This reverts the revert commits d264dbf1
and e97dc693.

Even though the GNOME Shell extension's logic to avoid double-starting Florence
is flawed, it is only flawed when one 0. uses Florence; 1. uses a system where
Florence needs more than 1 second to start; and 2. one waits more than 1 second,
but less than the time it takes to start before clicking on the icon again.

The alternate option this commit reverts has worse problems: Florence sometimes
is not hidden on startup, which affects all users, and requires an ugly
workaround in the test suite (which is broken on feature/stretch currently, so
this comes with a maintenance cost). Besides, it eats memory for everybody,
while I expect the vast majority of users don't actually use Florence.

So, all in all, I prefer going back to the behaviour previously implemented
in feature/stretch, that I had reverted earlier today.

This reverts commit 1d9d5609.

We've reintroduced the Pictures directory and friends.

Revert "Test suite: drop obsolete workaround for Florence sometimes not being hidden on startup (#11398)."

This reverts commit 7478982c.

We start it automatically again.

Revert "Don't automatically start Florence: the Florence Indicator GNOME Shell extension will start it the first time one tries to display it."

This reverts commit 98838908.

The GNOME Shell extension we use has logic to avoid double-starting Florence,
using a hard-coded delay. I've seen it fail (possibly due to our time sync'
system, or just because that hard-coded delay may cause robustness issues
depending on the hardware). When it fails, then the toggle logic is completely
broken in terms of UX, since hiding one instance of the virtual keyboard
shows the next one, etc.

Closes: #12052

Removing eatmydata did not affect build time when building in RAM.
Building on disk is a corner case so let's not bother optimizing it:
it'll be super slow anyway. Patches welcome to reintroduce eatmydata
without breaking the build :)

This reverts commit 851f3e7a.

This reverts commit 40332999.

session-c1.scope won't shut down before the default timeout (1min 30s),
and I didn't manage to change its timeout, nor to let shutdown.target
ignore it.

So let's change the default timeout globally. This feels hackish, but OTOH it
means that shutdown won't ever be blocked by any buggy service that takes
a while to stop, which can be seen as a safety feature in the context of Tails,
since we want to wipe memory afterwards. Maybe.

The idea is to enable:

 * users to report more useful bugs;
 * developers to better understand the failure we see in the test suite.

Due to #12061, we typically wait about 90 seconds before we start wiping the
memory, so let's factor this in as an additional constant, instead of as
a multiplication factor.

Install libc6-i386, needed for Tails Installer to run the syslinux:i386 from the ISO (refs: #11873).

Apparently that's needed for them to be taken into account.
It's weird, but we've seen that before for some other settings.

Fix-committed: #11853, #11850

It's required by openpgp-applet.

intrigeri authored Nov 10, 2016 and anonym committed Dec 21, 2016

Currently our devel branch FTBFS since Stretch got openpgp-applet 1.0,
that has a new dependency. For now let's revert to 0.9.1, that is the
version we installed before openpgp-applet 1.0 migrated to testing,
modulo "~bpo8+1". Ticket #11899 is there to track the upgrade to 1.0.

refs: #11899

When we skipped the 2.8 major release we didn't extend the 2016092002
snapshot's Valid-until.

Fix-committed: #12038

This reverts commit 79b6ad8a.

Closes: #11596

This reverts commit 52f4a370.

seahorse-nautilus is back in Stretch.

refs: #12004

This is not supported in GNOME anymore:

 * https://github.com/GNOME/gnome-settings-daemon/commit/69d9d8b1eda2d50acca5b4da58003afb8b82a7f1
 * https://bugzilla.gnome.org/show_bug.cgi?id=755953

… and it doesn't work on all hardware.

It contains fuser.