[Bug 12246] New: "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Summary: "make test" gives different results with and without
                    warn+heap
           Product: Wine
           Version: CVS/GIT
          Platform: Other
        OS/Version: other
            Status: NEW
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

The same sets of tests should pass, fail, hang, and crash regardless
of whether WINEDEBUG=warn+heap is set.
(Ideally they'd all pass, but that's orthogonal to this bug.)
Yet at the moment, this is not the case.  
To repeat:
 rm -rf .wine
 cd wine-git
 sh winetricks gecko
 make testclean
 make -k test > test-normal.log 2>&1 
 egrep '^make|Backtrace' test-normal.log > sum-normal.log
 make testclean
 WINEDEBUG=warn+heap make -k test > test-heap.log 2>&1 
 egrep '^make|Backtrace' test-heap.log > sum-heap.log
 diff -u sum-normal.log sum-heap.log 
The diff should not output anything, but at the moment it outputs

make[2]: Entering directory `/home/dank/wine-git/dlls/d3d9/tests'
+Backtrace:
+Backtrace:
+make[2]: *** [stateblock.ok] Error 5
+make[2]: Target `test' not remade because of errors.
 make[2]: Leaving directory `/home/dank/wine-git/dlls/d3d9/tests'
+make[1]: *** [d3d9/tests/__test__] Error 2
 make[2]: Entering directory `/home/dank/wine-git/dlls/d3drm/tests'
...
 make[2]: Entering directory `/home/dank/wine-git/dlls/gdi32/tests'
 make[2]: *** [bitmap.ok] Error 2
-make[2]: *** [font.ok] Error 2
...
 make[2]: Entering directory `/home/dank/wine-git/dlls/mshtml/tests'
+Backtrace:
+Backtrace:

and it hangs once (in mshtml/script).

In other words, something's fishy with tests d3d9/stateblock,
gdi32/font, and mshtml/script.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Dan Kegel  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |1.0.0

--- Comment #1 from Dan Kegel   2008-03-27 00:29:45 ---
I wouldn't be surprised if the d3d one depended
on graphics card, so ymmv.  But fixing the common
warn+heap induced problems seems like a must-do.
Nominating for 1.0.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Lionel Debroux  changed:

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

--- Comment #2 from Lionel Debroux   2008-03-27 02:35:50 ---
d3d9/stateblock has been failing here as well for three weeks. I reported the
problem as part of bug #9916. I have a GeForce Go 7600.
mshtml/script fails on 20080325 but it didn't on 20080322.

In bug #11644, I also reported two dinput tests that have always crashed for me
on two different computers when a specific piece of hardware is plugged in and
when warn+heap is activated. I haven't run these tests without warn+heap for at
least two months, but they didn't crash without warn+heap.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #3 from Austin English   2008-03-29 03:05:34 ---
+all also gives some funky results

austin@austin-desktop:~/wine-git$ cat 4.txt
make[2]: *** [datetime.ok] Error 5
make[2]: Target `test' not remade because of errors.
make[1]: *** [comctl32/tests/__test__] Error 2
make[2]: *** [sip.ok] Error 1
make[2]: Target `test' not remade because of errors.
make[1]: *** [crypt32/tests/__test__] Error 2
make[2]: *** [stateblock.ok] Error 5
make[2]: *** [visual.ok] Error 171
make[2]: Target `test' not remade because of errors.
make[1]: *** [d3d9/tests/__test__] Error 2
make[2]: *** [cstub.ok] Error 21
make[2]: Target `test' not remade because of errors.
make[1]: *** [rpcrt4/tests/__test__] Error 2
make[2]: *** [url.ok] Error 1
make[2]: Target `test' not remade because of errors.
make[1]: *** [urlmon/tests/__test__] Error 2
make[2]: *** [class.ok] Error 15
make[2]: *** [listbox.ok] Error 68
make[2]: *** [msg.ok] Error 23
make[2]: Target `test' not remade because of errors.
make[1]: *** [user32/tests/__test__] Error 2
make[1]: Target `test' not remade because of errors.
make: *** [dlls/__test__] Error 2
make: Target `test' not remade because of errors.

I haven't retested these individually yet, however, to check for race
conditions, etc. I'll be doing that soon, but working on bug 12263's regression
test ATM.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #4 from Lionel Debroux   2008-03-31 14:00:26 ---
I ran the tests this morning on wine-0.9.58-304-geb7f3b9.
Pretty much executable launched by make -k test barfs with an
Unhandled exception 0x000006bf
but execution of the tests (at least, most of them ?) continues OK:
err:rpc:RpcRaiseException handler continued execution
Due to the crash + backtraces + etc. and me compiling everything with -O0 to
save on compilation time, the whole test suite took at least one hour to
complete...

I've just updated my local git repository after today's commits, and I'm
currently compiling.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #5 from James Hawkins   2008-03-31 14:08:59 ---
Lionel, keep in mind that some of the tests are supposed to cause a crash.  I'm
not sure of the specific instance you're referring to though.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #6 from Lionel Debroux   2008-03-31 14:50:36 ---
Created an attachment (id=11770)
 --> (http://bugs.winehq.org/attachment.cgi?id=11770)
Beginning of trace of WINEDEBUG=warn+heap make -k test with
wine-0.9.58-358-g41bdea4

Indeed, I know that the kernel32 debugger test is supposed to crash ;)
However, since yesterday, if I activate warn+heap, I'm now seeing _every_ test
crash (before ignoring the unhandled exception anyway, it seems), even after a
rm -Rf ~/.wine !
That didn't happen on 20080325, when I last ran a WINEDEBUG=warn+heap make
test.

Judging from the backtraces, the problem is with the service starting code,
which has been changed very recently.

I've attached the trace of the first test, before me killing make with CTRL+C.
This one is after starting with no ~/.wine directory.
(FWIW, the whole Wine test suite with all crashes generates a ~25K-line trace)

Does this happen only on my computer ?

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #7 from James Hawkins   2008-03-31 14:56:50 ---
hmm looks like a problem with the recent creation of the services.exe module
and moving all the service control code over to that from advapi32.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #8 from Dan Kegel   2008-03-31 16:33:50 ---
Yeah, pretty awful regression today in warn+heap, everything crashes
just like your log says.  Oh Rob....

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #9 from Dan Kegel   2008-03-31 22:03:34 ---
The regression was already there on the 29th, I just
didn't look at the results of the valgrind run that day :-(
http://kegel.com/wine/valgrind/logs-2008-03-29/vg-advapi32_cred-diff.txt
...
http://kegel.com/wine/valgrind/logs-2008-03-29/vg-cabinet_fdi-diff.txt
...
http://kegel.com/wine/valgrind/logs-2008-03-29/vg-gdiplus_brush-diff.txt
etc etc.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #10 from Rob Shearman   2008-04-01 03:16:33 ---
Created an attachment (id=11780)
 --> (http://bugs.winehq.org/attachment.cgi?id=11780)
Fix Valgrind and warn+heap errors caused by recent svcctl patches

Silence the warnings.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #11 from Lionel Debroux   2008-04-01 03:49:24 ---
April fool patch, eh ? ;)

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Rob Shearman  changed:

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

--- Comment #12 from Rob Shearman   2008-04-01 10:07:13 ---
Yep ;-)

I've sent a real patch to wine-patches:
http://www.winehq.org/pipermail/wine-patches/2008-April/052567.html

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Rob Shearman  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #11780|0                           |1
        is obsolete|                            |

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #13 from Austin English   2008-04-16 16:45:21 ---
Also seeing this in shell32/shlexec and winmm/wave.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #14 from Dan Kegel   2008-04-24 21:47:13 ---
There seems to be some progress.  Using the attached script,
I find only one difference between make test with and without
warn+heap: d3d9/stateblock.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #15 from Dan Kegel   2008-04-24 21:48:11 ---
Created an attachment (id=12451)
 --> (http://bugs.winehq.org/attachment.cgi?id=12451)
Script to run tests both with and without warn+heap.  Cleans .wine between runs
now.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #16 from Dan Kegel   2008-05-12 12:27:07 ---
d3d9/stateblock is still the only test that gives
different results here depending on whether it was run
with warn+heap.

valgrind shows only a few errors on that file; see
http://kegel.com/wine/valgrind/logs-2008-04-17/vg-d3d9_stateblock.txt
Line numbers have changed since then; I'll do a fresh run and add a link here.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #17 from Dan Kegel   2008-05-13 14:32:25 ---
Fresh line numbers right here:
http://kegel.com/wine/valgrind/logs-2008-05-12/vg-d3d9_stateblock.txt

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Dan Kegel  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|-unknown                    |directx-d3d

--- Comment #18 from Dan Kegel   2008-05-13 16:55:12 ---
Oooh, and running it on a different machine gave a juicy crash rather than
a failure:

Unhandled exception: page fault on write access to 0xaaaaaaaa in 32-bit code
(0x601e822e).
Backtrace:
=>1 0x601e822e (0x0032f8d8)
  2 0x612c2ad3 IWineD3DDeviceImpl_ResourceReleased+0x509(iface=?,
resource=0x154518) [dlls/wined3d/device.c:7408]
  3 0x612f86b4 IWineD3DResourceImpl_CleanUp+0x109(iface=?)
[dlls/wined3d/resource.c:90]
  4 0x61321b42 IWineD3DSurfaceImpl_Release+0x13a(iface=0x154518)
[dlls/wined3d/surface.c:467] 
  5 0x612c1d76 IWineD3DDeviceImpl_SetRenderTarget+0x97(iface=0x12b900,
RenderTargetIndex=0x0, pRenderTarget=?) [dlls/wined3d/device.c:6713]
  6 0x6126a916 IDirect3DDevice9Impl_SetRenderTarget+0x58(iface=?,
RenderTargetIndex=?, pRenderTarget=?) [dlls/d3d9/device.c:685]
  7 0x60cf9196 revert_render_target+0x34(device=?, data=0x32fe64)
[dlls/d3d9/tests/stateblock.c:313]
  8 0x60cf8b6d execute_test_chain+0x5d(device=0x129f38, test=0x32fc50,
ntests=0x5, event=0x32fe04, nevents=0x3, event_arg=0x32fe64)
[dlls/d3d9/tests/stateblock.c:180] 
  9 0x60cfacdb func_stateblock+0x735() [dlls/d3d9/tests/stateblock.c:469]

This agrees fairly closely with the valgrind results on another machine
(with wine from git only a few hours out of sync, if any):

Invalid read of size 4
    at  ActivateContext (context.c:962)
    by  IWineD3DDeviceImpl_ResourceReleased (device.c:7408)
    by  IWineD3DResourceImpl_CleanUp (resource.c:90)
    by  IWineD3DSurfaceImpl_Release (surface.c:465)
    by  IWineD3DDeviceImpl_SetRenderTarget (device.c:6711)
    by  IDirect3DDevice9Impl_SetRenderTarget (device.c:685)
    by  revert_render_target (stateblock.c:313)
    by  execute_test_chain (stateblock.c:180)
    by  func_stateblock (stateblock.c:467)

Please, some D3D guy, fix :-)

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

shakaran  changed:

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

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Michael Karcher  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]-
                   |                            |berlin.de

--- Comment #19 from Michael Karcher   2008-06-03 09:27:14 ---
My patch
 http://www.winehq.org/pipermail/wine-patches/2008-June/055516.html
got committed to git. For me it fixes the stateblock bug. This was the only
warn+heap problem remaining for me. Can this bug be closed now?

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Alexandre Julliard  changed:

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

--- Comment #20 from Alexandre Julliard   2008-06-03 09:32:20 ---
Marking fixed.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #21 from Austin English   2008-06-03 13:12:19 ---
Created an attachment (id=13639)
 --> (http://bugs.winehq.org/attachment.cgi?id=13639)
wtest-out

I still get a few differences in user32.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Austin English  changed:

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

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

--- Comment #22 from Alexandre Julliard   2008-06-03 13:33:15 ---
Most likely timing issues, warn+heap makes things slower.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Alexandre Julliard  changed:

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

--- Comment #23 from Alexandre Julliard   2008-06-06 10:39:56 ---
Closing bugs fixed in 1.0-rc4.

[Bug 12246] "make test" gives different results with and without warn+heap

http://bugs.winehq.org/show_bug.cgi?id=12246

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|CVS/GIT                     |unspecified