Blender Git Loki

Changes to footage information panel

- Display additional information about channels
and buffer type (float/byte).
- Don't show frame number beyong sequence length.
- Also fixed issues with footage length calculation,
so it's pronbably will be needed to reload some
of existing footages.

Cycles / Filter functions:
* Avoid some variable castings here to save some calculations.

Cycles:
* Remove unused film_response table code.

Cycles / OSL:
* Remove custom template from r52885, not needed anymore as OSL 1.3 is a requirement and we do not use a custom branch anymore.

Cycles OpenCL: a few fixes to get things compiling after kernel changes,
for Apple OpenCL on OS X 10.8 and simple AO render.

Also environment variable CYCLES_OPENCL_TEST can now be set to CPU, GPU,
ACCELERATOR, DEFAULT or ALL values to test particuler devices.

Cycles bump node: change the Strength value to work better, previously it would
give results that were either too weak or too strong, this makes it give more
predictable results. The downside is that it breaks backwards compatibility but
the previous behavior was almost broken.

fix [#35280] blender crashes when setting Environment Map imagefile

use standard name for operator properties - 'props'.

bmesh: optimize bmesh_vert_separate, redice allocs (best cast it wont do any allocs).
gives approx 16% overall speedup to edgesplit modifier.

also reduce size of smallhash stack, was 521, which got doubled and was quite large on the stack. reduce to 64.

Upgrade for the add_search node operator. This now uses the same basic system as the regular add_node operator, with enum items generated from the common node categories system (nodeitems_utils module). This means that any node listed in the regular node Add menu can now also be added via searching, including node groups and the like. The search operator also uses the subsequent transform to make insertion a bit more streamlined.

can't use alloca in inline functions (fills up stack - I thought compiler would be smart here).

also reserve the exact number of vert/face/edge/loops when creating a bmesh during undo.

bmesh: avoid using BLI_array macros for every face when converting to bmesh, replace with alloca

bmesh speedup: skip free-realloc while running CustomData_bmesh_merge() when nothing is changed (happens quite often that there is nothing to do).

Fix for add_node operator: needs to have SKIP_SAVE set for the settings collection to prevent it from re-using settings from a previous call. This would otherwise lead to python exceptions when those properties don't exist for the other node type (point in case: node_tree only is defined for group nodes).

Partial revert of own commits r56604 and r56603:
* Reverted the changes to code comments, as suggested by Campbell. It makes it more hard to follow.
* Only keep changes to actual UI messages.

Fix for do_versions bug with node groups: The special case of direct input-to-output connections was not handled correctly. In this case both the tonode and fromnode pointers in old node groups are NULL.

Fix suggested by Campbell Barton: use %r instead of %s when building the python expression for the node operator settings to ensure correct escaping.

fix for 2 errors introduced since release
- shape key NULL pointer dereference.
- use uninitialized variable for bmesh free.

also update credits and merge dissolve flag assignment.

fix rna_info, python method to C function wasn't being tested for. (broke changelog generator)

update autogenerated changelog. also correct typos in the generator.