Inconsistency between CRuby and TruffleRuby with pattern matching

[Th3-M4jor]

When using pattern matching on a HashWithIndifferentAccess from ActiveSupport there appears to be a difference in behavior.

Given the below example, using ActiveSupport 7.2.1

require 'active_support/all'

test = HashWithIndifferentAccess.new('a' => true)

case test
in {a: b}
  b
else
  false
end

When run on CRuby 3.3.5 it will return true, while on TruffleRuby 24.2.0-dev-512427a6 it'll return false

This appears to work on CRuby because of three things:

• HashWithIndifferentAccess extends Hash
• Hash#deconstruct_keys returns self
• The resulting iseq calls .key? followed by .[] on the returned object

[eregon]

Thank you for the report.

This is because

truffleruby/src/main/java/org/truffleruby/parser/YARPPatternMatchingTranslator.java

Line 313 in 5aa5898

RubyNode valueOrUndefined = HashGetOrUndefinedNodeGen.create(key, readTemp);

uses the original Hash#[] definition and does not do a dynamic call to []/key?.
It's a lot more efficient to not call both of these methods.

I think this should be fixed in the semantics of pattern matching in CRuby, i.e. to not call methods dynamically as it's significantly more expensive and makes overall pattern matching slower.
(and the same thing for array pattern matching)
But I realize that's an uphill battle to change that.

Is this important in practice, does this get used in applications?

From https://github.com/rails/rails/blob/cca4db3db4a3a122ca3d9090f4cfd6754aed6487/activesupport/lib/active_support/hash_with_indifferent_access.rb#L388-L390 HashWithIndifferentAccess stores keys as Strings.
It seems Hash pattern matching does not currently accept String keys unfortunately (it's a SyntaxError).

I suppose one approach here would be to check if the receiver class is Hash and in that case use HashGetOrUndefinedNode, and if it's not then use key? + [].

[eregon]

We should fix this for compatibility.