AtlasImageImporter::populateChunks() bug

Welcome to Tom's bug of the day!


Today's bug is in AtlasImageImporter::populateChunks()...

if(mSource->getNumberChannels() == 0)
         {
            atc->mFormat = AtlasTexChunk::FormatJPEG;
            atc->layerCount = 1;
 
            // Construct the bitmap for this tile.
            GBitmap *gb = new GBitmap[1];
            gb->allocateBitmap(tileSize, tileSize);
 
            ColorI col;
 
            for(S32 leafX=0; leafX<tileSize; leafX++)
               for(S32 leafY=0; leafY<tileSize; leafY++)
               {
                  mSource->getColor(leafX + x*tileSize, leafY + y*tileSize, col.red, col.green, col.blue, col.alpha);
                  gb->setColor(leafY, leafX, col);
               }
 
            atc->bitmap = gb;
         }
         else
         {
            // It's a multilayer image, so set things up a bit differently
            atc->mFormat = AtlasTexChunk::FormatPNG;
 
            const S32 channelCount = mSource->getNumberChannels();
            atc->layerCount = (channelCount + 3) / 4;
 
            // Set up the chunk's bitmaps.
            atc->bitmap = new GBitmap[atc->layerCount];
            for(S32 i=0; i<atc->layerCount; i++)
               atc->bitmap[i].allocateBitmap(tileSize, tileSize, false, GFXFormatR8G8B8A8);
 
            // Ok, now sample each pixel & write out into the bitmaps.
            U8 *alphas = new U8[channelCount];
            ColorI c;
 
            for(S32 leafX=0; leafX<tileSize; leafX++)
            {
               for(S32 leafY=0; leafY<tileSize; leafY++)
               {
                  mSource->sampleChannels(leafX + x*tileSize, leafY + y*tileSize, alphas);
 
                  // Now store the alphas into their respective bitmaps.
                  for(S32 alphaStep=0; alphaStep < channelCount; alphaStep+=4)
                  {
                     c.red   = alphas[alphaStep+0];
                     c.green = (alphaStep + 1 < channelCount) ? alphas[alphaStep+1] : 0;
                     c.blue  = (alphaStep + 2 < channelCount) ? alphas[alphaStep+2] : 0;
                     c.alpha = (alphaStep + 3 < channelCount) ? alphas[alphaStep+3] : 0;
 
                     atc->bitmap[alphaStep/4].setColor(leafX, leafY, c);
                  }            
               }

Spot the bug? Of course not!

Look specifically at the innermost loops for both sides of the main if statement...

gb->setColor(leafY, leafX, col);

... and ...

atc->bitmap[alphaStep/4].setColor(leafX, leafY, c);

Notice how the first setColor() is leafY, leafX and the second is leafX, leafY. You would think that the first one is bugged and leafX, leafY is correct... and you would be very wrong!

The bug is in the PNG version of the branch and the fix is...

// Now store the alphas into their respective bitmaps.
   for(S32 alphaStep=0; alphaStep < channelCount; alphaStep+=4)
   {
      c.red   = alphas[alphaStep+0];
      c.green = (alphaStep + 1 < channelCount) ? alphas[alphaStep+1] : 0;
      c.blue  = (alphaStep + 2 < channelCount) ? alphas[alphaStep+2] : 0;
      c.alpha = (alphaStep + 3 < channelCount) ? alphas[alphaStep+3] : 0;
      
      atc->bitmap[alphaStep/4].setColor(leafY, leafX, c); // fixed!
   }

Why is this correct? You shouldn't ask questions like this... it makes Atlas angry. ;)