Its unlikely you want to do short -> int, int -> float etc, conversion during swapping (if its needed we could have a non type checking macro).
Double that the optimized assembler outbut using SWAP() remains unchanged from before.
This exposed quite a few places where redundant type conversion was going on.
Also remove curve.c's swapdata() and replace its use with swap_v3_v3()
checked this function against the previous method using random rotation inputs and compared results, while this isnt exactly the same the results are very close and acceptable in both cases, also checked baking actions that the resulting FCurves are good and give matching rotations.
now customdata is interpolated into a temp variable and applied at the end of each layer interpolation function.
So this now works for CDDM customdata interpolation and avoids duplicating the customdata when the source and destination overlap.
result. This wasn't really guaranteed anyway, because of numerical precision and
possible asymmetry in the triangulation, but we can do a bit better.
Now we bias the choice of how to split a quad into two triangles slightly
towards one of two possibilities, so that in case they are equal, floating point
errors do not decide the direction and symmetry is preserved.
Behaves in exactly the same way as bright/contrast compositor node.
Some code could be de-duplicated, like contrast formula and mask
influence, but wouldn't call it stopper for commit since it's
already needed for grading Mango.
Blender original code (NaN likely) was relying on strcmp to sort the list of controllers.
As it happens, in strcmp 10 < 2, thus the list was never in the right order.
The curious thing is that in 2.49 it worked, but I think it did because the make_unique_prop_names function was called twice. Strange ;)
The solution is to use blender BLI_natstrcmp to do natural sorting
The vertex shapekey index is now no longer copied, and propagation of offsets
in the basis to other shapekeys is disabled if new vertices were added. The
reason being that the propagation will only be done for the old vertices leaving
the new ones behind, and so doing e.g. subdivide + translate on the basis would
create a mess on other shape keys.
Shader preview job localizes material without referencing it as a user,
so don't need to unreference itself as a user from this material.
Added BKE_material_free_ex function which could skip user dereferencing.
This also removed old hack with mtex users.
