Build Runtime and test it

[sergio-costas]

This PR builds the runtime against a new SDK, and tests it against several snaps. Specifically, it tests:

• firefox
• thunderbird
• darktable
• cheese
• gimp
• gnome-calculator
• gnome-system-monitor
• gnome-text-editor
• evince
• kicad
• chromium
• shotwell
• telegram
• mattermost

It uses YARF OCR capabilities to detect that the application has been launched correctly. After it detects it, it waits five seconds and test again, to ensure that it is still working (thus to avoid the case where the window does appear but the application crashes immediately).

If a test fails, it uploads an artifact with the YARF logs, to check if it's a test problem or a PR problem.

[3v1n0]

Is this good to go now?

A part squashing the last commit with the initial one, I feel that we're good here.

[sergio-costas]

I still want to add the tests themselves: to install several snaps, launch them, and ensure that they run instead of failing.

[3v1n0]

I still want to add the tests themselves: to install several snaps, launch them, and ensure that they run instead of failing.

Fair.. Well, can we just add some very dummy test for now (just ensure that some simple app starts and closes without failing)?

So we can unblock #294 or can we land that without this?

[sergio-costas]

@3v1n0 I squashed all the commits, and it's ready for review.

[3v1n0]

Thanks for this and lots of tests.

There are some things to re-enable, but first I've some cleanups to suggest.

Also, checking the logs, maybe it's also worth to install the xdg-desktop-portal (gtk or gnome)? As they are used by some snaps and in general it's a requirement component to get the snaps working

[3v1n0]

Maybe temporarily revert 09aca1c :)

[sergio-costas]

@3v1n0 mattermost was updated this last friday, and now it doesn't launch. It returns a GPU problem. Any idea?

[3v1n0]

I would use a clearer name what this is about

[sergio-costas]

What do you propose, instead?

[3v1n0]

I would still like to launch a proper dbus session, or even the whole systemd --user session where we test, so that we actually have the portals running as we are getting currently:

(evince:7698): Gdk-WARNING **: 11:53:00.832: Failed to read portal settings: GDBus.Error:org.freedesktop.DBus.Error.UnknownMethod: No such interface “org.freedesktop.portal.Settings” on object at path /org/freedesktop/portal/desktop

[sergio-costas]

I would still like to launch a proper dbus session, or even the whole systemd --user session where we test, so that we actually have the portals running as we are getting currently:

I tried over and over, but it fails due to a cgroups problem.

[sergio-costas]

I would still like to launch a proper dbus session, or even the whole systemd --user session where we test, so that we actually have the portals running as we are getting currently

In fact, I'm already configuring the DBUS environment variable to point to the already-running dbus-session instance, and both dconf-daemon and xdg-desktop-portal binaries are being automatically launched, so that error must be due to a different reason.

[sergio-costas]

The problem with mattermost is in mattermost itself, so we have to wait for a new version.

[3v1n0]

I would still like to launch a proper dbus session, or even the whole systemd --user session where we test, so that we actually have the portals running as we are getting currently

In fact, I'm already configuring the DBUS environment variable to point to the already-running dbus-session instance, and both dconf-daemon and xdg-desktop-portal binaries are being automatically launched, so that error must be due to a different reason.

Could be because the gtk portal does not implement it? Should we try the gnome one?

[sergio-costas]

I restored all the flags.

[sergio-costas]

@3v1n0 Maybe the problem with Evince is this? Apr 15 09:24:34 runnervm35a4x dbus-daemon[14816]: apparmor="DENIED" operation="dbus_method_call" bus="session" path="/org/freedesktop/portal/desktop" interface="org.freedesktop.host.portal.Registry" member="Register" mask="send" name="org.freedesktop.portal.Desktop" pid=14871 label="snap.evince.evince" peer_pid=15231 peer_label="unconfined"

[sergio-costas]

@3v1n0 Ok, I restored everything. When mattermost is fixed, I think that we can merge this.

[sergio-costas]

@3v1n0 Ok, all tests are passing!!!!! The only open conversation is the name for the YARF logs files.

[aleasto]

mmh, sketchy. we could avoid lying about being a GNOME desktop instead, and it should default to xdg-desktop-portal-gtk already.

On the other hand, I know for a fact that some third party applications check for XDG_CURRENT_DESKTOP=GNOME before enabling some features so maybe this more realistic than not lying about being GNOME...