backport of the _sysconfigdata.py module (issue 13150) breaks the build on darwin

[doko42]

BPO	17512
Nosy	@Yhg1s, @doko42, @ronaldoussoren, @pitrou, @benjaminp, @ned-deily
Files	workaround.diff backport.diff

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/doko42'
closed_at = <Date 2016-09-12.02:50:49.815>
created_at = <Date 2013-03-21.22:56:23.527>
labels = ['build']
title = 'backport of the _sysconfigdata.py module (issue 13150) breaks the build on darwin'
updated_at = <Date 2016-09-12.02:50:49.814>
user = 'https://github.com/doko42'

bugs.python.org fields:

activity = <Date 2016-09-12.02:50:49.814>
actor = 'berker.peksag'
assignee = 'doko'
closed = True
closed_date = <Date 2016-09-12.02:50:49.815>
closer = 'berker.peksag'
components = ['Build']
creation = <Date 2013-03-21.22:56:23.527>
creator = 'doko'
dependencies = []
files = ['29537', '29543']
hgrepos = []
issue_num = 17512
keywords = ['patch']
message_count = 10.0
messages = ['184919', '184923', '184928', '184944', '184945', '184946', '184970', '185063', '185072', '186331']
nosy_count = 7.0
nosy_names = ['twouters', 'doko', 'ronaldoussoren', 'pitrou', 'benjamin.peterson', 'ned.deily', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue17512'
versions = ['Python 2.7']

[doko42]

The backport of bpo-13150 to the 2.7 branch breaks the build on darwin. There is a special case for providing the _sysconfigdata module in _generate_posix_vars before get_platform is called (darwin only) and before the module is actually written.

However get_platform is called before, in site.addbuilddir(). Wondering if this call can either be avoided, or worked around.

options:

• backport bpo-586680 too (http://hg.python.org/cpython/rev/4742e7aea2f5/)
• fall back to generic values for osname, release, machine

any other options?

[doko42]

untested work-around. we don't need the builddir to generate the _sysconfigdata.py.

[benjaminp]

Backporting 4742e7aea2f5 would seem to be the simplest and most "correct" solution.

[doko42]

here's the backport, tested on Linux only. sorry, can't test this on MacOSX before Sunday.

[ronaldoussoren]

Reverting the backport would also fix the breakage.

Is the addition of _sysconfigdata not a new feature? It changes the place where sysconfig (and distutils sysconfig?) looks for configuration data, and hence breaks all instructions that mention that you can change the Makefile in {sys.prefix}/lib/python2.7/config to globally affect the compilation flags (although there hopefully aren't a lot of those)

[ronaldoussoren]

Guess what, I hadn't read the thread on this python-dev yet :-(

[benjaminp]

I backed it out until it can be done without breaking OSX. d174cb3f5b9e

[benjaminp]

Can landed corrected in 2.7.5.

[doko42]

tested the backport on MacOSX, fixing the build.

[python-dev]

New changeset be3b4aa2ad28 by doko in branch '2.7':

• Issue bpo-13150, bpo-17512: sysconfig no longer parses the Makefile and config.h
  http://hg.python.org/cpython/rev/be3b4aa2ad28