Ah, good thinking Ben, thanks. I've had a quick look, and the rubygems
included in ruby version 2.3.1 tarball includes all of the symlinks fixes
made in PR#1209, so *maybe* we don't need rubygems version 2.5.2 after all.
I'm just testing to see if upgrading the ruby package to 2.3.1 fixes
everything.

-r

P.S. I certainly didn't mean to insinuate that this was the fault of
Ricardo (or you, for that matter). Thanks to all contributors!


On Wed, 27 Apr 2016 at 20:26 Ben Woodcroft <woodibe@gmail.com> wrote:

> Hi Rob,
>
> On 27/04/16 20:39, Rob Syme wrote:
> > Hi all
> >
> > Running `guix environment --ad-hoc ruby-ansi` fails for me (and
> > others). The package definition uses the ruby-build-system but fetches
> > the tar.gz directly from github instead of from rubygems. I'll try and
> > find time to have a closer look, but if Ricardo has any time, he might
> > be able to debug it faster than me.
>
> I think this the error is very similar to that described here, and
> occurs during 'gem install'
> https://github.com/rubygems/rubygems/issues/1448
>
> which is fixed in rubygems 2.5.2, but unfortunately ruby comes with
> 2.5.1. Today a new version of ruby was released, but sadly, still 2.5.1.
> So, I think this means we will need to either incorporate the patch(es)
> that fix this into the ruby package, or package rubygems alongside ruby
> somehow. Or, since there is only a problem during build time, perhaps
> clobber the gem from the ruby package with a rubygems package during
> build time. Unless you have any better ideas?
>
> Thanks for reporting this. I'm not sure we can blame Ricardo for this.
> Rather, it is all my fault.
> ben
>