core_surface.t: test/data/icon.bmp is really 4 bits per pixel

[lighterowl]

I wasn't able to track down the exact commit in SDL that caused this to start failing, but things are pretty clear after looking at the file that this test uses, i.e. test/data/icon.bmp. Here's a dump of the file's headers :

00000000: 424d 4202 0000 0000 0000 4200 0000 2800  BMB.......B...(.
00000010: 0000 2000 0000 2000 0000 0100 0400 0000  .. ... .........
00000020: 0000 0002 0000 6d0b 0000 6d0b 0000 0300  ......m...m.....
00000030: 0000 0300 0000                           ......

According to Wikipedia, the first 14 bytes are the standard "Bitmap file header". Four-byte integer at offset 0x0E is 0x28 which means BITMAPINFOHEADER so the number of bits per pixel is stored as a two-byte integer at offset 0x1C, which is indeed 4.

[ppisar]

I noticed this change when my distribution replaced a wrapper for SDL2 with a wrapper for SDL3. Your fix is on the right track, but compatibility with SDL1.2 and SDL2 which wrongly return 8-bit surface would be ideal. Either by accepting both (4, 8) values, or detecting what SDL version is used.