can't use clang -g to assemble .s file with .file directive

[llvmbot]

Bugzilla Link	11740
Resolution	FIXED
Resolved on	Apr 19, 2016 19:48
Version	trunk
OS	Linux
Blocks	llvm/llvm-bugzilla-archive#24345
Reporter	LLVM Bugzilla Contributor
CC	@compnerd

Extended Description

$ cat x.s
.file "x.ll"
$ clang -g x.s -c -o x.o
x.s:1:2: error: input can't have .file dwarf directives when -g is used to generate dwarf debug info for assembly code
.file "x.ll"
^

Yes, I'd like to pass -g to the compiler even when its input is a .s file. Even if the compiler isn't generating debug info so to speak. Please don't error on this!

[llvmbot]

Looks like the check for this in GenericAsmParser::ParseDirectiveFile() in lib/MC/MCParserAsmParser.cpp is in the wrong place. It it confusing the cpp .file directive and the dwarf .file directive. The check needs to me moved inside the else clause when FileNumber != -1.

[llvmbot]

Got this fixed and will commit this in the morning when I get back into the office. My tree is out of date and am rebuilding now but need to head home now.

[llvmbot]

Fixed with r147951.

[llvmbot]

cat a.c

void foo() {}

clang -g a.c -S -o a.s
clang -g -c -o a.o a.s

a.s:8:2: error: input can't have .file dwarf directives when -g is used to generate dwarf debug info for assembly code
.file 1 "a.c"
^
a.s:8:8: error: file number already allocated
.file 1 "a.c"
^

[llvmbot]

It is probably not related to clang.

I got a similar bug like this one
https://sourceware.org/bugzilla/show_bug.cgi?id=6428

The difference is that for clang, '-g' also triggers the problem.

[compnerd]

./bin/clang -g -x c -S - -o - <<< 'extern int silence_pedantic_warnings;' | ./bin/clang -g -x assembler-with-cpp -c - -o /dev/null -Werror -Wall -Wextra -Wpedantic

SVN r266560

Closing again, please re-open if I had something wrong with the test.

[llvmbot]

backtrace

[llvmbot]

Hi Saleem,

I think your solution does not work for my llvm trunk 266834.

The backtrace of the error is as the attachment.

[compnerd]

That error doesn't seem to be related to this. That is an redefined label. Please file a new bug with a reproduction. You can assign it to me if you like. Even if the input is invalid, it shouldn't assert.

[llvmbot]

mentioned in issue llvm/llvm-bugzilla-archive#24345