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.

classification
Title: Append system paths in setup.py instead of prepending
Type: behavior Stage: resolved
Components: Build Versions: Python 2.7
process
Status: closed Resolution: fixed
Dependencies: Superseder:
Assigned To: Nosy List: christopher.hogan, iritkatriel, r.david.murray, zach.ware
Priority: normal Keywords: patch

Created on 2015-08-24 14:55 by christopher.hogan, last changed 2022-04-11 14:58 by admin. This issue is now closed.

Files
File name Uploaded Description Edit
append-sys-paths.patch christopher.hogan, 2015-08-24 14:55 review
issue24923.diff zach.ware, 2016-01-25 16:36 review
Messages (6)
msg249054 - (view) Author: Chris Hogan (christopher.hogan) * Date: 2015-08-24 14:55
Setup.py evaluates what's given in LDFLAGS and CPPFLAGS and CFLAGS. These variables are the usual mechanism to communicate custom paths/libs/defs to a build process. However, setup.py puts system paths in front of custom paths which makes it impossible to use these variables for providing search paths to overwrite system settings.

This patch puts system paths at the end of the list to avoid this problem.  Is there a rationale for prepending system paths?
msg258623 - (view) Author: Zachary Ware (zach.ware) * (Python committer) Date: 2016-01-19 21:50
I've been doing some research on this, the below is pretty much a brain-dump.

Related issues: issue232609, issue420565, issue10520.

This looks to me like the ideal solution is to assign a priority to each entry and sort by that priority before using it; give system locations low priority, /usr/local (sys.prefix) locations slightly higher priority, and user-defined paths highest priority.  That may be far more complex than it's worth, though.

It looks like setup.py doesn't actually look at CFLAGS at all, just CPPFLAGS and LDFLAGS.

Chris, is the behavior of 3.x any better for you?  There are some differences between the 2.7 and 3.x implementations of add_dir_to_list (most obviously, relative paths are always kept at the front of the list; see issue10520).

The naive solution looks to me to be to just move the CPPFLAGS/LDFLAGS checking further down detect_modules(), past all other calls to add_dir_to_list.  A test build with that change seems to work for me, but I'm not familiar enough with setup.py and how it works to call it correct.
msg258903 - (view) Author: Chris Hogan (christopher.hogan) * Date: 2016-01-25 15:00
Zach, 

3.x had the desired behavior.  We didn't have to make any changes.
msg258908 - (view) Author: Zachary Ware (zach.ware) * (Python committer) Date: 2016-01-25 16:36
In that case, does this patch do what you want it to do?
msg258912 - (view) Author: Chris Hogan (christopher.hogan) * Date: 2016-01-25 17:23
I should be able to try it out today or tomorrow.  I'll let you know.
msg382207 - (view) Author: Irit Katriel (iritkatriel) * (Python committer) Date: 2020-11-30 23:00
Fixed in Python 3 (in https://github.com/python/cpython/commit/807bd0a8a47191f7aa25cd9d5619bbc46a3f2a2a). Too late for 2.7.
History
Date User Action Args
2022-04-11 14:58:20adminsetgithub: 69111
2020-11-30 23:00:03iritkatrielsetstatus: open -> closed

nosy: + iritkatriel
messages: + msg382207

resolution: fixed
stage: patch review -> resolved
2016-01-25 17:23:53christopher.hogansetmessages: + msg258912
2016-01-25 16:36:24zach.waresetfiles: + issue24923.diff

messages: + msg258908
stage: patch review
2016-01-25 15:00:38christopher.hogansetmessages: + msg258903
2016-01-19 22:56:07r.david.murraysetnosy: + r.david.murray
2016-01-19 21:50:31zach.waresetnosy: + zach.ware
messages: + msg258623
2015-08-24 14:55:21christopher.hogancreate