Mesa properly complained that S3TC isn't supported on 3D textures using
GL error (good), while AMD and NV had both their own unique data
corruption/random shuffling (bad!).
BPTC is available only on desktop, will have to wait until ASTC HDR is
more widely available.
Mesa drivers (rightfuly) complained that S3TC is not supported on 3D
textures, the packing had weird behavior on NVidia but it passed w/o
problems on AMD. Now should be okay on all three, yay!
Pre-DSA code path needs to pass specific slice of a cube map to all
getters instead of just GL_TEXTURE_CUBE_MAP. I did that properly for
image size query, which weirdly enough, had its own implementation, but
forgot to do that in compressed image getters and, because I have DSA
drivers, never tested that on pre-DSA contexts.
Using single implementation of image size with explicit target
parameter now.
Passing 0 as bufSize to glGetObjectLabel() is not allowed by the spec
even when just querying the size, thus passing the maximum. Might
hopefully fix the label queries on AMD drivers.
In the particular case of AnyImageConverter the function needs to access
manager and load/instantiate plugin using it, which is non-const
operation. More generally, it puts unnecessary restrictions on what the
plugin can and cannot do.
Provide a way to convert compressed images to data/file (i.e. saving
DXT5-compressed image data to DDS file), improve feature flags so that
the plugin can properly advertise what's supported (for example some
plugin may just be able to compress RGBA to DXT5, but not to save that to
DDS file).
This is backwards-incompatible API breakage (renamed enum value), but
because the original API wasn't in any official release yet, I'm not
doing any deprecation and backwards compatibility.
I can't believe that you can just forget to include some file and it
will SILENTLY PASS and then crashes horribly at runtime because each
translation unit thought that given member function pointer had
completely different size. THIS IS NOT ACCEPTABLE.
It worked flawlessly when crosscompiled from Linux, but compiling that
natively causes the linker to loudly complain about undefined references
to ConfigurationValue structs. I think it worked well under 4.9.2 (but
that mess had a slew of other ugly problems, such as complete inability
to produce non-crashing C++11 code under x64). The linker error is
caused only because I tried to reduce binary bloat, so I'm just
disabling that for MinGW and screw that. No problem under MSVC.