| Age | Commit message (Collapse) | Author |
|---|
|
Change-Id: I4359b7042f98586e2c9f5529d83d769cdf3d033c
Reviewed-on: https://gerrit.libreoffice.org/85775
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
|
|
Change-Id: Ie24f56fed811b3b317d6c20e15289e7cecfb738d
Reviewed-on: https://gerrit.libreoffice.org/81157
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
|
|
Found with bin/find-unneeded-includes
Only removal proposals are dealt with here.
There were some changes since last run and some omitted files
Change-Id: I666ac8ed7d06684e252ca590e3d7d454e9e10975
Reviewed-on: https://gerrit.libreoffice.org/70497
Tested-by: Jenkins
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
|
|
If a task is still in the scheduler priority queue and its
priority is changed, it won't be moved into the correct queue.
We have to track the priority of the scheduled task, so we can
warn the developer to fix the code and actually handle re-start
correctly.
Since we don't want to traverse the whole Scheduler queues on
priority change (which sometimes get very long) to remove the
wrong data item, we'll just invalidate it, if a priority change
is detected.
This also reverts commit d24b264c4a47 ("vcl opengl: avoid task
priority warning on cursor blink"), which tried to avoid the
warning, which was just half right and independent of the broken
priority change handling.
LO doesn't change priorities of scheduled tasks normally, so
that bug didn't turn out to have much impact, I guess.
Change-Id: I6e46b518a7c3532047c619c013bd8597f73ed7a6
Reviewed-on: https://gerrit.libreoffice.org/69249
Tested-by: Jenkins
Reviewed-by: Jan-Marek Glogowski <glogow@fbihome.de>
|
|
Found with bin/find-unneeded-includes
Only removal proposals are dealt with here.
Change-Id: Ib32ea54a3fa690a0722fa75ddb4121ec78c0b64f
Reviewed-on: https://gerrit.libreoffice.org/65386
Tested-by: Jenkins
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
|
|
This way we don't have to search the whole list for a higher
priority event, if an immediate Task is found. This probably
helps bugs like tdf#119724 and tdf#119428.
Change-Id: Ic5685193d1bedb6996cf46f0ee2cba42190ff7cc
Reviewed-on: https://gerrit.libreoffice.org/60572
Tested-by: Jenkins
Reviewed-by: Jan-Marek Glogowski <glogow@fbihome.de>
|
|
Change-Id: Ifd83affcb16209f4134c725640fbd95077c8ab0f
Reviewed-on: https://gerrit.libreoffice.org/59099
Tested-by: Jenkins
Reviewed-by: Caolán McNamara <caolanm@redhat.com>
Tested-by: Caolán McNamara <caolanm@redhat.com>
|
|
A task has to get the SchedulerLock to remove itself from the
Scheduler list. This doesn't work, if the Task is static, as the
static Scheduler might be destroyed earlier. In this case we fail
with the following backtrace:
#0 SchedulerMutex::acquire
#1 Task::~Task
#2 __run_exit_handlers
Thanks to Michael Stahl to catching this backtrace.
As a workaround this marks static tasks, so they ignore the
SchedulerMutex in the destructor, We also mark all scheduled Tasks
as "static" in DeInitScheduler, as their cleanup was already done.
In the end all Tasks should be removed from static objects.
Change-Id: I38be3206378b9449193efaccbc96896ac8de9478
Reviewed-on: https://gerrit.libreoffice.org/42574
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Jan-Marek Glogowski <glogow@fbihome.de>
|
|
This is based on glibs classification of tasks, but while glib uses
an int for more fine grained priority, we stay with our enum.
1. Timers start with DEFAULT priority, which directly corresponds
with the previous HIGH priority
2. Idles start with DEFAULT_IDLE priority instead of the previous
HIGH priority, so idle default becomes "really run when idle".
As RESIZE and REPAINT are special, and the DEFAULTS are set, there
is just one primary decision for the programmer: should my idle
run before paint (AKA HIGH_IDLE)?
If we really need a more fine-grained classification, we can add it
later, or also switch to a real int. As a result, this drops many
classifications from the code and drastically changes behaviour,
AKA a mail merge from KDE is now as fast as Gtk+ again.
Change-Id: I498a73fd02d5fb6f5d7e9f742f3bce972de9b1f9
|
|
All relevant information is also provided by UpdateMinPeriod and
the calculations were even duplicated. This also includes dropping
Scheduler::UpdateMinPeriod, as this is now reduced to a simple
comparison and assignment, as we simply ignore larger returned
sleep times.
Change-Id: I13852e3e63daead451bf7fcb98be9b1d44bd7abd
|
|
Idles are just instant timers, which should most time have a low
priority, By dropping most special idle handling we'll just
schedule by priority.
This also reverts SalYieldResult back to a bool, which just
indicates if any event was processed.
Change-Id: Ia0b91b06dffb77af066f01838d8f9483523bf67d
|
|
This refactors some more scheduler code by merging the code from
ProcessTaskScheduling into CalculateMinimumTimeout and keeping the
ProcessTaskScheduling name.
It replace the bHasPendingIdles information with a HasPendingTasks
function, based on the sleep timeout and invoke time.
To drop IsIdle() we simply account Idles again and stop the system
timer, so we instantly run our application loop again. This makes
all Tasks really run immediate instead of waiting 1ms.
Change-Id: I690a8754e6b115a53e74ef777e988b66b4e5037f
|
|
Change-Id: I54534787b8cfa4c47dc09dde9c38a7893df9d367
|
