Hi Philipp,
Philipp Lohmann - Sun Germany wrote:
Kay Ramme - Sun Germany - Hamburg wrote:
Agreed, what about dropping support for "gdk_threads_set_lock_functions"
then and to always acquire (which AFAIK is already the case for Gnome <
2.4) the GDK mutex (gdk_threads_enter / ..._leave) before calling into
GDK? You are actually right, that from a coding standpoint the
difference isn't too big, but it IMHO is from testing point of view.
Michael's idea was that we should have a guranteed way of synchronizing
the lock count between gdk mutex and solar mutex. In theory dispatching
events (in multiple threads) could let you end up with a lock count that
is different from the gdk lock count, perhaps leaving you with a locked
solarmutex but actually unlocked gdk and/or vice versa. This would not
be a problem with in vcl itself, but as soon as you have other gtk based
in process services (like the file dialog or a gnome-vfs based UCB
content provider, a gtk based GUI from a theroetical UNO service), this
would probably lead to problems, because these would call the gtk
dispatch functions themselves which would then (being the same gtk
instance) dispatch also events to vcl; which now would acquire the solar
mutex again and gets out of sync with the gdk mutex; resulting in
probable deadlock.
may I summarize the requirement this way, that VCL must cooperate (read:
properly synchronize) with GTK/GDK in a way, that GTK/GDK may be used by
any other subsystem (e.g. a Uno component) independently of VCL? So, it
seems (because of the need to call "gdk_threads_set_lock_functions"
before "gdk_threads_init") that VCL must be initialized first, before
any other subsystem (which obviously is not a problem with the current
centralistic approach, but may become one if thinking in components
terms only (which I agree is hypothetical :-) ).
Since at the time (and probably still) this is a rather rare problem, we
kept 2.2 (not quite so old at the time) alive and had the secure
synchronizing mechanism from 2.4 upwards with basiclly no extra cost.
Understood. I certainly don't plan to break anything, just trying to
find my way through the code. So, I am pretty sure, that most testers
(and probably developers as well) are pretty unaware of the fact, that
they may need to test cooperation with Gnome < 2.4 and Gnome >= 2.4 .
Just my 2 cents, pl
Again, thank you very much for your help
Kay
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
