GTK and WPE share a number of components, like GStreamer and Soup, leading to common expectations for both bots in a number of tests (For example, Charlie Turner found 661 EME tests with common expectations between both ports). Currently: GTK: platform/gtk -> platform/wk2 -> generic GTK-wayland: platform/gtk-wayland -> platform/gtk -> platform/wk2 -> generic WPE: platform/wpe -> platform/wk2 -> generic Goal: GTK: platform/gtk -> platform/glib -> platform/wk2 -> generic GTK-wayland: platform/gtk-wayland -> platform/gtk -> platform/glib -> platform/wk2 -> generic WPE: platform/wpe -> platform/glib -> platform/wk2 -> generic
Created attachment 397961 [details] Patch
Comment on attachment 397961 [details] Patch Didn't run testwebkitpy before. Will update. Sorry.
Created attachment 397971 [details] Patch
Comment on attachment 397971 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=397971&action=review Nice! In the future, gtk-wayland expectations should go away since Wayland is default for ages. We can add gtk-x11 that inherits from main gtk results, like gtk-wayland does now. > LayoutTests/platform/glib/TestExpectations:9 > +# If you get different test results than the bots for some tests > +# even when using flatpak, then you can re-try using a > +# chroot of the same distro and version the bots use (see below). > +# If doing that fixes the issue, it means we need to identify the library > +# that is causing the difference on the results, and add it to Flatpak. That was true in jhbuild, not for flatpak. flatpak doesn't expose host dependencies at all, so the only way the host system should affect the test results would be via the very few things that flatpak exposes from the host system, most notably fonts.
(In reply to Michael Catanzaro from comment #4) > In the future, gtk-wayland expectations should go away since Wayland is > default for ages. We can add gtk-x11 that inherits from main gtk results, > like gtk-wayland does now. Yeah. This could be done once we swap the main bots to wayland too (GTK-Release/GTK-Debug). (In reply to Michael Catanzaro from comment #4) > Comment on attachment 397971 [details] > That was true in jhbuild, not for flatpak. flatpak doesn't expose host > dependencies at all, so the only way the host system should affect the test > results would be via the very few things that flatpak exposes from the host > system, most notably fonts. True. For layout tests, even fonts should not be an issue, as flatpak bundles webkitgtk-test-fonts directly inside /usr/share and the WTR InjectedBundle searches there first. I'll remove this paragraph altogether.
Created attachment 397986 [details] Patch without jhbuild paragraph
(In reply to Lauro Moura from comment #5) > True. For layout tests, even fonts should not be an issue, as flatpak > bundles webkitgtk-test-fonts directly inside /usr/share and the WTR > InjectedBundle searches there first. I'll remove this paragraph altogether. Yes. But keyword is "should." :) The host fonts are exposed, and in theory could cause layout test failures if the tests are not careful to use only fonts provided by webkitgtk-test-fonts.
BTW this looks good to me. I'd give r+, except I think cgarcia will want to approve a change of this magnitude.
Committed r261383: <https://trac.webkit.org/changeset/261383> All reviewed patches have been landed. Closing bug and clearing flags on attachment 397986 [details].
<rdar://problem/63021855>