[cfe-dev] Issues with https://reviews.llvm.org/D26082

[Renato Golin via cfe-dev]

On 12 January 2017 at 20:52, Jonathan Coe via cfe-dev
<cfe-dev at lists.llvm.org> wrote:
> Issues have been reported in https://reviews.llvm.org/D26082 which I have
> locally reproduced.
>
> What's the best course of action?

Hi Jonathan,

This is before we branched, so whatever action we take need to be
back-ported to the 4.0 branch.

Adding Hans FYI.


> I can revert the change until problems are fixed. I don't think libclang's
> Python bindings tests get run as part of CI.

Is this affecting buildbots or some kind of external CI?

It seems to me that we should test Python's bindings as a standard CI
run, if they're added in Clang in any way.

I'm not proposing running Python itself, as that can be hard to match
system expectations, but some kind of unit test to make sure the
bindings are correct on all supported versions of Python would go a
long way of avoiding this scenario in the future.

As to revert vs. not revert, it depends on what's broken and how quick
is the fix. The questions I always ask myself are:

1. Is the breakage stopping any CI look from reporting further bugs?
This also depends on how often the CI loop runs and how much relies on
it.
2. Is there someone looking at the problem? If no, then reverting it
the *only* course of action. If yes, we'll need time-frames for a
possible fix before deciding to revert.

If we hadn't crossed the branch barrier, I'd be all for reverting it.
But we have. So, to avoid multiple reversions and re-applies and
multiple RCs, I think we need to be a bit more cautious here.

cheers,
--renato