[llvm-dev] [10.0.0 Release] Release Candidate 1 is here

[Rainer Orth via llvm-dev]

Hi Hans,

> On Thu, Feb 6, 2020 at 5:24 PM Rainer Orth via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
>>
>> Hi Hans,
>>
>> > It took a bit longer than planned due to master being a somewhat
>> > unstable at the branch point, but Release Candidate 1 has now been
>> > tagged as llvmorg-10.0.0-rc1.
>> >
>> > Source code and docs are available at
>> > https://prereleases.llvm.org/10.0.0/#rc1
>> >
>> > Pre-built binaries will be added there as they become available.
>>
>> I've already uploaded Solaris 11.4/AMD64 and Solaris 11.4/SPARCv9
>> binaries to releases-origin.llvm.org some time ago.
>
> Sorry, I missed this. Did you send an email when you uploaded? Can you

I forgot, probably assuming that uploads to releases-origin.llvm.org
would be noticed (or even picked up) automatically.

> please share the SHA1 or similar hashes so I can verify that I get the
> right ones on my end?

Sure:

b029dbf9216ce5a2ebdf3219c05e3cc27ce2b2dd  clang+llvm-10.0.0-rc1-amd64-pc-solaris2.11.tar.xz
32056e671b4f25f23764b14a2b07e75b2bdb444a  clang+llvm-10.0.0-rc1-sparcv9-sun-solaris2.11.tar.xz

>> While amd64-pc-solaris2.11 results are en par with the LLVM 9.0.0 ones
>> (some failures fixed or xfailed since then)
>>
>>   Expected Passes    : 54488
>>   Expected Failures  : 180
>>   Unsupported Tests  : 2409
>>   Unexpected Failures: 23
>>
>> the sparcv9-sun-solaris2.11 ones are horrible compared to the 9.0.0 ones:
>>
>> * 9.0.0 final:
>>
>>   Expected Passes    : 48477
>>   Expected Failures  : 180
>>   Unsupported Tests  : 1443
>>   Unexpected Passes  : 1
>>   Unexpected Failures: 382
>>
>> * 10.0.0 rc1:
>>
>>   Expected Passes    : 47959
>>   Expected Failures  : 186
>>   Unsupported Tests  : 1615
>>   Unexpected Passes  : 1
>>   Unexpected Failures: 4278
>>
>> A large number of those only occur for non-SPARC targets (something I
>> usually don't test at all), and even the buildbot is way way better with
>> only 51 unexpected failures.  However, that one only does a 1-stage
>> build where errors due to Bug 42535 don't occur.  Still, comparing
>> master results on Solaris 11.5/SPARC for all targets between a 1-stage
>> build with gcc 9.1.0 and a 2-stage build shows a similar pattern:
>>
>> * 1-stage:
>>
>>   Expected Passes    : 53734
>>   Expected Failures  : 205
>>   Unsupported Tests  : 2495
>>   Unexpected Passes  : 1
>>   Unexpected Failures: 228
>>
>> * 2-stage:
>>
>>   Expected Passes    : 45756
>>   Expected Failures  : 205
>>   Unsupported Tests  : 2495
>>   Unexpected Passes  : 2
>>   Unexpected Failures: 7303
>>
>> I fear this is effectively impossible to analyze let alone fix for the
>> 10.0.0 release.
>
> Thanks for the report!
>
> I guess since this doesn't seem to be well covered by continuous
> testing, it's not going to be good :-/

Right: I simply don't have the cpu cycles to run 2- or even 3-stage
builds on sparc, and as I mentioned a fast majority of issues only
occurs there.  All I can afford is 1-stage rebuilds with gcc 9 plus
check-all, and those look reasonable.

> Should we highlight the status of Solaris/Sparc support in the release
> notes somehow?

I guess so, especially since the results are so much worse then the
9.0.0 ones.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University