Raise if ruby platform is forced and there are no ruby variants

[deivid-rodriguez]

What was the end-user or developer problem that led to this PR?

We have been reported a case where Bundler generates a lockfile when using force_ruby_platform that then makes Bundler crash when run on a different platform.

The case was reported by @stex here: #5422 (comment).

When using force_ruby_platform on a Gemfile like this:

source "https://rubygems.org"

gem "sorbet", "0.5.9889"

the following Gemfile.lockfile is generated

GEM
  remote: https://rubygems.org/
  specs:
    sorbet (0.5.9889)
      sorbet-static (= 0.5.9889)
    sorbet-static (0.5.9889-universal-darwin-14)
    sorbet-static (0.5.9889-universal-darwin-15)
    sorbet-static (0.5.9889-universal-darwin-16)
    sorbet-static (0.5.9889-universal-darwin-17)
    sorbet-static (0.5.9889-universal-darwin-18)
    sorbet-static (0.5.9889-universal-darwin-19)
    sorbet-static (0.5.9889-universal-darwin-20)
    sorbet-static (0.5.9889-universal-darwin-21)
    sorbet-static (0.5.9889-x86_64-linux)

PLATFORMS
  ruby

DEPENDENCIES
  sorbet (= 0.5.9889)

BUNDLED WITH
   2.3.12

And using that lockfile under a M1 docker container (which has platform "aarch64-linux") leads to a crash because no compatible gem can be found for that arch.

Closes #5422.

What is your fix for the problem, implemented in this PR?

We should give a better error message when dealing with this Gemfile.lock file anyways, but in my opinion, the above lockfile should not be generated at all because the user is explicitly saying "I don't want platform specific gems" by specifying force_ruby_platform, and sorbet-static can't satisfy that.

I prepared a patch to implement the above and was really happy because it seemed to work well and it led to removing this line:

https://github.com/rubygems/rubygems/blob/2faada63ae65d8559bcba9e083f3c35ba6c18fdc/bundler/lib/bundler/match_platform.rb#L18

Which was the specific line giving @lloeki issues when trying to add non gnu libc support at #4488, so this could also unblock that.

However... It does not work on truffleruby because... truffleruby needs its own special behaviour that it's hard to support. In particular, truffleruby declares its own platform as ruby (like force_ruby_platform does) because it's ABI incompatible with CRuby, so it doesn't want platform specific gems installed. However, it also keeps a hardcoded list of gem names where platform specific gems apparently work, and it's supposed to actually install platform specific gems in those cases.

I will investigate how to keep truffleruby working fine and still be able to introduce this, but I'm opening a WIP PR for now because I'm not sure when I'll get to it.

Make sure the following tasks are checked

• Describe the problem / feature
• Write tests for features and bug fixes
• Write code to solve the problem
• Make sure you follow the current code style and write meaningful commit messages without tags

[lloeki]

However... It does not work on truffleruby because... truffleruby needs its own special behaviour that it's hard to support. In particular, truffleruby declares its own platform as ruby (like force_ruby_platform does) because it's ABI incompatible with CRuby, so it doesn't want platform specific gems installed. However, it also keeps a hardcoded list of gem names where platform specific gems apparently work, and it's supposed to actually install platform specific gems in those cases.

Oh my. Makes me think of Alpine-packaged Ruby in some ways.

[deivid-rodriguez]

Alright, I managed to figure this out, so I'll set this as ready.

In the end, it was not a truffleruby specific issue 😳. It was just that a spec only run on truffleruby was the one that caught the problem, but the issue also happened when running the same test case on other rubies.

I did have a lot of trouble with truffleruby because my fixes was just not being picked up there, until I found that it monkeypatches Bundler in a very confusing way.

[deivid-rodriguez]

This might fully close #5422, it does fix the only reproducible case reported there. I'll set this PR to close that ticket and if still reproducible after upgrading, people can reopen it.

[eregon]

I did have a lot of trouble with truffleruby because my fixes was just not being picked up there, until I found that it monkeypatches Bundler in a very confusing way.

I am sorry about that. The reason for that way is to patch older Bundler versions (which need that fix), we should only do something if Bundler is older than when that fix was added.

[deivid-rodriguez]

Yeah, I came to the same conclusion in a1843d6 commit's message.

I wanted to provide a fix but you know, too many things, and renaming the method on our side made sense anyways.