Fixed issue 580

[1bsyl]

This should fixed the issue #580

• Don't use SDL_surface 'refcount', but 'userdata'
• Use imgref and not imgref + 1

[SheridanR]

Looks good to me, well done.

[perkinslr]

Commit ce9b618 breaks the minimap (corrupted texture, eventual segfault) if I apply this patch to the 3.3.4 branch (last version available via Humble). If it doesn't break later versions, that's fine, this comment will hopefully help anyone else in my position. But it's also probably worth double-checking that it doesn't corrupt the minimap in the current master and 3.3.7 branch (just start a new game, if there's a problem, you'll get "static" on the bottom half of the minimap).

Without this commit, I haven't noticed any other issues, so I'm not sure what problem it is supposed to solve. That said, the rest of this PR works as intended (and thanks for fixing the broken textures!)

[1bsyl]

@perkinslr
about ce9b618

You can test the issue with latest (head) SDL_ttf, or simulate it by creating an intermediate surface with pitch != width

Though, the patch may be wrong because:

• GL_UNPACK_ROW_LENGTH is by default in OpenGL whereas it's a only an extension for OpenGL ES2.
• this was assumed that bytes per pixel is 4 ...

To solve some potential issue, you can also manually pack the surface, if needed.
like:
https://github.com/libsdl-org/SDL/blob/main/src/render/opengles2/SDL_render_gles2.c#L1503