Report unsupported textures (NPOT textures)

[elexis]

0 A.D. currnetly requires textures to be of the size 2^n * 2^n (POT). bubble.png was a PNG image, sized 24px * 24px and caused a re-release of Alpha 23 as it crashed the packaged OSX version when entering the gamesetup from the lobby, but reportedly no other platforms, see #5164 / r21822 / ​Phab:D1498.

So either images of arbitrary resolutions should be supported or these files should be reported by some new script.

Notice that it affects mod developers equally.

(If I recall correctly, Vladislav mentioned that most code supports it already, but only mipmap code doesn't?)

POT - Power-Of-Two

NPOT - Non-Power-Of-Two

[Stan]

According to Aeonios here: https://wildfiregames.com/forum/index.php?/topic/24372-script-to-find-non-power-of-two-textures/

NPOT doesn't work on ATI. Really it doesn't work well in general, and is not something we should encourage. Packing into an atlas does fix that though.

[Vladislav Belov]

Replying to stanislas69:

According to Aeonios here: https://wildfiregames.com/forum/index.php?/topic/24372-script-to-find-non-power-of-two-textures/

NPOT doesn't work on ATI. Really it doesn't work well in general, and is not something we should encourage. Packing into an atlas does fix that though.

It depends on video card and driver versions, also compression presenting makes sense too.

(If I recall correctly, Vladislav mentioned that most code supports it already, but only mipmap code doesn't?)

It seems so, only mipmaps have such assert.

[Stan]

It depends on video card and driver versions, also compression presenting makes sense too.

Indeed, it does, that's I guess the reason why he said it didn't work on ATI cards specifically. But I don't know if it's still relevant today ​https://www.khronos.org/opengl/wiki/NPOT_Texture

I agree though that it's a bad idea to use NPOT textures. I haven't seen a single AAA game using something else.

[elexis]

Also there are optimizations for NPOT textures if what I recall from what I heard once upon a time is correct.

[Vladislav Belov]

Replying to elexis:

Also there are optimizations for NPOT textures if what I recall from what I heard once upon a time is correct.

I suppose for POT textures.

[elexis]

We really don't want to run into that issue again that we introduce a NPOT texture and only conditionally break only OSX without noticing.

So we either need a good test/reporter to prevent this rare callstack or a more failsafe error handling (such as returning an empty texture and doing a LOGERROR).

Calling CTexture::TryLoad for the dds texture loads the function affected by the callstack. The POT test ceil_log2 from the decode_sd function should be done for all platforms.

The test could be ran either in some test_TextureManager.cpp when passing --test-texture or within ArchiveBuilder.cpp.

[Stan]

What about doing it when using the archive builder ?

Ah my bad just saw the last line of your post.