@mguenther provided a set of patches that fixed GIF reading, which was completely broken (returning all zeros) so far. See discussion on !24 (merged).

This bug tracks the problem pointed on !24 (merged) that one cannot "transcode" a test image using the GIF implementation available in this package. It seems that the fact we always quantize the image as we write it makes the transcoding test fail. Possible fixes to this problem are one of:

• Modify the GIF image writing to relax on the quantization effects
• Relax the transcoding test so that we accept, in the case of GIFs, more error