[PATCH] D43779: [Tooling] [0/1] Refactor FrontendActionFactory::create() to return std::unique_ptr<>

[Alexander Kornienko via Phabricator via llvm-commits]

alexfh added a comment.

In https://reviews.llvm.org/D43779#1061043, @lebedev.ri wrote:

> Hmm.
>  Got back to this issue.
>
> Not reproducible with gcc-4.8.5 and gcc-4.9.3 in ubuntu 16.04 chroot.
>  *Reproducible* with gcc-4.8.4 and gcc-4.9.2 in debian oldstable (Jessie) chroot.
>  So one might assume that it was fixed.
>
> I did try to creduce the crasher, but not not much success, since gcc takes 10+ sec to crash.


That's unfortunate, but it could be even worse. My experience with creducing crashes is that the process can take up to several days. In this case, however, we know what changes cause the crash and it should be possible to construct the test case manually instead of creducing it.

> Should i simply try to re-commit and see if the buildbots got upgraded so this is no longer an issue?

I wouldn't expect buildbots to have been upgraded without someone doing this on purpose. If you have the list of buildbots that crashed, you could look at their recent logs to figure out which GCC version they are using now. It *might* be fine to require a patched version of GCC, but in that case you would have to:
0. Ask llvm-dev+cfe-dev whether anyone has concerns in raising the minimum required version of GCC

1. Fix the documentation, in particular this list: https://llvm.org/docs/GettingStarted.html#software and these instructions: https://llvm.org/docs/GettingStarted.html#getting-a-modern-host-c-toolchain
2. Get buildbot maintainers to upgrade their GCC to at least the new required version

An alternative would be to try working around the bug.


Repository:
  rL LLVM

https://reviews.llvm.org/D43779