[llvm-bugs] [Bug 28024] New: ppc64 (be and le) don't have a working std::call_once

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=28024

            Bug ID: 28024
           Summary: ppc64 (be and le) don't have a working std::call_once
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Core LLVM classes
          Assignee: unassignedbugs at nondot.org
          Reporter: chandlerc at gmail.com
                CC: hfinkel at anl.gov, kbarton at ca.ibm.com,
                    llvm-bugs at lists.llvm.org
    Classification: Unclassified

This appears to be a bug in the libstdc++ implementation of std::call_once, but
it may be somewhat specific to ABI issues as there are differences between
stage1 and stage2 in LE.

Here are the build bot builds that started failing when I landed r271800 that
switch LLVM to use std::call_once on various implementations:
http://lab.llvm.org:8011/builders/clang-ppc64le-linux-multistage/builds/1169
http://lab.llvm.org:8011/builders/clang-ppc64be-linux-multistage/builds/2703

These fail in unit tests that use INITIALIZE_PASS which defines a function that
uses std::call_once. The failure happens when the implementation of
std::call_once throws std::system_error with the '-1' return of a call to
pthread_once. I don't know why pthread_once would fail here. I've tried to
reproduce any failure to properly initialize the std::once_flag (which contains
as a member the pthread_once_t variable passed to pthread_once) but to no avail
on my system.

Hopefully someone who works on PPC can figure this out.

Also, when I tried to disable this in r271821, it worked for ppc64le, and for
stage2 in ppc64be, but not for stage1 for some reason! Really weird:
http://lab.llvm.org:8011/builders/clang-ppc64be-linux-multistage/builds/2707

Again, I can't figure out how or why this would happen. Hopefully some PPC
folks can figure this out.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160606/01f108f9/attachment-0001.html>