Signed-off-by: Gabriel Ivăncescu <gabrielopcode@gmail.com>

Signed-off-by: Gabriel Ivăncescu <gabrielopcode@gmail.com>

Signed-off-by: Gabriel Ivăncescu <gabrielopcode@gmail.com>

This is in the same vein as e106bbdd
and further fallout from 2ddb6b66.

It can be unnecessary at best and unsupported at worst (e.g. no
ARB_texture_multisample or MultisampleTextures setting disabled).

To wined3d_context_gl_apply_fbo_state_explicit(). It's not really
related to blitting in principle; what it does is attaching specific
textures to the FBO instead of the d3d render targets, which was the
"original" use of FBOs in wined3d.

BTW even that original use case (currently handled by
context_state_fb()) is not using render_targets[] directly anymore and
we end up kind of abusing the blit_targets[] arrays in struct
wined3d_context_gl. Maybe we could rename that array as well.

All the callers (i.e. the blitters) also call
context_gl_apply_texture_draw_state() that does that part of the setup
anyway.

It really isn't supposed to be its responsibility. All the callers of
the function (which are blitters) take care of it already.

Port of c0ab5570 to the ARB FP blitter.

Port of c0ab5570 to the FFP blitter.

None of the GL states set by wined3d_context_gl_apply_blit_state()
should matter for glReadPixels(), aside from the FBO read binding. So
just do that instead.

Some wined3d git archaeology suggests that the
wined3d_context_gl_apply_blit_state() call was added right before Wine
1.2 to workaround various driver issues with glReadPixels() that in
practice was erroneously affected by some GL states. If those kind of
issues are still a thing, it might be necessary to reintroduce some
limited state reset, possibly tied to a quirk.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54588

As shown by the testbot, doubling is not always sufficient.

It fixes EMRPOLYDRAW16 printing when -O2 option is used (in GCC 12.3.1). The
compiler assumes that apts table has 0 or 1 elements and optimizes the loop
out.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55486

We can't do much without unicode mapping.

With recent changes leave_handler writes to some of its parameter's
fields (via DS_sig and ES_sig) on FreeBSD and NetBSD - which fails
since it is declared const. Accordingly strip const-ness.

They are not used for anything anymore and just impose limit of 10
devices of a given type (capturer / renderer) without doing bound
checking.