[9.0.0 Release] Release Candidate 6 is here

Hello everyone,

9.0.0-rc6 was just tagged from the release_90 branch at r372100. In
the Git monorepo, it's tagged as llvmorg-9.0.0-rc6.

Source code and docs are available at https://prereleases.llvm.org/9.0.0/#rc6

This is the same as rc5 plus one very minor change (r371969) that
still seemed good to pick up.

I'm not allocating extra time for testing this one, expecting to tag
'final' in a day or two unless any new problem arises. If you still
have cycles to spare, testing is still very much appreciated of

As usual, please file bug reports about any issues you find, marking
them blocking of https://llvm.org/PR42474


I've updated OpenMandriva to it, all arches, no problems.

Even RISC-V is starting to work (though still a long way from being able to replace gcc on that arch). Unless someone asks me to, I won't send bug reports for what's still failing there because the code already has lots of comments saying it's broken (e.g. abi is always assumed to be ilp32 even when the target system is lp64d).

One RISC-V thing that's broken without a corresponding comment [at least one that I've found] is C++ linkage -- passing the right flags makes C applications compile (and work) with clang, but clang++ fails:
# clang++ -march=rv64imafdc -mabi=lp64d test.cpp
/usr/bin/ld: error in /tmp/test-24247e.o(.eh_frame); no .eh_frame_hdr table will be created


Hi Bernhard,

Thank you for trying the RISC-V backend, and looking for comments in the backend before filing issues.

The RISC-V backend has only recently become stable, and there are indeed some bugs in the trickier parts of the backend.

We (lowRISC and other RISC-V backend contributors) would welcome you to file bugs against the backend, but we do not feel these bugs should block the release, as they are not regressions.


Nothing new on Gentoo.

Obtained the same results as the rc4 run that was considered non-blocking.

Failing Tests (1):
Clang Tools :: clang-tidy/run-clang-tidy.cpp

Expected Passes : 62200
Expected Failures : 256
Unsupported Tests : 1741
Unexpected Failures: 1


Same as before.



This has now been tagged as the final 9.0.0 release. In the Git
monorepo, it's tagged as llvmorg-9.0.0.

The official release announcement will follow as soon as the source
tarballs and docs are ready on the web page.

Testers: please start building the final binaries, and I'll add them
as they become ready.

Many thanks for your hard work!

Is it expected that after moving 7.1.0 and 8.0.1 to GitHub, 9.0.0 is
again distributed via releases.llvm.org? Not that I care either way but
it's really frustrating to have to update URLs back and forth.

tarballs was done as a test. I agree it could probably have been
communicated better.

Windows is ready:

$ sha256sum LLVM-9.0.0-win*.exe

They were built with the attached batch file.

It seems in the no-asserts 32-bit build, two tests pass unexpectedly:

Unexpected Passing Tests (2):
  LLVM :: ExecutionEngine/MCJIT/stubs-sm-pic.ll
  LLVM :: ExecutionEngine/OrcMCJIT/stubs-sm-pic.ll

I think that's safe to ignore.

build_llvm_900-final.bat|attachment (5.44 KB)

FWIW - I've released a build of my MinGW toolchain/distribution with the 9.0.0 release now: https://github.com/mstorsjo/llvm-mingw/releases/tag/20190920
(Normally I use random trunk versions after a bit of testing.)

It's clearly not an official llvm/clang release or anything like that like the other ones posted here in this thread, but still a convenient package of compiler and sdk, and available for both x86 and arm windows (in addition to use as cross compiler from linux).

// Martin

The assembly sequence for this is not clear. I was thinking

-release 9.0.0
-triple x86_64-pc-linux-gnu

But perhaps without -test-asserts. Looking at

gives some manual items under ‘Release Process’ but perhaps those
are now contained in an automated sequence.

Neil Nelson

I know it's too late to report problems but I'm wondering if anybody
else sees additional test failures after switching from git checkout to
source tarballs. I'm currently investigating as to why.

Uploaded ubuntu x86 binaries:

4361dc68e47b9972f0ec7db0c20411f60df90f28 clang+llvm-9.0.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
085f223d48fd6fcf372f54ccdf8bbac403c279a2 clang+llvm-9.0.0-x86_64-linux-gnu-ubuntu-16.04.tar.xz
b1627073e533975f0b4d4366b5f6e4dd44c7c6d8 clang+llvm-9.0.0-x86_64-linux-gnu-ubuntu-18.04.tar.xz

Xubuntu/Ubuntu 19.04 uploaded

sha256sum clang+llvm-9.0.0-x86_64-pc-linux-gnu.tar.xz

Neil Nelson

Xubuntu/Ubuntu 19.04 uploaded

sha256sum clang+llvm-9.0.0-x86_64-pc-linux-gnu.tar.xz

Neil Nelson

Uploaded SLES11:
810a828c7874d2753049493fa80f8228a0167067 clang+llvm-9.0.0-x86_64-linux-sles11.3.tar.xz

For the final release of 9.0.0, I used two patches, from:

* https://bugs.llvm.org/show_bug.cgi?id=42892 - After r356631, Sanitizer-i386-Test faills to link on FreeBSD
* https://bugs.llvm.org/show_bug.cgi?id=42894 - FreeBSD needs -pthread link flag for dynamic ASan tests


SHA256 (clang+llvm-9.0.0-amd64-unknown-freebsd11.tar.xz) = 2a1f123a9d992c9719ef7677e127182ca707a5984a929f1c3f34fbb95ffbf6f3
SHA256 (clang+llvm-9.0.0-i386-unknown-freebsd11.tar.xz) = 2d8d0b712946d6bc76317c4093ce77634ef6d502c343e1f3f6b841401db8fa56

Main test results on amd64-freebsd11:





it has been a few days since 9.0 was released. I'm missing the macOS binary. Do you have any status update on that? Thank you!