Re: ruby-mide, SMIE and token priority

[Dmitry Gutov]

On 07.11.2013 06:30, Stefan Monnier wrote:
> > bar.foo(tee) do
> >    bar
> > end
>
> What is the AST corresponding to this code?
> I mean, is it a method call on "bar" with method "foo" and 2 arguments
> (tee and do...end)?  I.e. (. bar foo tee (do-end bar))?

More or less, yes. "bar" could also be a method, but the structure would 
be the same.

> > The problem seems to be that "." is considered the parent token of
> > "do".  Probably because the left priority of "do" is not a number.
>
> SMIE parses it as (. bar (call foo tee (do-end bar))).
>
> > It would be more natural if the parent of "do" was ";" on the preceding
> > line, instead.
>
> You mean parse it as (call (. bar foo tee) (do-end bar))?

Guess so. The argument in favor of doing it this way is the grammar is 
already a bit skewed toward indentation (for example, " @ " has higher 
priority than "."), but that's likely something to be fixed, too.

> > Is it at all possible to change the grammar this way?
>
> You'd probably have to use a trick similar to the " @ " used on the
> space between the method name and the multiple-args.

Ah, okay. Sounds not very efficient, performance-wise.

> > Or should we just handle this as a special case in the `ruby-smie-rules',
> > and in case of "do", instead of delegating to the parent, skip the parents
> > until we find ";"?
>
> Depends: the rule of thumb is that if you have to choose between "make
> the grammar match the AST" or "make the grammar match the indentation",
> then better choose "match the AST" and then adjust the indentation via
> the smie-rules.
>
> So if the indentation you want is indeed not faithful to the AST, then
> indeed it's better to fix it in smie-rules.

Let's try this way, then.