You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In order to unblock #1878, patchelf needs to be updated.
Each update of patchelf usually comes with various regressions.
Rather than applying the update and wait for regressions bugs to pop-up as in #1421, let's wait for a bit of beta testing reports.
This issue serves for reporting success/failing updates.
Please add 🚀 reaction if you tested this successfully, otherwise please report the regression upstream linking to this issue.
In order to test this, patchelf can be updated in the image by running pipx install -f patchelf==0.19.1.0rc1.
If you use cibuildwheel, it could be as easy as adding it to your linux before-all command:
Besides raising the minimum version, a smoke test should exercise the specific behavior that requires patchelf 0.19.1. Since manylinux publishes several architecture and policy variants, the check should cover every image family that installs it. The documentation could also state whether the binary comes from the distribution or is built from source.
Besides raising the minimum version, a smoke test should exercise the specific behavior that requires patchelf 0.19.1. Since manylinux publishes several architecture and policy variants, the check should cover every image family that installs it.
This is done through auditwheel tests for now. There's no easy way to add this here (except in an unmerged PR).
The documentation could also state whether the binary comes from the distribution or is built from source.
Works great and fixes several issues with older versions of patchelf including this one: NixOS/patchelf#639
I understand the caution, but this version seems overall better and it is still possible for people to workaround things by installing another version on top in their CI config.
I used patchelf 0.19.1.0 w/ cibuildwheel to build wheels of freeze-core (local tests only), and it worked.
In cx_Freeze, enable it to be used in marcelotduarte/cx_Freeze#3415 and the functional tests passed.
Activity
[-]Update `patchelf>=0.19.0`[/-][+]Update `patchelf>=0.19.1`[/+]hiSandog commentedon Jul 11, 2026
Besides raising the minimum version, a smoke test should exercise the specific behavior that requires
patchelf0.19.1. Since manylinux publishes several architecture and policy variants, the check should cover every image family that installs it. The documentation could also state whether the binary comes from the distribution or is built from source.mayeut commentedon Jul 11, 2026
This is done through auditwheel tests for now. There's no easy way to add this here (except in an unmerged PR).
It already does say where the
patchelfbinary comes from: https://github.com/pypa/manylinux#image-contentfrankier commentedon Aug 6, 2026
Works great and fixes several issues with older versions of patchelf including this one: NixOS/patchelf#639
I understand the caution, but this version seems overall better and it is still possible for people to workaround things by installing another version on top in their CI config.
mayeut commentedon Aug 8, 2026
So far, no one reported anything breaking.
The next run of the update workflow will update patchelf to 0.19.1.0
marcelotduarte commentedon Aug 12, 2026
I used patchelf 0.19.1.0 w/ cibuildwheel to build wheels of freeze-core (local tests only), and it worked.
In cx_Freeze, enable it to be used in marcelotduarte/cx_Freeze#3415 and the functional tests passed.
mayeut commentedon Aug 15, 2026
#1982 has been merged with the updated patchelf