+ builder: emit implicit hash passed to a method call as kwargs

[iliabylich]

Closes #761.

[iliabylich]

@marcandre Could you take a look, too, please? (can't find you in "Reviewers" dropdown)

[marcandre]

The implementation surprises me. It was too difficult to emit the right node type from the get go?

[marcandre]

^hash^kwargs

[iliabylich]

It was too difficult to emit the right node type from the get go?

Yes and no, it's relatively simple. We rewrite either the last arg (if it's a hash) or the 2nd last (if it's a hash AND the last argument is a block). Rules are not trivial, and so the implementation takes 5 lines. The rest of the diff is about docs/tests.

(just in case) If you are talking about rewriting it in the associate method (that constructs a hash node) that's simply incorrect, not all hashes are kwargs. And there's no node for parameters (that'd be very helpful here)

[marcandre]

I meant that looking in ruby30.y there are 4 definitions that call associate; I was expecting the calls in primary and aref_args to remain unchanged, and those in opt_call_args and in call_args to change to associate_kwargs (or whatever you'd want to call the new method). Would that not work?

[iliabylich]

Yes, I had a similar idea, but there's a single exception - indexasgn:

             lhs: ...
                | primary_value tLBRACK2 opt_call_args rbracket
                    {
                      result = @builder.index_asgn(val[0], val[1], val[2], val[3])
                    }

Here opt_call_args can't be kwargs because this foo[bar: 42] = baz doesn't take kwargs at all.

Also by looking deeper at the grammar I've realized that it would affect super(foo: 42), and we need to modify it too 😄 . I'll push an update

[iliabylich]

Or maybe it's safer to rewrite it at the very beginning and un-rewrite in indexasgn. Yes, it seems to be the only exception, all other branches end up doing call_method/super/yield

[iliabylich]

I give up on moving it to the grammar level. On 1.8 (huh) aref_args rule is a part of indexing (that does support kwargs) and regular array literals at the same time (that don't support it). Making 1.8 special would require us to rewrite a ton of tests to separate 1.8/SINCE_1_9 examples.

Added kwargs to super as mentioned above, going to merge it as is, feel free to propose better ideas.