Opened 8 years ago
Closed 4 years ago
#771 closed defect (wontfix)
Python discovery is sometimes wrong
Reported by: | tdsmith | Owned by: | strk |
---|---|---|---|
Priority: | minor | Milestone: | 3.10.0 |
Component: | Build/Install | Version: | 3.5.0 |
Severity: | Unassigned | Keywords: | |
Cc: |
Description
The Python discovery logic in configure is a little surprising. If python-config exists, asking it for include paths with python-config --includes
may be more robust than using find in the parent of the directory holding the Python binary to locate a Python.h.
Here is a report of a situation where, building the geos python extensions against OS X's system python, pypy's Python.h was discovered instead of Python's, which caused trouble: https://github.com/Homebrew/homebrew/issues/49725
Also, extension modules built on OS X generally should not explicitly provide -lpython to the linker and should use -undefined dynamic_lookup instead, so that it can be used with any compatible Python interpreter. I've written a blog post that describes this problem more extensively: http://blog.tim-smith.us/2015/09/python-extension-modules-os-x/
Thanks! Tim
Change History (2)
comment:1 by , 4 years ago
Milestone: | → 3.10.0 |
---|
comment:2 by , 4 years ago
Resolution: | → wontfix |
---|---|
Status: | new → closed |
We seem to have evolved beyond this ticket by just not having our own python support anymore, it'll all delegated to pygeos and shapely projects.