Although I think that stabs have at least some kind of  usable format and the DWARF format must have been invented by a really sick mind, there's actually no alternative to DWARF2 on x64.

To be compatible to x64 binaries, we will at some place need to support 64bit SEH, which is table based rather than frame based. So we need the internal mechanism for unwinding etc and a possibility to create the unwind info in a windows compatible way.
DWARF2 should be a useful source for this. So we would need a tool that would parse DWARF2 unwind info and create proper Windows unwind info from that..

If we already create a tool that will convert DWARF2 unwind info to a windows format, we could also do it with line numbers. We could either convert them into the rsym format or CV4 or COFF. IMO we should do it like windows does.

Whatever these are just my 50 cent.

Timo


Samuel serapion schrieb: 
In my recent endeavors into porting ReactOS to the amd64 architecture I have
discovered that the the x86_64-pc-mingw32 target uses DWARF debugging
symbols by default.
The reason for this change was very simple, stabs are strictly 32bit. A
stabs record only has 32 bits to represent an address, it represents less
information, was never properly standardized, so gdb is pretty much the only
debugger which can handle gcc stabs in their full glory, and it is larger to
boot. DWARF is better in every aspect. There is no reason to use stabs these
days.


encoded         � does i386-mingw32 support DWARF2 ?
NightStrike     � the plan is that gcc4.3 will
NightStrike     � more specifically, it will also drop support for sjlj

Dwarf 2 Unwind frames for mingw32/cygwin
patch<http://gcc.gnu.org/ml/gcc-patches/2006-12/msg00133.html>

mingw32-gcc4.3 packages are so patched afaics:
http://sourceforge.net/project/showfiles.php?group_id=2435&package_id=241304


Right now ReactOS has a tool called rsym that parses stabs and coff symbol
data.

arty            � rsym was made to keep compact line number info


Ask yourself:

Could we eliminate an entire build step that is imposed on all ros binaries?
How hard would it be to teach the kernel about DWARF? (If we really wanted
to)


sedwards        � wine dbghelp supports pdb too
sedwards        � CV, PDB, Stabs and Dwarf
sedwards        � as far as I know

Reading Material:

http://dwarfstd.org/dwarf-2.0.0.pdf
http://dwarfstd.org/Dwarf3.pdf
  

_______________________________________________
Ros-dev mailing list
Ros-dev@reactos.org
http://www.reactos.org/mailman/listinfo/ros-dev