I made a mistake when I updated tex_dds.cpp in rP23305.
The previous code had an unreachable line 464. Indeed at line 436 DDPF_ALPHAPIXELS (without DDPF_RGB excluded from the previous conditional) was used to determine whether the file was uncompressed greyscale. Thus the condition at line 463 could never be true.
As a consequence, before the upgrade, no DXT1a file would have been properly decoded: they would all be mistaken for a 8bpp greyscale, and would throw an error line 445 (DXT1a is 4bpp (actually dwRGBBitCount is not even guaranteed to be valid with compressed formats)).
This never happened since rP14015 so we probably never use DXT1a and do not need to support it.
When I updated our code to take into account the DDPF_ALPHA flag which describes uncompressed 8bpp greyscale, I kept the logic of that section intact without noticing that some code was unreachable. That let me to misunderstand the meaning of DDPF_ALPHA and to write a wrong comment.
As a consequence, right now, DXT1a files are still not supported, however loading them will fail silently: they will be mistaken for DXT1 files and their 1bpp alpha will be ignored.
This patch fixes the logic and allows us to properly decode DXT1a.