Re: 01/03: llvm: Build with RTTI by default.

[Eric Bavier <ericbavier@centurylink.net>]

[-- Attachment #1: Type: text/plain, Size: 2093 bytes --]

On Mon, 19 Nov 2018 00:05:18 +0100
ludo@gnu.org (Ludovic Courtès) wrote:

> Hello,
> 
> Mark H Weaver <mhw@netris.org> skribis:
> 
> > ericbavier@centurylink.net (Eric Bavier) writes:
> >  
> >> bavier pushed a commit to branch master
> >> in repository guix.
> >>
> >> commit fc9dbf41311d99d0fd8befc789ea7c0e35911890
> >> Author: Eric Bavier <bavier@member.fsf.org>
> >> Date:   Fri Oct 5 22:58:43 2018 -0500
> >>
> >>     llvm: Build with RTTI by default.
> >>     
> >>     Increases size of llvm@6 by 2.5MiB (2.5%), but saves building specialized
> >>     llvm's with rtti enabled for packages that require RTTI when linking with the
> >>     llvm libraries.
> >>     
> >>     * gnu/packages/llvm.scm (llvm-3.8-with-rtti): Remove variable.
> >>     (llvm)[arguments]: Add '-DLLVM_REQUIRES_RTTI=1' to #:configure-flags.
> >>     * gnu/packages/audio.scm (faust-2)[native-inputs]:
> >>     'llvm-3.8-with-rtti' -> 'llvm-3.8'.  
> >
> > Unfortunately, this seems to have broken Rust, and thus IceCat.  See:
> >
> >   https://hydra.gnu.org/build/3179552
> >
> > I don't have time to look at this right now, but please also keep in
> > mind the following commit:
> >
> >   commit f8cba3ff4fc6aed0c539700aa47b6e8f4c25c34d
> >   Author: Marius Bakke <mbakke@fastmail.com>
> >   Date:   Sat Nov 17 22:54:03 2018 +0100
> >   
> >       gnu: mesa: Stay on RTTI-less LLVM.
> >       
> >       * gnu/packages/llvm.scm (llvm-without-rtti): New public variable.
> >       * gnu/packages/gl.scm (mesa)[inputs]: Use that instead of LLVM.
> >
> > I would suggest simply reverting both of these commits for now, and
> > building this change out on another branch, possibly 'core-updates'.  
> 
> This commit also triggered more than 1K rebuilds per architecture, so
> Marius “semi-reverted” it in f8cba3ff4fc6aed0c539700aa47b6e8f4c25c34d
> after we discussed it on IRC.

Sorry about all the trouble.  Since posting the patches I had forgotten
how many packages were affected.  I suppose this should have been
pushed to a staging branch.

`~Eric

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]