[PATCH] cmake: Export symbols in all tools that support plugins.

[Reid Kleckner]

Go for it.  Somebody could play around with protected ELF visibility if
they want to win back the ~3%.
http://www.airs.com/blog/archives/307


On Wed, Dec 4, 2013 at 2:56 PM, Will Dietz <w at wdtz.org> wrote:

> Interesting.
>
> Before this change 'ninja check' takes 1:47 on my machine, after the
> patch takes 1:50.
> (Minimum of 5 runs, times were pleasantly consistent, "-j1" added to
> LIT_ARGS)
>
> Note that in Make we presently export symbols in llc and lli, so we
> already take this hit on
> 'make check' when not using CMake.  llvm-stress doesn't currently
> export, but it should
> (since it supports plugins) but that shouldn't impact the testing use case.
>
> Accordingly IMO a configuration option for managing test times isn't worth
> it,
> but it might be useful for making the installed (stripped) binaries
> smaller.
> If this route is taken, I'd suggest disabling plugin loading in a
> friendlier
> way than removing -export-dynamic and giving undefined reference errors
> should a user attempt to load a plugin.
>
> Anyway, updated patch attached to address the inconsistencies.  Okay to
> commit?
>
> ~Will
>
>
> On Wed, Dec 4, 2013 at 12:58 PM, Rafael Espíndola
> <rafael.espindola at gmail.com> wrote:
> > On 4 December 2013 13:49, Reid Kleckner <rnk at google.com> wrote:
> >> Can you make sure that this doesn't slow down 'make check' on Linux too
> >> much?  I worry that we --export-dynamic for llc, lots of symbol uses
> will
> >> become dynamic PLT lookups and startup time will regress.
> >>
> >
> > Good point. If it does, we should probably add a --enable-plugins (and
> > cmake equivalent) and replace CLANG_IS_PRODUCTION with it.
> >
> > Cheers,
> > Rafael
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131204/7906dd88/attachment.html>