On Mon, Mar 20, 2017 at 4:09 PM, Daniel Berlin <dberlin at dberlin.org>
wrote:> Interesting.
> I 3-stage basically every day on darwin, and have not gotten a single
> failure yet.
> i just use cmake + ninja though
> Are you sure it's not related to the computer involved?
>
These builds are on a MacPro 3,1 using Xcode 8,2,1 as the system
compiler. The same build approach...
http://fink.cvs.sourceforge.net/viewvc/fink/dists/10.9-libcxx/stable/main/finkinfo/languages/llvm39.info?revision=1.2&view=markup
has been used to test
llvm/clang/clang-tools-extra/compiler-rt/openmp/polly trunk for
several years now and this is the first time I've seen instability in
the file comparison of the stage2/stage3 build.
Jack
ps I've repeated the build of
llvm/clang/clang-tools-extra/compiler-rt/openmp/polly for the 4.0.0
release and it still passes the file comparison of the stage2/stage3
without issue. The fact that the exact list of files failing the
stage2/stage3 comparison varies seems especially concerning.
>
> On Mon, Mar 20, 2017 at 1:08 PM, Jack Howarth via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
>>
>> Do any of the current build bots for llvm.org perform 3-stage
>> bootstraps with file comparison of the stage2 and stage3 object files
>> and generated headers? On x86_64-apple-darwin16 using the fink
>> projects llvm packaging methodology (crafted by David Fang), I am
>> seeing non-deterministic file comparison failures in current trunk
>> that goes back as far as r296837.
>>
>> http://bugs.llvm.org/show_bug.cgi?id=32342
>>
>> The non-deterministic nature of this bug is making it rather difficult
>> to pinpoint the actual commit that triggered it.
>> Jack
>> ps The issue seems to be non-deterministic in both its appearance and
>> the exact subset of stage2/stage3 object files that differ when it
>> triggers.
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>