[PATCH] D33852: Enable __declspec(selectany) on linux

Hans Wennborg via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Mon Aug 28 13:53:03 PDT 2017


hans added a comment.

In https://reviews.llvm.org/D33852#854213, @aaron.ballman wrote:

> In https://reviews.llvm.org/D33852#854176, @Prazek wrote:
>
> > In https://reviews.llvm.org/D33852#853982, @aaron.ballman wrote:
> >
> > > In https://reviews.llvm.org/D33852#853410, @Prazek wrote:
> > >
> > > > Sorry for so late fixes, but it would be good to put it in 5.0
> > >
> > >
> > > I do not think this should be in 5.0, as I believe we're only accepting regression fixes at this point.
> >
> >
> > This is a regression. __declspec(selectany) works completely fine on linux with 4.0. Without it clang-5.0 will be useless for any major windows project compiled on linux.
> >
> > edit:
> >  here is a regression: https://reviews.llvm.org/D32083
>
>
> This is also adding new functionality that has had zero testing because it removes *all* target-specific checking for the attribute. Under the previous functionality (changed in https://reviews.llvm.org/D32083), this still required some mention of microsoft something (it went from requiring microsoft extensions to be enabled to instead require a Windows target) -- that's been entirely removed from this patch so now you can use this attribute for all target architectures, so it's not purely fixing a regression. Given how late we are in the release cycle, I am uncomfortable with this going in to 5.0, but I'd have no problems letting it bake for a bit and putting it into 5.1 (or 5.0.1, however we're naming bug releases these days).


I agree. The idea of taking this so late (and it hasn't even landed in trunk yet) makes me nervous. I think 5.0.1 would be more appropriate.


https://reviews.llvm.org/D33852





More information about the cfe-commits mailing list