Move MSYS2 to GitHub Actions #4642
Conversation
| @@ -660,6 +661,7 @@ def test_variation_get(self): | |||
| {"name": b"Size", "minimum": 0, "maximum": 300, "default": 0} | |||
| ] | |||
|
|
|||
| @pytest.mark.skipif(is_mingw(), reason="epsilon too high for meaningful test") | |||
There was a problem hiding this comment.
I'm not sure what is causing the difference, but I am getting epsilon around 33 for both variation tests. Checking manually, I see there is a significantly different kerning after the first letter, but the output looks correct to me.
Note: MSYS2 is currently using FreeType version 2.10.2.
There was a problem hiding this comment.
Please could you attach images for each case?
There was a problem hiding this comment.
Sure.
test_variation_set_by_name (epsilon 31.5340 on line 686 of PR):
a.png (MSYS):
b.png (target - variation_adobe.png):
test_variation_set_by_axes (epsilon 39.6856 on line 720 of PR):
a.png (MSYS):
b.png (target - variation_adobe_axes.png):
Basic and Raqm layout are pixel identical. In fact, it looks like there is a bug in the tests where the used layout_engine is not being explicitly set and defaults to Raqm in both cases.
# Conflicts: # .appveyor.yml # .github/workflows/test-windows.yml
|
I decided to test Cygwin, and aside from Raqm (which was not found) and two test which should have been skipped, everything passes on this PR. It turns out that |
|
Thank you! |
For #3606 and #4495 (comment).
Move MSYS2 from AppVeyor to faster GHA, now that it is available.
Changes proposed in this pull request:
_WIN32instead of_MSC_VERin_imagingft.cto fix FreeType compile.libraqm-0as alternate Raqm filename for MinGW.OPJ_STATICon MinGW to fix python setup.py install: Jpeg2KDecode.c: undefined reference to `opj_stream_create' #2848. Inspired by the patch from MSYS2 repo: https://github.com/msys2/MINGW-packages/blob/master/mingw-w64-python-pillow/002-shared-openjpeg.patchTODO:
libtiffis failing to open files for some reason (I tried both 4.1.0 and 4.0.10), disabled for now.Looking at the MSYS2 repo, they don't seem to have a patch for
libtiff, yet somehow that version passes tests. Edit: https://trac.wxwidgets.org/ticket/17286 may be related.