[compiler-rt] r218545 - Effectively revert r217284 on Darwin due to issues with -gmlt there

[David Blaikie]

On Mon, Sep 29, 2014 at 11:39 AM, Adrian Prantl <aprantl at apple.com> wrote:

>
> On Sep 29, 2014, at 11:28 AM, David Blaikie <dblaikie at gmail.com> wrote:
>
>
>
> On Mon, Sep 29, 2014 at 6:17 AM, Alexander Potapenko <glider at google.com>
> wrote:
>
>> On Sat, Sep 27, 2014 at 2:22 AM, David Blaikie <dblaikie at gmail.com>
>> wrote:
>> >
>> >
>> > On Fri, Sep 26, 2014 at 2:58 PM, Alexey Samsonov <vonosmas at gmail.com>
>> wrote:
>> >>
>> >> Author: samsonov
>> >> Date: Fri Sep 26 16:58:33 2014
>> >> New Revision: 218545
>> >>
>> >> URL: http://llvm.org/viewvc/llvm-project?rev=218545&view=rev
>> >> Log:
>> >> Effectively revert r217284 on Darwin due to issues with -gmlt there
>> >
>> >
>> > Not sure if this is the right direction - but maybe I'm confused about
>> what
>> > was discussed in the other thread.
>> >
>> > I assume Apple uses/cares about -gmlt working on their platform and it
>> > currently doesn't. (I could be wrong though - maybe they don't care
>> about
>> > -gmlt) This change would just paper over that.
>> >
>> > But if -gmlt's behavior on Apple isn't important to anyone, I'm fine
>> with
>> > this - keeps the code simpler on my side of things.
>> >
>>
>> Not sure about whether Apple cares about -gmlt (is that the new
>> preferred name for -gline-tables-only?) or not, but without this
>> change a bunch of asan tests has been failing on Darwin.
>>
>
> Yeah, I'm just concerned this'll suppress the failure but leave the bug.
> But I'll leave it to Apple folks to decide what they care
> about/priorities/timelines when it comes to -gmlt.
>
>
> I’m a little confused now. Alexey, are the Darwin issues that you cited in
> the commit message the ones discussed in r218129 (dsymutil skipping
> compile units without subprograms?) or is this a separate issue?
>

Yes, this is the same issue. This change causes the ASan buildbot to pass
-g on darwin and -gline-tables-only on non-Darwin.


> We definitely do care about -gmlt working on Darwin, and we also care
> about Asan working on Darwin.
>

Then you/I/we/someone should patch -gmlt to not do that optimization on
Darwin & revert this patch to continue testing -gmlt + ASan.

Well, maybe. It's not inherent that compiler-rt should be used to test
-gmlt, but given that my understand was that -gmlt was implemented for
ASan's convenience/use (and other things that just need
backtracing/symbolication) I asked Alexey to change the compiler-rt testing
to use -gmlt so I could make changes to it and be sure I wasn't regressing
compiler-rt's needs. That seems as applicable to Darwin (if you guys care
about -gmlt + ASan, specifically), otherwise you might want to setup some
other bot with -gmlt + <whatever workload you care about being able to do
stuff with -gmlt>, though I imagine it'll boil down to the same bugs caught
in basically the same way.

- David


>
> -- adrian
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140929/9806ee2b/attachment.html>