I am not sure it should always be enabled. In our case, we use it as there where issues with AIX malloc and hotshot when building without it, but Python seems to globally work fine without, and we did not compare performances. Like you, I don't know why this was overridden, so it might be safer to keep the override and just add the #ifdef _LINUX_SOURCE_COMPAT. This also applies to Python 2.7. |
|||
| msg280732 - (view) | Author: Roundup Robot (python-dev)  |
Date: 2016-11-14 05:39 | |
New changeset 7598e65a924e by Martin Panter in branch '3.5': Issue #28000: Fix gethostbyname_r() usage on AIX with _LINUX_SOURCE_COMPAT https://hg.python.org/cpython/rev/7598e65a924e New changeset 77cb8fa47138 by Martin Panter in branch '2.7': Issue #28000: Fix gethostbyname_r() usage on AIX with _LINUX_SOURCE_COMPAT https://hg.python.org/cpython/rev/77cb8fa47138 |
|||
| msg280735 - (view) | Author: Martin Panter (martin.panter) * |
Date: 2016-11-14 07:12 | |
I committed your patches for 3.5+ and 2.7. |
|||
| msg280927 - (view) | Author: Matthieu S (sarterm) * | Date: 2016-11-16 08:20 | |
Thanks ! |
|||
| History | |||
|---|---|---|---|
| Date | User | Action | Args |
| 2022-04-11 14:58:35 | admin | set | github: 72187 |
| 2016-11-16 08:22:36 | sarterm | set | versions: + Python 3.7 |
| 2016-11-16 08:21:15 | sarterm | set | status: open -> closed resolution: fixed |
| 2016-11-16 08:20:27 | sarterm | set | status: closed -> open resolution: fixed -> (no value) messages: + msg280927 versions: - Python 3.7 |
| 2016-11-14 07:12:22 | martin.panter | set | status: open -> closed versions: + Python 3.7 messages: + msg280735 resolution: fixed stage: patch review -> resolved |
| 2016-11-14 05:39:13 | python-dev | set | nosy:
+ python-dev messages: + msg280732 |
| 2016-11-08 12:08:14 | sarterm | set | messages: + msg280295 |
| 2016-09-08 06:06:18 | martin.panter | set | messages:
+ msg274972 versions: + Python 2.7 |
| 2016-09-08 05:58:51 | martin.panter | set | versions:
+ Python 3.6 nosy: + martin.panter messages: + msg274971 stage: patch review |
| 2016-09-07 16:00:25 | ericvw | set | nosy:
+ ericvw |
| 2016-09-07 15:21:43 | sarterm | create | |