[Bug 53629] New: mixed gdi/dx9 rendering leads to a failure to render

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53629

Bug ID: 53629
           Summary: mixed gdi/dx9 rendering leads to a failure to render
           Product: Wine
           Version: 7.15
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: d3d
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

The summary is a guess and probably a wrong one.

wine version is set to 7.15, but it was first observed around 7th Oct 2020.

Anyway, in two games of a single dev (not really an engine, unless in the
home-cooked sense) the games start, show a standard window then fail to render
anything more. Mouse cursor still disappears upon entering the window and
reappears upon leaving.

One is steamid 1289960, other currently pending (there are demos available for
the original, non-steam versions).

Initially I thought it was something like the Diablo problem, even though ddraw
isn't involved, yet there was something about child windows in 'win' channel.

Then I've ran apitrace.

After first Present call, a Reset call is made repeatedly and each time it
fails.

Also of note is the fact that a native d3dx9_31 is required, otherwise a crash
with something looking like a stack corruption ('Unhandled page fault on read
access to CCCCCCCC at address ...') happens immediately.

Potentially of note: the hack makes things work with gl renderer, vulkan is
still broken, though that might be due to the child window bug; gdi renderer
makes things crash immediately with or without the hack.

If reset_enum_callback (in dlls/d3d9/device.c) is made to return D3D_OK in both
Texture and Surface step, things seem to work correctly, but that looks like a
leak-inducing step.

In vulkan case, console is flooded with repeated:

<threadid>:err:d3d:adapter_vk_draw_primitive Failed to apply draw state.
<threadid>:err:d3d:wined3d_context_vk_apply_draw_state Failed to get graphics
pipeline.

[Bug 53629] mixed gdi/dx9 rendering leads to a failure to render

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53629

--- Comment #1 from Rafał Mużyło  ---
On further examination it seems the need for native d3dx9_31 is due to
vkd3d-shader being utter crap at parsing the shaders (pretty basic
vs_3_0/ps_3_0 stuff).

Upon failure to parse a NULL is returned (well, most likely one of shader's
return parameters is set to NULL) and the game is unable to deal with that.

I wonder if some of resources not being released are like that due to being
bound to shaders and those in d3d9_device_Reset should be left alone.