I incorporated the "proper" solution to
PDFBOX-1075 into my regression tests and the file which made me raise that issue got broken again. It has pictures with indexed color spaces, which now are returned as black. The indexed color space is one-bit. The lookup table has two colors: black and white. With the current trunk, the black pixels remain black (0,0,0), but the white pixels are returned as (1,1,1), which in RGB is nearly just as black. The text on the picture is obviously unreadable.
On a second look it seems to me that the offending line is here (in PDPixelMap.getRGBImage):
I think it's wrong, as in an indexed color space, "bpc" is not "bits per component", but "bits per index" i.e. bits per the integer which is interpreted as the index to the color lookup array. I think that the base color space's color model should be initialized with a different number. I came up with a following calculation:
The baseColorModelBPC is calculated as the total number of bits in the lookup array divided by the total number of components in all colors. This seems to work for my offending file and causes no regressions with files from
PDFBOX-1075, PDFBOX-1010 and PDFBOX-706.
What is weird though is a line which is later:
Doesn't this effectively assume that baseColorModelBPC should always be 8? If the base color model bpc were anything else than 8, then this code wouldn't be able to handle it correctly anyway. Or am I overlooking something?
I'll attach a patch which works for me. Note that simply changing
has exactly the same effect (fixed my problem, no regressions in those three earlier issues). Please decide what makes more sense.