(X86?) Backend generates duplicate labels

[llvmbot]

Bugzilla Link	2074
Resolution	WONTFIX
Resolved on	Jan 20, 2009 16:06
Version	trunk
OS	Linux
Attachments	failing testcase
Reporter	LLVM Bugzilla Contributor
CC	@asl

Extended Description

It appears as if the label counter is not being incremented somewhere.

--
ddunbar@ozzy:src$ llvm-as -f invalid-duplicate-labels.ll
ddunbar@ozzy:src$ llvm-ld --disable-opt -native -o rt invalid-duplicate-labels.bc
rt.s: Assembler messages:
rt.s:46: Error: symbol `.Llabel1' is already defined
llvm-ld:

[lattner]

Debug info problem?

[llvmbot]

Yep.

[asl]

Proposed fix
This fixed for me with the patch attached. Before we assumed, that every function can have dbg.start_function() call, which is definitely incorrect.

Fixme in the comment can be connected with llvm/llvm-bugzilla-archive#2075 .

Evan, please have a look.

[llvmbot]

This seems like the wrong fix and will likely break debugging (at least on Mac OS X) for this particular function. Can you ind out why there wasn't a dbg.start_function()?

[asl]

Well, but anyway we shouldn't dereference invalid pointer here, even if code is invalid :)

Maybe Verifier should check/reject such stuff

[llvmbot]

Perhaps add an assertion? I'd rather the code generator poop than generating bad stuff. But this is debugging info so we shouldn't be that strict. I'd still want to track down the llvm-gcc bug. Anton, do you think you can take a look there?

[llvmbot]

Mark this WONTFIX now that debug info has changed significantly.

[asl]

mentioned in issue llvm/llvm-bugzilla-archive#2075

[asl]

mentioned in issue llvm/llvm-bugzilla-archive#2075