Opened 2 months ago

Closed 6 weeks ago

Last modified 6 weeks ago

#14911 closed enhancement (fixed)

llvm-12.0.0

Reported by: Xi Ruoyao Owned by: Douglas R. Reno
Priority: normal Milestone: 10.2
Component: BOOK Version: SVN
Severity: normal Keywords:
Cc:

Description (last modified by Xi Ruoyao)

New major version.

We need to wait for rustc-1.52.0, or make a rustc patch for it.

Change History (13)

comment:1 by Xi Ruoyao, 2 months ago

Description: modified (diff)

I was wrong: even rustc-1.51.0 won't work out of the box.

comment:2 by Xi Ruoyao, 2 months ago

Version: SVN

comment:3 by Douglas R. Reno, 8 weeks ago

Type: taskenhancement

Change "task" to "enhancement" on recent tickets.

in reply to:  2 comment:4 by ken@…, 7 weeks ago

Replying to Xi Ruoyao:

Arghh, I didn't see that comment. I've just been doing a manual build (tried installing 12.0.0 with my normal script, ignoring tests, but that failed somewhere without any indication of exactly where.

With the manual build I ran the tests:

Failed Tests (13):
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedInvalidatingPasses
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/FunctionCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/FunctionCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedInvalidatingLoopNestPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedInvalidatingPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/ModuleCallbacksTest.InstrumentedSkippedPasses
  MemorySanitizer-X86_64 :: fstat.cpp
  SanitizerCommon-msan-x86_64-Linux :: Posix/lstat.cpp
  libFuzzer :: trace-malloc-unbalanced.test

I've now looked at what I wrote in the wiki for rustc : system llvm, when it has worked, was a quicker and smaller build with the "shared" libs. I was hoping to see if that was still true.

comment:5 by ken@…, 7 weeks ago

I have no idea if the additional test failures are important, but I suggest we should leave this for a while. If the inability to build current rust releases continues when rustc-1.52.0 is released, I guess that Arch or fedora will find a fix.

comment:6 by Douglas R. Reno, 7 weeks ago

Hi Ken,

Once rustc-1.52.0 comes out, can you report here if it works?

comment:7 by Douglas R. Reno, 6 weeks ago

Owner: changed from blfs-book to Douglas R. Reno
Status: newassigned

Grab stuff that needs to be synced to LFS and some other stuff to work on today.

comment:8 by Douglas R. Reno, 6 weeks ago

I'm working on this first today. Since rustc-1.52 will work with it natively, there shouldn't any problems there. I'll get back to you on test results when the tests are done running, I just did a build without tests for the figures. So far, there are a few new programs, and a few removed programs.

comment:9 by Douglas R. Reno, 6 weeks ago

Release notes can be found here:

https://releases.llvm.org/12.0.0/docs/ReleaseNotes.html - LLVM itself

https://releases.llvm.org/12.0.0/tools/clang/docs/ReleaseNotes.html#what-s-new-in-clang-12-0-0 - Clang

There don't seem to be any significant changes for us x86 users other than new features.

comment:10 by Douglas R. Reno, 6 weeks ago

I wound up with 20 test failures in this run

Failed Tests (20):
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedInvalidatingPasses
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/CGSCCCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/FunctionCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/FunctionCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedInvalidatingLoopNestPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedInvalidatingPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedPasses
  LLVM-Unit :: IR/./IRTests/LoopCallbacksTest.InstrumentedSkippedPasses
  LLVM-Unit :: IR/./IRTests/ModuleCallbacksTest.InstrumentedSkippedPasses
  LLVM :: tools/llvm-libtool-darwin/L-and-l.test
  LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/many_threads_detach.cpp
  LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/many_tls_keys_thread.cpp
  LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/new_array_with_dtor_0.cpp
  LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/pointer_to_self.cpp
  LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/print_suppressions.cpp
  MemorySanitizer-X86_64 :: fstat.cpp
  SanitizerCommon-msan-x86_64-Linux :: Posix/lstat.cpp
  libFuzzer :: only-some-bytes.test
  libFuzzer :: trace-malloc-unbalanced.test


Testing Time: 3878.31s
  Unsupported      : 17096
  Passed           : 56605
  Expectedly Failed:   128
  Failed           :    20

The tests took a lot longer than they did in previous releases, and I ended up in swap at some point (I have 16GB of RAM, but it looks like it hit 2GB in swap, so that would be 18GB of total RAM usage at one point).

comment:11 by Douglas R. Reno, 6 weeks ago

Resolution: fixed
Status: assignedclosed

in reply to:  10 ; comment:12 by ken@…, 6 weeks ago

Replying to Douglas R. Reno:

The tests took a lot longer than they did in previous releases, and I ended up in swap at some point (I have 16GB of RAM, but it looks like it hit 2GB in swap, so that would be 18GB of total RAM usage at one point).

A late comment: the tests use all cores (or more probably all+2 if all>=4). If I'm measuring for the book, I end up taking cores > #3 offline for these situations, and for big packages on small machines I may have to kill graphical browsers.

For building with 8 cores (in practice, SMT) 16GB is often a bit pokey.

in reply to:  12 comment:13 by Douglas R. Reno, 6 weeks ago

Replying to ken@…:

Replying to Douglas R. Reno:

The tests took a lot longer than they did in previous releases, and I ended up in swap at some point (I have 16GB of RAM, but it looks like it hit 2GB in swap, so that would be 18GB of total RAM usage at one point).

A late comment: the tests use all cores (or more probably all+2 if all>=4). If I'm measuring for the book, I end up taking cores > #3 offline for these situations, and for big packages on small machines I may have to kill graphical browsers.

For building with 8 cores (in practice, SMT) 16GB is often a bit pokey.

That's a good point. I only have four cores, but in the future, I'll remember that

Note: See TracTickets for help on using tickets.