[Bug 36348] New: valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

Bug ID: 36348
           Summary: valgrind shows a couple leaks in d3d9/tests/d3d9ex.c
           Product: Wine
           Version: 1.7.17
          Hardware: x86
                OS: Linux
            Status: NEW
          Keywords: download, source, testcase
          Severity: normal
          Priority: P2
         Component: directx-d3d
          Assignee: [e-mail]
          Reporter: [e-mail]

==3976== 16 bytes in 1 blocks are definitely lost in loss record 396 of 1,973
==3976==    at 0x7BC50A9B: RtlAllocateHeap (heap.c:255)
==3976==    by 0x4F58C13: Direct3DCreate9Ex (d3d9_main.c:63)
==3976==    by 0x4A557EB: create_device.constprop.1 (d3d9ex.c:59)
==3976==    by 0x4A55992: test_reset (d3d9ex.c:684)
==3976==    by 0x4A5A5CB: func_d3d9ex (d3d9ex.c:1218)
==3976==    by 0x4ADCD4C: run_test (test.h:584)
==3976==    by 0x4ADD13B: main (test.h:654)
==3976==

==3976== 12,068 bytes in 1 blocks are definitely lost in loss record 1,915 of
1,973
==3976==    at 0x7BC50A9B: RtlAllocateHeap (heap.c:255)
==3976==    by 0x506F901: wined3d_create (wined3d_main.c:95)
==3976==    by 0x4F654AE: d3d9_init (directx.c:647)
==3976==    by 0x4F58C30: Direct3DCreate9Ex (d3d9_main.c:66)
==3976==    by 0x4A557EB: create_device.constprop.1 (d3d9ex.c:59)
==3976==    by 0x4A55992: test_reset (d3d9ex.c:684)
==3976==    by 0x4A5A5CB: func_d3d9ex (d3d9ex.c:1218)
==3976==    by 0x4ADCD4C: run_test (test.h:584)
==3976==    by 0x4ADD13B: main (test.h:654)
==3976==

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |valgrind

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

--- Comment #1 from Austin English  ---
Also:
../../../tools/runtest -q -P wine -T ../../.. -M d3d9.dll -p d3d9_test.exe.so
d3d9ex && touch d3d9ex.ok
==6443== 16 bytes in 1 blocks are definitely lost in loss record 150 of 936
==6443==    at 0x7BC49E0E: notify_alloc (heap.c:254)
==6443==    by 0x7BC4E044: RtlAllocateHeap (heap.c:1715)
==6443==    by 0x4947EB9: Direct3DCreate9Ex (d3d9_main.c:63)
==6443==    by 0x4877696: test_wndproc (d3d9ex.c:2114)
==6443==    by 0x487B247: func_d3d9ex (d3d9ex.c:2929)
==6443==    by 0x4907463: run_test (test.h:584)
==6443==    by 0x490789D: main (test.h:666)
==6443==

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

--- Comment #2 from Henri Verbeet  ---
Created attachment 51301
  --> https://bugs.winehq.org/attachment.cgi?id=51301
patch

I think the attached patch should help. I don't see how the leaks from comment
0 happen, as far as I can see here the allocations are properly free'd again.

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

--- Comment #3 from Austin English  ---
(In reply to Henri Verbeet from comment #2)
> Created attachment 51301 [details]
> patch
>
> I think the attached patch should help. I don't see how the leaks from
> comment 0 happen, as far as I can see here the allocations are properly
> free'd again.

The leaks from comment #0 aren't occurring in 1.7.41. The patch does help,
though a new issue appears after that:
==31771== Conditional jump or move depends on uninitialised value(s)
==31771==    at 0x4E9551A: normalize_nonclientmetrics (sysparams.c:635)
==31771==    by 0x4E97938: SystemParametersInfoW (sysparams.c:1583)
==31771==    by 0x4E9A583: GetSystemMetrics (sysparams.c:2414)
==31771==    by 0x4E9A6FB: GetSystemMetrics (sysparams.c:2433)
==31771==    by 0x4EAE481: WINPOS_GetMinMaxInfo (winpos.c:774)
==31771==    by 0x4EA7291: WIN_CreateWindowEx (win.c:1565)
==31771==    by 0x4EA7A8D: CreateWindowExW (win.c:1748)
==31771==    by 0x521FE3D: grab_clipping_window (mouse.c:372)
==31771==    by 0x5220590: clip_fullscreen_window (mouse.c:520)
==31771==    by 0x520C6E1: update_windows_on_desktop_resize (desktop.c:210)
==31771==    by 0x4EAAF39: EnumWindows (win.c:3291)
==31771==    by 0x520CAB8: X11DRV_resize_desktop (desktop.c:285)
==31771==    by 0x524759C: xrandr12_set_current_mode (xrandr.c:362)
==31771==    by 0x5233031: X11DRV_ChangeDisplaySettingsEx (settings.c:420)
==31771==    by 0x4E9B40E: ChangeDisplaySettingsExW (sysparams.c:2795)
==31771==    by 0x4B6A790: wined3d_set_adapter_display_mode (directx.c:3965)
==31771==    by 0x4BF4C92: wined3d_swapchain_activate (swapchain.c:1207)
==31771==    by 0x4B5D3E2: device_process_message (device.c:5035)
==31771==    by 0x4C0DECF: wined3d_wndproc (wined3d_main.c:404)
==31771==    by 0x4EB44C9: ??? (winproc.c:174)
==31771==  Uninitialised value was created by a stack allocation
==31771==    at 0x4E9548B: normalize_nonclientmetrics (sysparams.c:626)
==31771==

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

--- Comment #4 from Henri Verbeet  ---
(In reply to Austin English from comment #3)
> The leaks from comment #0 aren't occurring in 1.7.41. The patch does help,
> though a new issue appears after that:

Does that happen consistently? (As opposed to being some kind of race.) It's
slightly odd that this patch would make that issue show up, although it looks
like a user32 thing to me either way.

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

--- Comment #5 from Austin English  ---
(In reply to Henri Verbeet from comment #4)
> (In reply to Austin English from comment #3)
> > The leaks from comment #0 aren't occurring in 1.7.41. The patch does help,
> > though a new issue appears after that:
>
> Does that happen consistently? (As opposed to being some kind of race.) It's
> slightly odd that this patch would make that issue show up, although it
> looks like a user32 thing to me either way.

Not consistent, though I have been able to reproduce it. After
https://source.winehq.org/git/wine.git/commitdiff/ca00d72da3d959df1e6a6d619e...,
running d3d9ex.c is causing the display to change resolution and it fails to
return to its original size. xrandr -s was useful originally, but later
failures required a restart of X.

When the user32 failure occurs, the resolution is very wrong.

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

joaopa  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #6 from joaopa  ---
What about this bug with current wine(3.20)?

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #7 from Austin English  ---
(In reply to joaopa from comment #6)
> What about this bug with current wine(3.20)?

Let's call it fixed.

[Bug 36348] valgrind shows a couple leaks in d3d9/tests/d3d9ex.c

https://bugs.winehq.org/show_bug.cgi?id=36348

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #8 from Alexandre Julliard  ---
Closing bugs fixed in 4.2.