@yui-knk you have a few regressions in spec:ruby:fast involving this. They are sometimes a bit pedantic on Ruby dispatch but at least one error looks like it may be recursing a lot.

@enebo I fixed regressions. The rest of failures are encoding related test cases, which also fail on the head of jruby-9.1:

• TestTranscode#test_windows_1255
• Integer#chr without argument when Encoding.default_internal is not nil and self is greater than 255 raises RangeError if self is invalid as a codepoint in the default internal encoding
• Integer#chr with an encoding argument raises RangeError if self is invalid as a codepoint in the specified encoding FAILED

@yui-knk jcodings was updated yesterday and is the fallout. New updated data for encoding support. The regressions need to be examined still.

not sure about these changes - I think we might have clashed on blindly following MRI
that f_to_r with the "Fix spec:ruby:fast" msg doesn not tell much and is not part of the PR's desc

+ there's a Rational regression (test failing) on master, @yui-knk not sure but might be relate to these?
or maybe do you recall if that f_to_r is necessary (a legit fix)?
(was actually doing the opposite on master - trying to avoid the dynamic call-site when not needed - for a rational instance having to need to go through to_r just seems like smt I would try to avoid 😈 )

@kares the line below basically explains that change: "MRI does not call to_r when a1 is not a float nor a string." @yui-knk seems to be following MRI logic exactly and this fixed an MRI test (unsurprisingly).

As for master, I have not checked but I half wonder if perhaps 2.4 changed behavior (and we should not have merged this fix to master)? His PR actually removes call to_r in that particular case for 2.3.x, but original f_to_r code always performs the dyncall so maybe 2.4.x+ requires always dyncalling now?

sorry, the first commit is nice and clear but the following isn't ... that is what I was hoping to understand why
anyway I will check it later whether its related to the failing test on master

The second commit is needed to fix

• "Numeric#denominator works with Numeric subclasses" in spec/ruby/core/numeric/denominator_spec.rb
• "Numeric#numerator works with Numeric subclasses" in spec/ruby/core/numeric/numerator_spec.rb

These specs expect "#to_r" is called when #denominator/#numerator is called for Numeric subclass.

By the way #5022 will fix failing test on master.