use libffi instead of older vendored version #48
Conversation
mattip
requested review from
isuruf,
ocefpaf,
ohadravid and
omerbenamram
as code owners
|
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
|
@conda-forge-admin, please rerender |
|
ping @h-vetinari |
|
Hi! This is the friendly automated conda-forge-webservice. |
|
Cool, thanks for figuring this out Matti! I don't understand the dependence on virtualenv though? I see you're carring the same diff in the patch here, why would we need to wait for pypa/virtualenv#2141? Also, the patch doesn't seem to apply cleanly on top of 7.3.5 (+ the existing patches)... |
|
We don't need to wait for virtualenv, upstream PyPy does. So even though I made this a "pre-7.3.6-0008.patch", it may need to be carried over past v7.3.6. I will fix the diff. |
|
There seems to be an issue (best guess from the trace) with how the shared lib extension is applied? Not sure where along the way |
|
Probably an issue with the compilation LIB directive to add the path. I am trying to debug locally |
recipe/patches/pre-7.3.6-0008.patch
Outdated
| --- a/pypy/tool/release/package.py Mon Apr 12 07:11:48 2021 +0200 | ||
| +++ b/pypy/tool/release/package.py Sun Jul 11 12:46:37 2021 +0300 | ||
| --- a/pypy/tool/release/package.py 2021-4-12 07:11:48.000 +0200 | ||
| +++ b/pypy/tool/release/package.py Sun Jul 11 12:46:37.000 2021 +0300 |
There was a problem hiding this comment.
Aside from the fact that git (& patch) should easily handle the other dates as well, this looks inconsistent (now a mix of both formats). In any case, the patch seems to have been applied without issue:
Patch analysis gives:
[[ RA--D1L-VE ]] - [[ patches/fficurses.patch ]]
[[ RA--D1LO-E ]] - [[ patches/tklib_build.patch ]]
[[ RA--D1LOVE ]] - [[ patches/tk_app.patch ]]
[[ RA--D1L-VE ]] - [[ patches/lzma_build.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/site-and-sysconfig-conda.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/distutils-install.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/distutils-sysconfig-get_python_lib.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/0009-runtime_library_dir_option-Use-1st-word-of-CC-as-com.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/0012-Fix-find_library-so-that-it-looks-in-sys.prefix-lib-.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/0013-Disable-new-dtags-in-unixccompiler.py.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/0020-Use-ranlib-from-env-if-env-variable-is-set.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/pre-7.3.6-0001.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/pre-7.3.6-0002.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/pre-7.3.6-0003.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/pre-7.3.6-0004.patch ]]
[[ RA-MD1LOVE ]] - [[ patches/pre-7.3.6-0005.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/pre-7.3.6-0006.patch ]]
[[ RA-MD1L-VE ]] - [[ patches/pre-7.3.6-0007.patch ]]
[[ RA-MD1L--E ]] - [[ patches/pre-7.3.6-0008.patch ]]
There was a problem hiding this comment.
yeah, I got confused by a failure at some point and tried to modify this by hand. I think patch ignores the date altogether, so it shouldn't matter
|
@isuruf, if fine for you, would be good to merge this for continuing with conda-forge/numpy-feedstock#238 |
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)It turns out PyPy was vendoring a very old version of libffi for msvc. Conda provides a newer version, so use it. This seems to fix the segfaults in conda-forge/numpy-feedstock#238. While the patch comes from an upstream branch, the branch itself requires pypa/virtualenv#2141 so I am optimistically marking this as part of the next pypy release.