I wonder if this might cause RenderDoc to show different results when investigating driver bugs and stuff like that.
Here you could avoid the move constructor by appending first and then taking a reference to it.
I think using an explicit move here can prevent the compiler from doing a copy elision?
It looks like projmat is already used in some places.
BTW, BLI_math_geom.h functions are a perfect example of the issues with this terminology.
We have "perspective" used for both…
I think I may have think that
winmatwaspersmat.
I get confused about this all the time as well. I always have to check what each one means.
I wish we had something like view_mat,…
It skips one normalization, but other than that it's the same. I've reverted the change.
I misread what the comment meant. I've added it back with a small clarification.