More robust mipmap fix

korman/exporter/material.py

@@ -272,13 +272,20 @@ class MaterialConverter:
             compression = plBitmap.kDirectXCompression
             dxt = plBitmap.kDXT5 if key.use_alpha or key.calc_alpha else plBitmap.kDXT1
 
-            # HACK ATTACK!
+            # Major Workaround Ahoy
             # There is a bug in Cyan's level size algorithm that causes it to not allocate enough memory
-            # for the color block in a 1x1 DXT5 mipmap. To fix that, we will reduce the number of levels
-            # by one, such that we don't even produce a 1x1 mipmap. DXT images need to have at least 4 pixels anyway...
-            # We can't fix the real source of the bug because the values generated by the algo are
-            # used to read from the file. Can't fixed close source games :(
-            numLevels -= 1
+            # for the color block in certain mipmaps. I personally have encountered an access violation on
+            # 1x1 DXT5 mip levels -- the code only allocates an alpha block and not a color block. Paradox
+            # reports that if any dimension is smaller than 4px in a mip level, OpenGL doesn't like Cyan generated
+            # data. So, we're going to lop off the last two mip levels, which should be 1px and 2px as the smallest.
+            # This bug is basically unfixable without crazy hacks because of the way Plasma reads in texture data.
+            #     "<Deledrius> I feel like any texture at a 1x1 level is essentially academic.  I mean, JPEG/DXT
+            #                  doesn't even compress that, and what is it?  Just the average color of the whole
+            #                  texture in a single pixel?"
+            # :)
+            if key.mipmap:
+                # If your mipmap only has 2 levels (or less), then you deserve to phail...
+                numLevels = max(numLevels - 2, 2)
 
             # Grab the image data from OpenGL and stuff it into the plBitmap
             with _GLTexture(image) as glimage: