Meshimage could be made to work properly for any transformation, if we created a bad mockup of a triangle rasterizer. Alternatively we know that the meshimage must be drawn per pixel, so could pseudo-rasterize a map of pixel positions to original positions.
This would probably only be for (a) PNG output or (b) if rasterize=true. SVG and PDF could either use the rasterization approach with 3x resolution, or just use the direct mesh as they do now. That's up for debate!
Meshimage could be made to work properly for any transformation, if we created a bad mockup of a triangle rasterizer. Alternatively we know that the meshimage must be drawn per pixel, so could pseudo-rasterize a map of pixel positions to original positions.
This would probably only be for (a) PNG output or (b) if rasterize=true. SVG and PDF could either use the rasterization approach with 3x resolution, or just use the direct mesh as they do now. That's up for debate!