Starting with the following .blend file on the command line hangs with a black non-full-screen window, but starting without a filename and then opening this same file works.
`../build_linux/bi…
Fixed now. Bit of a panic there. Any opinions on whether the math library is wrong?
It looks like int2 / float does not produce a float2, it might be producing an int2. This might be considered a bug in the math library?
Updated with new version to address comments.
It would be nice if all the functions were consistent about whether minx/y are subtracted from the index!
Futher investigation made me not use math::interpolate_nearest_fl as I am worried about it making a different decision about 0.5 than the clipping code. Uses the old function which does Extend…
b9d722d422
Fix #124842: Fix clipping and Nearest sampling in COM_MemoryBuffer::read/read_elem_sampled
Existing issue. Bicubic gets the other pixel samples for the filter using Extend mode and does not implement Repeat.
The border condition is slightly different for Nearest than for the antialiased version. However it could calculate this 'm' value for both and use a different threshold.
I pushed a new version that skips the steps you already applied and removes the bouding box changes for Scale.
d3090f6e14
Fix #124842: Fix clipping and Nearest sampling in COM_MemoryBuffer::read/read_elem_sampled
Actually now I'm thinking of removing this. The pixels are ending up in the right place, it's just that the bounding box is somewhat wrong (which also means that a second scale to a fixed size…
Probably a good idea to force the size, my version could move one edge when not moving the other, which I think is what produced your errors. Any particular scale or input size that you observed…
You are probably right that alternating shift does not need to be put in unless the current scheme actually demonstrates a problem. Also I think all this will have to be changed to use `displayWind…