| msg339869 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-10 15:04 |
| This is something that probably shouts - boring - but back in 2012 it was a hot topic for linux2 and linux3. Maybe - as far back as 1996 (when AIX4 was new) "aix3" and "aix4" made sense. Whether that is true, or not - is pointless these days - for Python3. In the python code I have reviewed - for various reasons - I have never seen any code with "sys.platform() == "aixX" (where X is any of 5, 6, 7). There was a reference to "aix3" and "aix4" in setup.py (recently removed, and there is no replacement for aix5, aix6, or aix7 - not needed!) What I mostly see is sys.platform.startswith("aix"). The other form of the same test is sys.platform[:3] == 'aix' sys.platform is "build" related, e.g., potentially bound to libc issues. Even if this was the case (AIX offers since 2007 - official binary compatibility from old to new (when libc is dynamically linked, not static linked), was "unofficial" for aix3 and aix4). Yes, I am sure there are arguments along the line of "why change" since we have been updating it - always, and/or the documentation says so. linux2 had to stay, because there was known code that compared with linux2 (and that code was having problems when python was built on linux3 - hence the change to make sys.platform return linux2 for all Python3.2 and younger). FYI: in Cpython (master) there are no references to "aixX". All the references there are (in .py) are: michael@x071:[/data/prj/python/git/cpython-master]find . -name \*.py | xargs egrep "[\"']aix" ./Lib/asyncio/unix_events.py: if is_socket or (is_fifo and not sys.platform.startswith("aix")): ./Lib/ctypes/__init__.py: if _sys.platform.startswith("aix"): ./Lib/ctypes/util.py:elif sys.platform.startswith("aix"): ./Lib/ctypes/util.py: elif sys.platform.startswith("aix"): ./Lib/distutils/command/build_ext.py: elif sys.platform[:3] == 'aix': ./Lib/distutils/util.py: elif osname[:3] == "aix": ./Lib/sysconfig.py: elif osname[:3] == "aix": ./Lib/test/test_asyncio/test_events.py: if sys.platform.startswith("aix"): ./Lib/test/test_faulthandler.py: @unittest.skipIf(sys.platform.startswith('aix'), ./Lib/test/test_strftime.py: or sys.platform.startswith(("aix", "sunos", "solaris"))): ./Lib/test/test_strptime.py: @unittest.skipIf(sys.platform.startswith('aix'), ./Lib/test/test_locale.py: @unittest.skipIf(sys.platform.startswith('aix'), ./Lib/test/test_locale.py: @unittest.skipIf(sys.platform.startswith('aix'), ./Lib/test/test_fileio.py: not sys.platform.startswith(('sunos', 'aix')): ./Lib/test/test_tools/test_i18n.py: @unittest.skipIf(sys.platform.startswith('aix'), ./Lib/test/test_wait4.py: if sys.platform.startswith('aix'): ./Lib/test/test_c_locale_coercion.py:elif sys.platform.startswith("aix"): ./Lib/test/test_shutil.py:AIX = sys.platform[:3] == 'aix' ./Lib/test/test_utf8_mode.py: elif sys.platform.startswith("aix"): I'll write the patch - if I recall it should be a one-liner in configure.ac, but I think some discussion (or blessing) first is appropriate. Maybe even review whether other platforms no longer rely on the X for the platform. Hoping this helps! |
|
|
| msg339870 - (view) |
Author: STINNER Victor (vstinner) *  |
Date: 2019-04-10 15:05 |
| I like the idea. Would you like to propose a patch? I suggest to only make such change in Python 3.8 and properly document it. |
|
|
| msg339872 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-10 15:54 |
| On 10/04/2019 17:05, STINNER Victor wrote: > STINNER Victor <vstinner@redhat.com> added the comment: > > I like the idea. Would you like to propose a patch? I suggest to only make such change in Python 3.8 and properly document it. > > ---------- > nosy: +vstinner > > _______________________________________ > Python tracker <report@bugs.python.org> > <https://bugs.python.org/issue36588> > _______________________________________ > If I understand correctly, the change should be quite simple: diff --git a/configure.ac b/configure.ac index 73ee71c..9632add 100644 --- a/configure.ac +++ b/configure.ac @@ -404,6 +404,7 @@ then MACHDEP="$ac_md_system$ac_md_release" case $MACHDEP in + aix*) MACHDEP="aix";; linux*) MACHDEP="linux";; cygwin*) MACHDEP="cygwin";; darwin*) MACHDEP="darwin";; However, I am less familiar with (where) the appropriate documentation is. A pointer to the documentation is appreciated. |
|
|
| msg339895 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-04-10 22:09 |
| > However, I am less familiar with (where) the appropriate documentation is. A pointer to the documentation is appreciated. It would be nice to add a note to: * https://docs.python.org/dev/whatsnew/3.8.html#changes-in-the-python-api * https://docs.python.org/dev/library/sys.html#sys.platform In the source code, there are the files: * Doc/whatsnew/3.8.rst * Doc/library/sys.rst |
|
|
| msg339983 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-11 15:23 |
| Was: root@x064:[/data/prj/python/python3-3.8]./python Python 3.8.0a3+ (heads/bpo-28009-2-dirty:2fb2bc81c3, Apr 11 2019, 07:09:55) [C] on aix6 Type "help", "copyright", "credits" or "license" for more information. >>> import sys >>> sys.platform 'aix6' >>> import os >>> os.uname()[3] '7' ./python ../git/python3-3.8/Tools/scripts/patchcheck.py Getting base branch for PR ... origin/master Getting the list of files that have been added/changed ... 6 files Fixing Python file whitespace ... 0 files Fixing C file whitespace ... 0 files Fixing docs whitespace ... 0 files Docs modified ... yes Misc/ACKS updated ... NO Misc/NEWS.d updated with `blurb` ... yes configure regenerated ... yes pyconfig.h.in regenerated ... no On system I built on: root@x066:[/data/prj/python/python3-3.8]./python Python 3.8.0a3+ (heads/bpo-36588-aix-platform-dirty:2021d40faa, Apr 11 2019, 15:16:05) [C] on aix Type "help", "copyright", "credits" or "license" for more information. >>> import os,sys >>> sys.platform 'aix' >>> os.uname()[3] '6' And on AIX7 system: root@x064:[/data/prj/python/python3-3.8]./python Python 3.8.0a3+ (heads/bpo-36588-aix-platform-dirty:2021d40faa, Apr 11 2019, 15:16:05) [C] on aix Type "help", "copyright", "credits" or "license" for more information. >>> import os,sys >>> sys.platform 'aix' >>> os.uname()[3] '7' |
|
|
| msg340048 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-04-12 14:15 |
| New changeset 9d949f7796da612f1b588d18c6f041376992a9fc by Victor Stinner (Michael Felt) in branch 'master': bpo-36588: On AIX, remove major version from sys.platform (GH-12787) https://github.com/python/cpython/commit/9d949f7796da612f1b588d18c6f041376992a9fc |
|
|
| msg340049 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-04-12 14:16 |
| Do you want to work on a change to replace sys.platform.startswith("aix") to cleanup the stdlib and tests? Not sure if it's needed :-) It's up to you. |
|
|
| msg340061 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-12 15:28 |
| On 12/04/2019 16:16, STINNER Victor wrote: > STINNER Victor <vstinner@redhat.com> added the comment: > > Do you want to work on a change to replace sys.platform.startswith("aix") to cleanup the stdlib and tests? Not sure if it's needed :-) It's up to you. > > ---------- > > _______________________________________ > Python tracker <report@bugs.python.org> > <https://bugs.python.org/issue36588> > _______________________________________ > Sure. I'll do that. I shoul open an issue along the lines of "AIX: cleanup stdlib and tests and how sys.platform is utilized". But, should I just continue standard practice (sys.platform), or would this be a moment to move towards platform.system() (i.e., set the example to be to use "run-time" rather than "build-time"). Michael |
|
|
| msg340064 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-04-12 15:34 |
| > But, should I just continue standard practice (sys.platform), or would > this be a moment to move towards platform.system() (i.e., set the > example to be to use "run-time" rather than "build-time"). Oh, now I'm confused :-) I checked the Python test suite: some tests use sys.platform == "linux" or sys.platform in ("linux", ...), some tests uses sys.platform.startswith("linux"). In case of doubt, I suggest to do nothing :-) Leave the code unchanged :-) |
|
|
| msg340112 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-12 21:16 |
| On 12/04/2019 17:34, STINNER Victor wrote: > STINNER Victor <vstinner@redhat.com> added the comment: > >> But, should I just continue standard practice (sys.platform), or would >> this be a moment to move towards platform.system() (i.e., set the >> example to be to use "run-time" rather than "build-time"). > Oh, now I'm confused :-) I checked the Python test suite: some tests use sys.platform == "linux" or sys.platform in ("linux", ...), some tests uses sys.platform.startswith("linux"). > > In case of doubt, I suggest to do nothing :-) Leave the code unchanged :-) Agreed, in case of doubt - leave alone (never change a winning team). And, to make it a short reply - I'll get started, and we see where it leads us. Michael > > ---------- > > _______________________________________ > Python tracker <report@bugs.python.org> > <https://bugs.python.org/issue36588> > _______________________________________ > |
|
|
| msg340156 - (view) |
Author: Michael Felt (Michael.Felt) * |
Date: 2019-04-13 12:28 |
| On 12/04/2019 23:16, Michael Felt wrote: > Agreed, in case of doubt - leave alone (never change a winning team). > > And, to make it a short reply - I'll get started, and we see where it > leads us. I opened (https://bugs.python.org/issue36624) - before I "take off", some comments/discussion on what is desirable and achievable would be welcome. |
|
|