Only started, still more to do, just sketching at this point.

ListCachedUsers could theoretically be somewhat slow,
so set an unlmited timeout for it.

We were just returning an uninitialized variable.

Under current design and usage patterns, the library user is not
expected to keep the manager around, if only interested in signals
from ActUsers. Therefore additional references must be taken to
ensure that the object is still valid at the end of an async call,
and signals must be prevented from firing on finalized objects.

https://bugs.freedesktop.org/show_bug.cgi?id=46797

Passing 0 means 0 milliseconds, which is quite too short even on
modern computers.

https://bugs.freedesktop.org/show_bug.cgi?id=46794

Not doing so confuses the UserManager, which still considers the
user pending and blocks the emission of user-added/user-removed
signals.

https://bugs.freedesktop.org/show_bug.cgi?id=46794

Because the session proxy creating happened synchronously, seat
loading state was incremented, but nothing called load_seat_incrementally()
again, meaning that it never advanced to next step. Making it async
like the other operations fixes it.

https://bugs.freedesktop.org/show_bug.cgi?id=46794

https://bugs.freedesktop.org/show_bug.cgi?id=46283

https://bugs.freedesktop.org/show_bug.cgi?id=46283

https://bugs.freedesktop.org/show_bug.cgi?id=46283

Some of the recent code for iterating through
the results of dbus code uses i when isn't an
integer.  I find that really hard to wrap my head
around, and it appears to be causing crashes.

(I believe the loop condition should be *i not i)

This commit changes things up to be, in my opinion,
clearer.

https://bugs.freedesktop.org/show_bug.cgi?id=46170

Based on work by Robert Ancell,
https://bugs.freedesktop.org/show_bug.cgi?id=40364

Based on work by Robert Ancell,
https://bugs.freedesktop.org/show_bug.cgi?id=40364

long isn't necessarily 64bit.  On big endian
architectures, where it's smaller than 64bit,
the compute_object_path function will get get
computed incorrectly.

Based on patch by Michel Dänzer <daenzer@debian.org>

http://bugs.freedesktop.org/show_bug.cgi?id=45738

Also support the GNOME build api (NOCONFIGURE)

When the current icon is in /usr/share, and we reset the
icon, we try to delete a nonexisting icon file in
/var/lib/AccountsService/icons. Deal with this eventuality.

Since we do checks that seem to imply it can't be NULL, be
consistent and assume everywhere that it may be NULL.

Ed Schouten authored Oct 13, 2011 and Matthias Clasen committed Oct 17, 2011

The fgetpwent() function seems to be a Linux-specific extension. POSIX
does not specify such a function. I think we can even get rid of
fgetpwent() altogether, since we use it on /etc/passwd, which is already
the default.
Signed-off-by: Ed Schouten <ed@80386.nl>

https://bugs.freedesktop.org/show_bug.cgi?id=41747

Ed Schouten authored Oct 13, 2011 and Matthias Clasen committed Oct 17, 2011

POSIX does not specify a header file called <wait.h>. This seems to be a
Linux-specific feature. It's called <sys/wait.h>.
Signed-off-by: Ed Schouten <ed@80386.nl>

https://bugs.freedesktop.org/show_bug.cgi?id=41747

Ed Schouten authored Oct 13, 2011 and Matthias Clasen committed Oct 17, 2011

The <shadow.h> header is a Linux-specific extension. On FreeBSD, all
functionality is integrated into the <pwd.h> interface.
Signed-off-by: Ed Schouten <ed@80386.nl>

https://bugs.freedesktop.org/show_bug.cgi?id=41747

Ed Schouten authored Oct 13, 2011 and Matthias Clasen committed Oct 17, 2011

Not all operating systems out there support utmpx. Examples include
OpenBSD, but also FreeBSD prior to version 9.0.

Also, FreeBSD's utmpx (fortunately) does not support utmpxname(), to
prevent applications from hardcoding the path to the wtmp file. Use the
setutxdb() function to switch to the wtmp database.
Signed-off-by: Ed Schouten <ed@80386.nl>

https://bugs.freedesktop.org/show_bug.cgi?id=41747

We don't want to call cat on a FIFO, for instance.

If a user happens to be logged in at a VT, we don't
want to show them as "already logged in" or whatever.

This commit filters out vt sessions in the same way it
filters out sessions associated with the login screen.

The passwd file handling code clears the list of loaded
users so we need to always do wtmp on top of it, instead of
before it.

The stylesheet was producing invalid docbook in some places.

libaccountsservice was trying to launch 'actflexiserver', which
does not exist. Use 'gdmflexiserver' instead, which does exist.
This was a copy-and-paste error from when libaccountsservice was
first introduced, a year ago.

Often the two files are changed in concert, so it makes sense to
wait a bit before doing a heavy reload operation.

Suggested by Vincent.

https://bugs.freedesktop.org/show_bug.cgi?id=39413

We also emit relevant signals when loading the file.

https://bugs.freedesktop.org/show_bug.cgi?id=39413

It's better to consolidate references to file path in one
place up top.

Right now, if a user is not configured for automatic login, and
SetAutomaticLogin(false) is called for that user then we disable
automatic login system wide even if someone else is configured
for automatic login.

This commit changes the semantics of the call such that
SetAutomaticLogin(false) is a no op for users where automatic login
isn't already enabled.

Spotted by Vincent Untz <vuntz@gnome.org>

https://bugs.freedesktop.org/show_bug.cgi?id=39402