Greyscale to RGBA texture conversion fails silently (failing Atlas heightmap import)

[elexis]

After importing an image as a heightmap to atlas, the vertex with the highest coordinates is wrong.

For example if you load an entirely black image, you get a flat surface with one pixel at the corner up in the sky.

(See #5021 for the wrong border, this ticket is only about that one corner)

[elexis]

When loading a greyscale png, the code of r12308 first tries to convert it to RGBA. But that silently fails. tex.m_Flags consists of only TEX_GREY and TEX_TOP_DOWN after the tex.transform_to. Then the following heightmap import code still reads 3 channels per pixel, which results in an out of bounds read for the very last pixel.

So it's something wrong with the texture conversion code, not with the atlas heightmap import code apparently.

[Stan]

Does that code handle the other texture loading because I know some of our AO and spec textures use grayscale to save size.

[elexis]

yes

[elexis]

Workaround in r25843, duplicate at #6261

[Vladislav Belov]

The bug is inside tex.cpp, the code fails to an add alpha channel to a grayscale image. Even in case it's going to be converted to a color image:

if(transforms & TEX_ALPHA)
{
    // add alpha channel
    if(bpp == 24)
    {
        dstSize = (srcSize / 3) * 4;
        t->m_Bpp = 32;
    }
    // remove alpha channel
    else if(bpp == 32)
    {
        return INFO::TEX_CODEC_CANNOT_HANDLE;
    }
    // can't have alpha with grayscale
    else
    {
        return INFO::TEX_CODEC_CANNOT_HANDLE;
    }
}

[wraitii]

(would argue it's not necessarily a bug, but would probably be nice to implement. That said, the ticket is named incorrectly, as the failure is very much not silent.)

[Vladislav Belov]

Replying to wraitii:

(would argue it's not necessarily a bug, but would probably be nice to implement. That said, the ticket is named incorrectly, as the failure is very much not silent.)

Nope, it fails silently. So it's the bug.