FYI, Ruby 2.6.0 has a known bug that affects http. Details at Ruby Bug #15468 and ruby/ruby#2058.

There's not much to prepare, but maybe let's prepare for switching to it once 2.6.1 is released? (Assuming it releases with the fix.)

rubygems/rubygems#2587 affects us also – (some?) gems don't seem to build using our buildRubyGem stuff because of a regression in RubyGems. It's been fixed in master, but we probably want to wait for a Ruby release that includes a RubyGems release with the fix.

Also lsegal/yard#1218 seems to affect quite a few gems if they're built with documentation, but that's easy to work around.

I just merged #54939, but it doesn’t include the RubyGems fix. I don’t think we should merge this without it. Perhaps we could just update the RubyGems we ship with Ruby…

@GrahamcOfBorg test redmine gitlab

These use ruby, I guess?

Gitlab is best to be tested not on ofborg. The test timeouts because it loads a lot of of files via qemu's p9 files systems.

No, in this instance it's a real failure in dependencies.

I’ve been using this in my config for ages now, and haven’t seen any problems that were not addressed by #60394, so I intend to merge this in the next couple of days unless anyone objects.

I tried using this but saw the problem with redis-rack from #60676 again. I manually cherry-picked the rubygems update from staging and this.
If it doesn't affect any existing packages this shouldn't prevent you from merging this, but I suspect the problem with rubygems/rubygems#2587 is still present.

Thanks @petabyteboy. Given how annoying that issue is to work around for the people who are most likely to benefit from this change, I’ll hold off merging this for now and see what can be done about that issue.

After thinking about this, I think we should backport the fix to
RubyGems 3.0.3, and then make Ruby 2.6 the default Ruby.

Agreed.
I will test again with the rubygems update, this and alyssais@06bd012

Unfortunately the problem was still present, even with the patch applied:
https://github.com/petabyteboy/nixpkgs/commits/feature/ruby
Is there anything obvious I'm missing?

With all commits in staging concerning ruby and those from the PR you mentioned cherry-picked I still run into the same problem.

Edit: Nevermind, with this tree it worked: https://github.com/petabyteboy/nixpkgs/commits/feature/ruby

Edit: Nevermind, with this tree it worked: petabyteboy/nixpkgs:ruby@feature (commits)

Would you mind posting this sort of thing as a separate comment in future? I didn't see this until now because GitHub doesn't notify for edits, and had spent days wondering how it could still be broken.

Sorry, I will do that if I remember. There are no issues anymore from my side, as long as we apply the patch for rubygems#2587 before updating.

Is there any new or existing reason not to merge this?

I think before we merge this, we should upgrade ruby to 2.6.3 as well as rubygems and see if that addresses the issue.

ruby_2_6 is already on 2.6.3 in unstable.
The rubygems version 3.0.3 should already have the fix, but 3974254f6a20191b6b07621f6ea928be75dad121 and dfe7e42ff5844a4fe9901ce1be6125127085e044 are required to make use of our own rubygems. The rubygems included in ruby 2.6.3 still has the issue.
I could build mastodon successfully with https://github.com/petabyteboy/nixpkgs/commits/feature/ruby-testing.

Friendly reminder that the branch-off is in 11 days — do you want this in 19.09 and what do you think the chances of it being ready on time are?

I think it's very realistic to make it before the branch-off. All the fixes have been committed since months. If we can merge it before that would be great, but if it doesn't work out that's not too bad either.

Closing in favour of #67597.