Repository navigation
MNT: ci: MacOS: replace 'gobject-introspection' dependency with 'glib' - #29777
jayaddison wants to merge 30 commits into
Conversation
The `girepository` dependency previously provided by this dep has been migrated[1] into `glib`, and `glib` is listed as a dep of the `gtk4` package. [1] - https://docs.gtk.org/girepository/migrating-gi.html
This reverts commit 1949930.
This reverts commit 38cf554.
…rsion" This reverts commit 7520638.
These packages are available via PyPI (e.g. using `pip install`), but are also available -- and listed as dependencies of -- `gtk4` in Homebrew, which we already install.
|
(I probably should have temporarily disabled non-MacOS GitHub Actions workflow jobs during this pull request.. in any case, I am going to pause to wait for the latest set of CI results before pushing any more commits) |
|
There is some relevant documentation that may help to figure out how to proceed: matplotlib/doc/install/dependencies.rst Lines 57 to 60 in a9dc9ac |
Relates-to / replays commit 1bfc185.
This reverts commit 6bb8581.
This reverts commit 447dd90. Conflicts: .github/workflows/tests.yml
The `girepository` dependency previously provided by this dep has been migrated[1] into `glib`, and `glib` is listed as a dep of the `gtk4` package. [1] - https://docs.gtk.org/girepository/migrating-gi.html (cherry picked from commit 38052b9)
|
The latest status update here (52ee08e) is:
The relevant error in the build logs is found at: https://github.com/matplotlib/matplotlib/actions/runs/14048541275/job/39334454171#step:10:307 ...as already encountered during some of my previous testing last week. I'm not sure I want to expend much more CI compute resource on this (especially after realizing that I could and should have disabled Azure Pipelines during this PR). |
|
I'm still confused by the apparent absence of the Could the fact that I'm taking a break from this for a while. |
| lmodern \ | ||
| ninja-build \ | ||
| pkg-config \ | ||
| python3-cairo-dev \ |
There was a problem hiding this comment.
NB: this isn't directly related to the MacOS build; but if we do install PyGObject from PyPI without retrieving any dependencies -- as I suggest in 6d41f1a -- then I think we'll need this system package installed during Linux-based CI workflows.
|
Tests passed on MacOS from commit 6d41f1a, but that is deceiving; neither GTK3 nor GTK4 loaded successfully, and as a result, testing of those backends was skipped. edit: fixup for unintentional double-negative |
Also, remove a potentially-misleading comment.
…pendency [skip azp]
|
I think it'll make more sense to recommence this migration from a fresh branch; it doesn't seem worthwhile to rebase exploratory commits (because they are useful primarily in their original form) nor to attempt to merge significantly-conflicting upstream changes into this branch. |
PR summary
The
girepositorydependency previously provided bygobject-introspectionhas been migrated intoglib.To allow upgrades of
pygobject(currently restricted to<3.52), we need to resolve the failure case originally reported in #29732.Unblocking upgrades of
pygobjectallows this project to stay current and receive updates and fixes in updated versions of that library.Initially this is a draft PR where I'm only removing a legacy Homebrew
gobject-introspectiondependency.Unlike the Linux-based counterpart, where I do find various reports of problems due to an updated
girepositoryversion requirement, I don't find similar reports mentioning MacOS -- so perhaps our dependency list is simply out-of-date andgobject-introspectioncauses a conflict?PR checklist
May resolve #29732.