Palette order is rotated when specifying palettes with the hex format

[nitro2k01]

When specifying an explicit palette using -c hex it seems the palette entries are rotated within the palette. In particular it seems if the input order is 0123 the output order is 3012.

For example, I would expect the input data: (This is test1.hex)

ffffff
b2b2b2
4c4c4c
000000

to output the following GBC format data, with white first and black last:

FF7F D65A 2925 0000

However, the actual output is this:

0000 FF7F D65A 2925

Included are two test images and three test palettes, for a total of 6 test commands. The reason for all of these tests is to test whether the order was affected by either the first color seen in the image, or the brightness of the palettes. However, it seems to consistently rotate the order as described above, so really only 3 of the test cases are relevant, as the output from the black and white tests is identical.

Command invocation:

rgbgfx -o white-test1.2bpp -p white-test1.pal -b128 -m -u -c hex:test1.hex white-bg.png
rgbgfx -o white-test2.2bpp -p white-test2.pal -b128 -m -u -c hex:test2.hex white-bg.png
rgbgfx -o white-test3.2bpp -p white-test3.pal -b128 -m -u -c hex:test3.hex white-bg.png
rgbgfx -o black-test1.2bpp -p black-test1.pal -b128 -m -u -c hex:test1.hex black-bg.png
rgbgfx -o black-test2.2bpp -p black-test2.pal -b128 -m -u -c hex:test2.hex black-bg.png
rgbgfx -o black-test3.2bpp -p black-test3.pal -b128 -m -u -c hex:test3.hex black-bg.png

Here's the input and output I observed during for the other test cases:

test2.hex

ffffff
4c4c4c
b2b2b2
000000

black-test2.pal/white-test2.pal

0000 FF7F 2925 D65A

test3.hex

000000
ffffff
4c4c4c
b2b2b2

black-test3.pal/white-test3.pal

D65A 0000 FF7F 2925

Notes:

• In all tested cases, the input order 0123 is transformed to the output order is 3012. The brightness or order the colors are seen in the image don't seem to matter.
• Specifying a palette with -c gbc preserves the order in the input file in all my tests. I have not tested other palette formats than those.
• The generated tile data has been internally consistent with the palette data in all my tests, so used as is the data would create the correct image on the screen. (But in my use case it caused other problems with code that assumed that the order was as specified.)

Test cases:

gfxtest.zip

[Rangi42]

Very thorough report and testing, thank you!

[Rangi42]

Note to self: the HEX palette parsing function is very similar to the GPL one, so double-check both of those.