memorytexture test on big-endian regressed in "gsk: Use has_bgra in more places"
Since 4.12.4, the memorytexture test fails:
test: gtk:gdk / memorytexture start time: 16:40:53 duration: 0.64s result: killed by signal 6 SIGABRT ... 1..2072 # Start of memorytexture tests # Start of download_1x1 tests # Start of b8g8r8a8-premultiplied tests ok 1 /memorytexture/download_1x1/b8g8r8a8-premultiplied/local not ok /memorytexture/download_1x1/b8g8r8a8-premultiplied/gl - ERROR:../../../testsuite/gdk/memorytexture.c:819:compare_textures: 'gdk_memory_format_pixel_equal (format, accurate_compare, data1 + bpp * x, data2 + bpp * x)' should be TRUE
Big-endian ppc64 has a similar failure mode, suggesting that this is an endianness problem.
Bisecting indicates 41dbb7a7 as the first bad commit. The following commit 5ff9316e was meant to fix big-endian, but it appears that it didn't.
It looks as though this can be solved by always taking the "BGRA unsupported" code path on big-endian.