[Bug 25920] New: winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

Summary: winedbg does not work under clang
           Product: Wine
           Version: 1.3.12
          Platform: x86
               URL: http://test.winehq.org/data/58786dc8ea237cf646e62edb86
                    5dc260684eb784/wine_ae-ub1004-clang/shdocvw:ie.html
        OS/Version: Linux
            Status: NEW
          Keywords: download, source
          Severity: normal
          Priority: P2
         Component: dbghelp
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

Created an attachment (id=33039)
 --> (http://bugs.winehq.org/attachment.cgi?id=33039)
terminal output

When wine is compiled with clang, crashes do not get a good backtrace, instead,
I get something like:

../../../tools/runtest -q -P wine -M shdocvw.dll -T ../../.. -p
shdocvw_test.exe.so ie.c && touch ie.ok
wine: Unhandled page fault on write access to 0x00000028 at address 0x68e94653
(thread 0039), starting debugger...
Unhandled exception: page fault on write access to 0x00000028 in 32-bit code
(0x68e94653).
fixme:dbghelp:elf_new_wine_thunks Duplicate in L"oleaut32<elf>":
OLEFontImpl_IPersistPropertyBag_Load.sAttrName<68f7b37c-0000000a>
sAttrName<68f7b37c-0>
fixme:dbghelp:elf_new_wine_thunks Duplicate in L"oleaut32<elf>":
VarFormatPercent.szPercentBracket<68f853b6-00000006>
szPercentBracket<68f853b6-0>
...
fixme:dbghelp:elf_new_wine_thunks Duplicate in L"shdocvw_test<elf>":
test_ConnectionPoint.dw<685dc820-00000004> dw<685dc820-0>
fixme:dbghelp:elf_new_wine_thunks Duplicate in L"shdocvw_test<elf>":
CLSID_FileProtocol<685d4260-00000010> IID_IInternetSession<685d4260-0>
winedbg: Internal crash at 0x68641f3d

full output attached.

For building Wine with Clang, see http://wiki.winehq.org/Clang

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Blocks|                            |25825

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

--- Comment #1 from Austin English  2011-02-13 04:32:57 CST ---
Backtraces seem to work at -O0, see e.g.,
http://bugs2.winehq.org/attachment.cgi?id=33271 (bug 25808).

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

Eric Pouech  changed:

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

--- Comment #2 from Eric Pouech  2011-03-19 10:57:36 CDT ---
1/ confirming crash
2/ however, clang generates bogus dwarf information:

(compilation with -O2, clang version 2.8)
----8<---
[eric:~/work/output-wine/wine-clang]$ dwarfdump dlls/kernel32/kernel32.dll.so
<<....>>
.debug_line: line number info for a single cu
dwarfdump ERROR:  dwarf_srclines:  DW_DLE_ATTR_FORM_BAD (114)
<<....>>
---------
3/ seems confirmed by some web searches (on clang 2.8)
see for example:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20110221/117178....

4/ looking at the hexdump of the section show that the .debug_line section
doesn't even contain its header

5/ I'll send a patch to protect dbghelp against bogus .debug_line sections in
dwarf files. winedbg now generates the correct backtrace (even if they are a
tons of duplicate symbols info. some of them are harmless, some others may
require some more care. that would require another bug report.

6/ I didn't test for the yet to come 2.9 version (or its RC), but definitively
a clang bug.

A+

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

--- Comment #3 from Austin English  2011-03-21 15:18:19 CDT ---
(In reply to comment #2)
> 1/ confirming crash
> 2/ however, clang generates bogus dwarf information:
>
> (compilation with -O2, clang version 2.8)
> ----8<---
> [eric:~/work/output-wine/wine-clang]$ dwarfdump dlls/kernel32/kernel32.dll.so
> <<....>>
> .debug_line: line number info for a single cu
> dwarfdump ERROR:  dwarf_srclines:  DW_DLE_ATTR_FORM_BAD (114)
> <<....>>
>
> 3/ seems confirmed by some web searches (on clang 2.8)
> see for example:
> http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20110221/117178....
>
> 4/ looking at the hexdump of the section show that the .debug_line section
> doesn't even contain its header
>
> 5/ I'll send a patch to protect dbghelp against bogus .debug_line sections in
> dwarf files. winedbg now generates the correct backtrace (even if they are a
> tons of duplicate symbols info. some of them are harmless, some others may
> require some more care. that would require another bug report.

Thanks! I filed bug 26509 for that.
> 6/ I didn't test for the yet to come 2.9 version (or its RC), but definitively
> a clang bug.

That sounds a bit like http://llvm.org/bugs/show_bug.cgi?id=9268

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

Austin English  changed:

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

--- Comment #4 from Austin English  2011-04-02 18:36:45 CDT ---
Fixed now, thanks Eric!

[Bug 25920] winedbg does not work under clang

http://bugs.winehq.org/show_bug.cgi?id=25920

Alexandre Julliard  changed:

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

--- Comment #5 from Alexandre Julliard  2011-04-15 12:50:13 CDT ---
Closing bugs fixed in 1.3.18.