Remove :rasgn and :mrasgn types.

[marcandre]

See whitequark/parser#739

[marcandre]

@eregon Test runs hang on truffleruby (second time in a row). I stopped last run after 5 hours (I'm surprised github actions doesn't seem to have a cutoff?)

[eregon]

It's possible to specify a timeout for GitHub Actions:
https://docs.github.com/en/free-pro-team@latest/actions/reference/workflow-syntax-for-github-actions#jobsjob_idtimeout-minutes

I'll try to reproduce. It seems to only affect truffleruby-head, right?

[eregon]

It gets stuck in
https://github.com/rubocop-hq/rubocop-ast/blob/2788a1a0e129013f7e401a52b03a916b8dd0d495/lib/rubocop/ast/node_pattern/node.rb#L200-L204
(i.e., (0..Float::INFINITY).minmax takes forever)
Unfortunate that GitHub Actions doesn't send SIGINT or so, it would have shown that clearly.

TruffleRuby recently updated to 2.7, but didn't port that fix yet.

[eregon]

This is quite involved because I had to finish writing a spec for Range#minmax, synchronize specs across implementations and then actually fix it in TruffleRuby.

[eregon]

BTW, is there any reason not to simply use Range#begin/end here? It's very fragile to depend on RUBY_VERSION checks.

[marcandre]

Thanks for the investigation!

is there any reason not to simply use Range#begin/end here

I don't remember, and it looks like all the arity ranges are inclusive, so maybe not. Still, I'd rather write things properly; also I should probably be using refinements here. Does map(&:refined_method) work in truffleruby?

It's very fragile to depend on RUBY_VERSION checks.

Agreed, it's also not the best in general. I think I can use a different check like (0..4.2).minmax.last

[eregon]

Fixed in oracle/truffleruby@df78798, thanks for the ping.
So no need to change anything :)

I don't remember, and it looks like all the arity ranges are inclusive, so maybe not.

Exclusive ranges wouldn't work anyway, 2.7.2:

> (0...Float::INFINITY).minmax
TypeError: cannot exclude non Integer end value

Does map(&:refined_method) work in truffleruby?

It does. But implementation-wise, that's a proper nightmare (need to cache globally per execution context & per refinements), I'd recommend having an explicit block. (actually TruffleRuby has an inline cache for it now, but still a global cache in the case the Symbol wasn't always the same, that's the part that's unpretty about this construct).

[marcandre]

Fixed in oracle/truffleruby@df78798, thanks for the ping.
So no need to change anything :)

I don't remember, and it looks like all the arity ranges are inclusive, so maybe not.

Exclusive ranges wouldn't work anyway, 2.7.2:

Right. For parser's TreeRewriter, I break the integers into exclusive integer ranges and a last inifnite range (that I make inclusive because it doesn't work otherwise, but I think that's unfortunate): 0...15, 15...42, 42..Float::INFINITY. Works like a charm.

Does map(&:refined_method) work in truffleruby?

It does. But implementation-wise, that's a proper nightmare (need to cache globally per execution context & per refinements), I'd recommend having an explicit block.

Ok. Since TruffleRuby is fixed now, I could use a conditional refinement here to support MRI <= 2.6. Should be nicer.