Skip specs failing on TruffleRuby

[andrykonchin]

Some pattern matching specs are failing on the recent TruffleRuby release 24.0. We will investigate the issues - some of them are known. So the failing specs are temporary disabled on TruffleRuby.

[andrykonchin]

I suppose it's a bug in https://github.com/ruby/ruby/blob/master/test/ruby/test_pattern_matching.rb that was hidden - C.keys returned keys captured in some previous test case, that by chance matches C.new({}) and captures the nil value.

[palkan]

Good catch! Yeah, having case {} in and checking for C.keys doesn't make any sense here

[andrykonchin]

The fix in the Ruby repo - ruby/ruby#10757

[palkan]

Thanks!

One thing: could you please add links to the corresponding TruffleRuby repo issues where we skip it? (If any)

I check them from time to time to see if we can unskip the tests back.

[palkan]

Good catch! Yeah, having case {} in and checking for C.keys doesn't make any sense here

[andrykonchin]

There are no issues for not supported pattern matching edge cases. But failing ruby/spec and MRI tests are tracked in plain text files:

• https://github.com/oracle/truffleruby/blob/master/spec/tags/language/pattern_matching_tags.txt
• https://github.com/oracle/truffleruby/blob/master/test/mri/excludes/TestPatternMatching.rb

Would you like to have these links added (for every skipped test case)?

[andrykonchin]

BTW this test case differs a bit from the one in the ruby/spec repository. Not sure if it's done intentionally. Here the test relies on a way how CRuby accesses array elements - with #[] method. The ruby/spec's current version doesn't redefine the #[] method:

    it "accepts a subclass of Array from #deconstruct" do
      obj = Object.new
      def obj.deconstruct
        Class.new(Array).new([0, 1])
      end

      eval(<<~RUBY).should == true
        case obj
        in [1, 2]
          false
        in [0, 1]
          true
        end
      RUBY
    end

This version passes on TruffleRuby but the version with overridden #[] - doesn't.

[palkan]

Looks like the change to ruby/spec was made for TruffleRuby: ruby/spec@0e04268

Unfortunately, there are no additional details on that change.

MRI uses #[] (or opt_aref) to access elements returned by #deconstruct. So, you can do smth like this:

class A < Array
  def [](i) = at(-i-1)
end

a = A.new([1, 2])

case a
in [2, 1]
  true
end #=> true

I don't know if it's an intended feature or not; @k-tsj could you please take a look?

[k-tsj]

@palkan
Yes, it's intentional. If #[] is redefined, it should be called,.

[palkan]

@k-tsj Thank you!

@eregon ^ That's what I mentioned yesterday.

[eregon]

I see. I'm not sure it's a good idea to have that behavior in Ruby, maybe something to discuss on ruby-core.
Subclassing Array and redefining [] on its own is already clearly a bad idea, so it seems a niche edge case no one should use.

[palkan]

Would you like to have these links added (for every skipped test case)?

Let's add the links to the top of the spec file (ruby_pattern_matching_spec.rb).

[andrykonchin]

Done

[andrykonchin]

Is there anything that should be done in this PR to have it approved?

[palkan]

@andrykonchin Sorry, missed latest updates. Looks good, thanks!