GitLab

This conflicts with the one from the view switcher.

This makes the header work in both desktops and mobiles.

This prevents the window from being stupidly small.

Make it work well on desktops and phones alike.

Currently, every Flathub app is show as using a "legacy display manager"
whether they use x11 with --socket=x11, or wayland with
--socket=wayland and --socket=fallback-x11.

However, the fallback-x11 enables X11 only if the host doesn't support
anything else than x11, so strictly speaking, it's the host that uses
x11 and not usually the app.

So this commit stops from showing this warning if using x11 as fallback.

Spotted by Coverity.

Spotted by Coverity.

This means that global state changes, such as to the `no-interaction` flag
of the `FlatpakInstallation`, don’t escape outside the context of the
transaction currently being run.

This is a follow up to !283, using a suggestion by Matthias Clasen.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

That commit added extra padding in the primary menu that no other app has,
remove it.

This reverts commit 26c02f45.

In gs_flatpak_set_kind_from_flatpak(), we use AS_APP_KIND_LOCALIZATION
and AS_APP_KIND_GENERIC in addition to AS_APP_KIND_RUNTIME, all for
things which from Flatpak's perspective are runtimes (their ref starts
with "runtime/"). So change checks in a few places to correctly identify
Flatpak runtimes instead of checking for AS_APP_KIND_RUNTIME, which has
the effect of ensuring that applications properly show as needing an
update when their locale extension needs an update, e.g. because the
user chose a language other than the one the locale extensions already
have.

Together with an update of libflatpak to 1.7.3 when that is released,
this commit addresses the issue described in this MR:
https://gitlab.gnome.org/GNOME/gnome-software/-/merge_requests/430

Closes: #539

If the user double clicks on a .cab file on a USB stick to install it, it is
certainly not expected we delete the archive when it has been scheduled.

This resolves issue #967

The appdata dir is unlikely to exist on newer systems, so check for the
more-likely file first.

This actually properly resolves #918.
Missing stock icons occur quite frequently, as pretty much any icon can
be overridden by the theme and therefore stock icon entries are
plentiful, but no theme covers all icons.
Therefore, we'll add one that we know exist (the AppStream cached icon)
as backup.

This resolves #918
Missing stock icons occur quite frequently, as pretty much any icon can
be overridden by the theme and therefore stock icon entries are
plentiful, but no theme covers all icons.
Of course, this change also protects us against wrong icon data.

Many thanks to Laurent Bigonville for his work on this issue!

It causes a critical warning. `gs_app_get_name()` can legitimately
return `NULL`, so guard against that situation (at the possible expense
of some less-than-ideal sorting).
Signed-off-by: Philip Withnall <withnall@endlessm.com>

It was getting emitted a lot (possibly on every refine operation),
slowing things down a little, and clogging up the `--verbose` output.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

This won’t affect performance much, but does make the code a little
cleaner. There’s no need to query whether an element exists in a hash
table before removing it or updating it: instead, check the return
value from the remove/update function.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

`g_hash_table_get_{keys,values}()` are quite inefficient: they have to
iterate over the hash table and copy the keys/values into a newly
allocated linked list, return that, and then the calling code has to
iterate over that again.

Instead, use `GHashTableIter`, which does zero allocations and iterates
directly in the calling code.

This introduces no functional changes.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

Rather than querying them on every loop iteration. In a trace where I
switched between categories a few times, `gdk_pixbuf_get_width()` was
being avoidably called 61 million times.

This introduces no functional changes.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

It was being called 80 million times in a trace where I switched between
a couple of categories a few times, and taking 1% of all CPU time in the
trace. Inlining it will reduce some of the calling overheads.
Signed-off-by: Philip Withnall <withnall@endlessm.com>

Fixes https://gitlab.gnome.org/GNOME/gnome-software/-/issues/965

This is only used for local distro packages that do not ship appdata files.

Fixes https://gitlab.gnome.org/GNOME/gnome-software/issues/984

Fixes https://gitlab.gnome.org/GNOME/gnome-software/issues/983

As per discussion with jimmac on #fedora-design.