[llvm-commits] [patch] Add an option for not using dwarf's .loc directive

[Devang Patel]

On Nov 18, 2010, at 10:59 AM, Rafael Espíndola wrote:

> Hello,
> 
> The attached patches are incomplete, but they should be sufficient to
> fix the dawin 9 bots and better explain what I am doing. The idea is
> that clang will continue to use .loc, but llvm-gcc will go back to
> using a manually crafted .debug_line section.
> 
> Another option is to do it by target, but I assume you guys also want
> the benefits of .loc on darwin when possible.
> 
> The patch is incomplete in that it has only the plumbing, but is
> enough to disable .loc on llvm-gcc. It is also flexible if we want to
> enable it in llvm-gcc for some targets but not others in the future.

> The missing bit that I am working on right now is refactoring the MC
> code a bit so that it can print the debug_line info a regular
> streamer, not just an object streamer. If using an asm streamer it
> will have the same limitation as the old code in CodeGen (using
> DW_LNE_set_address), but at least there will not be as much code
> duplication.
> 
> Are these first 3 patches OK?

If you make a choice between clang vs. llvm-gcc then we will trade llvm-gcc-powerpc-darwin9 buildbot failures with llvm-gcc-x86-darwin10-gdb-testsuite build bot failures! Obviously, I care more about darwin10-gdb-testsuite results.

I suggest, for now while you make progress, you can entirely disable debug info for powerpc by updating following line in PPCMCAsmInfo.cpp
	 SupportsDebugInformation= true; // Debug information.
 This will make all buildbots green (I do not know of any powerpc buildbot focusing on debug info).

-
Devang

> 
> 
> Cheers,
> Rafael
> <llvm.patch><clang.patch><llvm-gcc.patch>