Also a little review of Dead objects in general.
The old code was calling 'erase' on a large vector, once for
every dead object (which are a lot, especially if you exit or
teleport), causing the whole (large) vector to be copied every
time. The new code only calls erase once per call (about once
a second at most now), erasing 20 to 100 objects at the END
of the vector. This is INCREDIBLY faster.
This makes sound beacons green when playing at
full volume, yellow when playing at a lower volume
and red when they are muted (aka, in another parcel
that you can't hear the sounds of).
Originally this was a debug patch as muted sound
sources used to be implemented by setting the volume
to zero, which happens to use like three times more
CPU: so, having a lot of muted sound source caused
the audio thread to never release a mutex anymore
(cause it was never idle anymore), causing the main
loop to hang, waiting on that mutex - dropping the
FPS drastically. Hence it was necessary to see which
sound sources were muted for debugging purposes.
(Since VWR-14914, muted source source are not played
at all anymore, so they do not take extra CPU).
It's still fun to see this extra information though,
now the patch exists anyway.
See http://redmine.kokuaviewer.org/issues/582
This happened only on opensim. The fixed was made
basically by MichelleZ. It doesn't harm anything, so
even though it appears to never happen on Agni
and it is unknown what bug is causing it, it make the
viewer resilient just-in-case...
It has happened to me that this hung the viewer. The cause is that
when you call parent->getParentFloater(child), then getParentFloater
which starts like this:
LLFloater* LLFloaterView::getParentFloater(LLView* viewp)
{
LLView* parentp = viewp->getParent();
starts with parentp == this. And the following logic causes
it to return 'viewp'!
I'm pretty sure that getParentFloater's whole design is wrong,
but I don't see another way to avoid this problem then with
this hack, except by making changes that do a lot more and
are a risk with regard breaking code elsewhere.
Added missing header to indra/newview/ascentkeyword.cpp now that
llviewerpluginmanager.h has a couple of header dependencies less.
Resolved Conflicts:
indra/llcommon/CMakeLists.txt
Proximity: addition of two independent things. Just included both.
It now is disabled if:
-Lacking hardware support for FBOs
-RenderDeferred not permitted in gpu feature table
-RenderAvatarVP not permitted in gpu feature table
-VertexShaderEnable is enabled and permitted in gpu feature table
-WindLightUseAtmosShaders is enabled and permitted in gpu feature table
Enabling deferred now requires WindLightUseAtmosShaders to be on beforehand, instead of blindly enabling it.
The previous hack wasn't thread-safe: read-only access would
access the reference counter multiple times at the same time,
which therefore would have to be thread-local to ever work.
The current solution just disables the calls to lock/unlock
for copyconstructed objects, which works if the copyconstructed
object isn't used anymore after the original is destructed.
This is the case then the copy construction only happens
upon passing a temporary to a function, which is the case.
Actually flush messages before terminating a plugin (upon
the shutdown message) and flush messages in the file- and
dirpicker before opening the blocking dialog. Flush debug
messages too (deeper into the code, just prior to the actual
blocking call).
Also, fix the context folder map to be a thread-safe
singleton and *attempt* to add support for default folders
to windows and Mac. The latter might even not compile yet
and definitely have to be tested (and fixed):
Opening a DirPicker in preferences --> Network and Set
the directory location of the cache. It should open a
Dialog window where you are already in the folder that
is the current cache directory setting (you can click
Cancel after verifying that this worked).
And, start to upload an image, select a file is some
directory (other than what it starts in). You can omit
the actual upload by clicking cancel in the preview.
Then upload again and now it should start in the same
folder as that you were just in. Possibly you need to
first open a file picker elsewhere with a different context
though, or windows might choose to open in the last
folder anyway while the code doesn't really work. Uploading
a sound before the second texture upload should do the
trick.
