Blender Git Loki

Smoke: Use better standards for inflow
a) Set particle system for inflow to " show unborn"
b) Set particle system for inflow to lifetime = 1

bmesh minor change - avoid increasing array sizes one by one and use iterator macros.

Animation version patching for Absolute Shape Keys ("speed" curve ->
"eval_time") now works

Fix [#30832] Weight on Bevel modifier has no effect for edges with weight.

Added missing CD_BWEIGHT layer for the modifier's DerivedMesh.

Windows x64 / Scons:
* Enable WITH_BF_OPENJPEG per default, it was enabled for 32 bit systems, why not on x64?
* This brings back JPEG 2000 support for win x64.

option to have inset face select the interior faces (mango request)

Ghost:
* Typo fix, preventing mingw to compile.
Found by XercesBlue in IRC:

code cleanup: add doxygen headers to bmesh operator files, also add own include so definitions dont get out of sync.

some pedantic load time speedup by not loading modules (math and subprocess in this case)

fix crash with stack memory usage for the knife tool.
knife cutting ~25000 faces would crash with 8mb stack set.

fix by removing generic iterators in functions which do recursive calls.

remove unused theme color

update python keywords (remove exec, print, add nonlocal)

code cleanup: header cleanup, remove commented workaround for mingw since its no longer needed.

Merged changes in the trunk up to revision 45431.

spherical panoramic fix
note: this was never correct and not a single soul bothered noticing that.
not really a surprise, this mode is more for debug than for anything else.
to test go to (game engine) Render -> Dome Mode -> Spherical Panorama

thanks to Aldo Zang to help me spotting the problem.

Fix windows compile error in previous commit.

Cycles / OpenCL:
* Enable __KERNEL_SHADING__ per default for OpenCL.
This enables basic shading (color, emission, textures...) for AMD cards.

You need the latest AMD catalyst driver in order to have this work.

Cycles: add rejection of inf/nan samples, in principle these should not happen
but this makes it more reliable for now.

Also add an integrator "Clamp" option, to clamp very light samples to a maximum
value. This will reduce accuracy but may help reducing noise and speed up
convergence.

DO not store Flipped flag for pose paste operator in operator redo storage,
so mixing pasteing flipped and non-flipped will work properly.

Fix #30804: nodetree.links.new() incorrectly disconnected or allowed to connect
multiple links to a socket when the arguments were passed in order:
(input socket, output socket) instead of (output socket, input socket)