GitLab

An info bar packed in the middle of the window requires its own window,
to be drawn properly.

Closes #1065
Closes !510

(cherry picked from commit 510266dd)

(cherry picked from commit 26204524)

(cherry picked from commit d5bef205)

(cherry picked from commit 94b33a4e)

(cherry picked from commit e7987360)

(cherry picked from commit e785e47f)

When getting the main ref of a related one, the ref kind used for
trying to match it to an installed one was always the "app" kind. This
may have been because initially, all related apps had a regular app as
their main one. However, runtimes do have extensions too, and thus the
main app in this case will be a "runtime" type one.

Thus, this patch gets the right ref kind for the main app based on
the ref name's prefix.

This commit fixes offline installation of .flatpak bundles, and fixes
offline installation of .flatpakref files that point to a file:// URL.
Currently such installations would be forever stuck in the "Pending
installation..." state. The check for a file:// prefix on the
gs_app_get_origin_hostname() return value is wrong because in case a
file:// URL is set, "localhost" is returned. See gs_app_func() in
lib/gs-self-test.c.

One might wonder if this could be implemented by checking if
gs_app_get_local_file() returns a non-NULL value, but
gs_app_set_local_file() is called for flatpakref files which are not
always offline friendly.

This commit comes from
https://github.com/endlessm/gnome-software/pull/557

Use the modal dialog queue of the GsShell to have the window closed
when the GsShell changes the page.

Closes #41
Closes !507

`anything_ran` is set to indicate that at least one plugin handled a
certain job (even if it returned an error). However, it’s not set on all
error paths in `gs_plugin_loader_call_vfunc()`. Ensure it is, to avoid
an incorrect error potentially being reported.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

The fedora-langpacks plugin disables itself if not on fedora. Skip the
self tests in that situation as well so they don't fail. The reason this
hasn't caused CI failures is that we have "--no-suite fedora-langpacks"
in .gitlab-ci.yml so that it doesn't run under Debian. But it still
seems worth skipping as people may run the tests manually, and we need
to skip it in the downstream Endless CI.

Signed-off-by: Philip Withnall <withnall@endlessm.com>

Otherwise the main context iteration where loading is marked as complete
has too much happening in it, and takes many cycles in GTK’s frame
clock. This makes the UI freeze for a noticeable fraction of a second.

Partially avoid that by splitting up some of the work here between two
main context iterations. The UI still freezes a bit; further work is
required.
Signed-off-by: Philip Withnall <withnall@endlessm.com>