Heads up: Rebuild for Ruby 1.9.3

[Mo Morsi]

On 01/25/2012 03:40 AM, Vít Ondruch wrote:
> Dne 25.1.2012 00:52, Rex Dieter napsal(a):
>> Mo Morsi wrote:
>>
>>> On 01/24/2012 04:50 AM, Bohuslav Kabrda wrote:
>>>> Hi,
>>>> since we finally got our Ruby 1.9.3 feature page [1] approved, we are
>>>> starting rebuild for Ruby 1.9.3. Everyone who owns a package that 
>>>> depends
>>>> on Ruby or Rubygems should rebuild it in the special Koji target
>>>> "f17-ruby". This target will be merged into rawhide just before the
>>>> branching, which will occur on 2012-02-07. After this deadline, you 
>>>> will
>>>> need to go through the standard Bodhi update process. The detailed
>>>> instructions on rebuilding the packages, including our suggested 
>>>> changes
>>>> can be found in [2]. Feel free to contact me or Vit Ondruch 
>>>> (vondruch at
>>>> redhat dot com) or write to ruby-sig mailing lists, if you run into 
>>>> any
>>>> kind of problems.
>>>>
>>>>
>>> I'm somewhat confused. Has the issue w/ sudo gem install been resolved?
>> Any citation or bug references for this problem you mention?
>>
>> -- rex
>>
>
> Let me clear the situation. First of all, this is not issue of "sudo 
> gem install". That command works just fine. That is issue with 
> Bundler. Unfortunately, Bundler is one big hack above RubyGems based 
> on some assumptions which are not true. It can be seen on its code 
> where is a lot of custom logic for specific version o RubyGems and 
> Bundler has a lot more quirks with regard of its usage on package 
> driven systems. It can break for example with every upstream change of 
> RubyGems. Now what is the issue:
>
> Ruby, speaking about upstream as well as 1.8 in Fedora, traditionally 
> broke the FHS. We tried to remove this breakage with Ruby 1.9. 
> RubyGems traditionally installs their binary extensions into the same 
> place as their platform independent code, incompatible with FHS. To be 
> somehow compatible with FHS, this was changed and for Ruby 1.8, the 
> binary extensions were installed into ruby_sitedir. That results in 
> conflicting binary RPMs which conflicts among their versions, although 
> this was never the case for RubyGems packages. So for Ruby 1.9.3 we 
> tried to work on this issue and we modified RubyGems to be able to 
> load the libraries, where the noarch part is placed in share and the 
> binary library in lib, so RPM packaged gems are installed into 
> /usr/share and their binary libraries into /usr/lib. Moreover, we 
> changed how RubyGems works, e.g. by default, "gem install" installs 
> gems into your home directory, so you don't need sudo to work with 
> gems anymore. However, if your are using "gem install" with root 
> users, they are installed into /usr/local/share and their binary 
> libraries are installed into /usr/local/lib.
>
> Now there comes Bundler. Assuming it knows everything about RubyGems, 
> they put together their custom logic how to populate Ruby's load 
> paths, which does not work well with changes we made into RubyGems, 
> i.e. it cannot add the binary extension path to the Ruby's load path, 
> resulting in failure when loading the gem. Nevertheless, this is going 
> to be handled in RPM packaged Bundler and it will work properly as 
> long as user uses RPM packaged Bundler. Once somebody installs 
> upstream version of Bundler, the Bundler fails naturally.
>
> So there are probably 4 solutions to this issue:
>
> 1) Always use Bundler provided by Fedora which will works as it should.
> 2) Force Ruby and RubyGems upstream to properly support FHS. I already 
> provided patches [1] but I need your support.
> 3) Revert the customized behavior of RubyGems and break FHS.
> 4) Treat root as regular user and install gem also into root's home 
> directory, but it obviously doesn't make sense.
>
> So obviously I have chosen 1) and trying to move forward with 2). That 
> works most of the times. It breaks only in situation, when you install 
> upstream Bundler using "sudo gem install" and later you want to use 
> some system wide gem with binary extension. If you install gems using 
> "gem install" into your home dir (preferred, default way), it works 
> every time. Mixing of RPMs with other package managers has its limits 
> and this is one of them.
>


OK so the only thing that will not work after everything is said and 
done is running 'sudo bundle install' if you've installed bunder via 
'gem install bundler' and not 'yum install rubygem-bundler' correct?

If that is the case, I can live with this as running 'bundle install' as 
sudo is discouraged upstream anyways. I've seen a few cases which this 
is used as a hacky workaround, but I think the intention on bundler is 
to vendorize dependencies locally and is not meant for system-wide 
management.

My only concern is that upstream practices will change yet again in a 
manner which our reworking of the load paths to conform to the FHS will 
be incompatible. Try as we might, I don't believe the Fedora ruby 
community has enough sway with the upstream ruby community to change 
their practices (yet) so I'm willing to go forward with this new 
direction provided that if it becomes incompatible w/ a widespread 
upstream practice down the road, we evaluate the situation and possibly 
make the necessary changes to play well w/ the Ruby community. At least 
until the day we can reasonably influence the direction of the upstream 
practices.

   -Mo