[llvm-commits] [llvm] r169886 - /llvm/trunk/lib/MC/MCAssembler.cpp

[Michael Spencer]

On Fri, Dec 21, 2012 at 7:56 AM, Eli Bendersky <eliben at google.com> wrote:
> On Fri, Dec 21, 2012 at 7:13 AM, Timur Iskhodzhanov <timurrrr at google.com> wrote:
>> For example, if fails to build this test on my ASan/Win bot:
>> http://code.google.com/p/address-sanitizer/source/browse/trunk/win/tests/malloc_right_oob_crash.cpp
>> this also fires up an assertion message box which hangs the bot :(
>>
>> Is it possible to quick-fix this or revert?
>
> WinCOFFStreamer seems undecided about using the RelaxAll flag. It
> overrides MCObjectStreamer's EmitInstruction but does not actually
> early-relax instructions in the presence of RelaxAll, so it's not
> clear to me why it needs this flag at all. Perhaps there's a bigger
> problem hiding here?
>
> Does anyone familiar with this streamer know the answer? Michael?
>
> Also I wonder why there are no regression tests that turn this up.
>
> Eli

It looks to me as it can simply use the MCObjectStreamer's
implementation of EmitInstruction. The COFF streamer was written
before we moved a lot of shared logic up to MCObjectStreamer, and
apparently COFF didn't get updated in that change.

Tests that break this would be nice.

- Michael Spencer