Add support for picking fields with `@`

as discussed at https://github.com/pegjs/pegjs/issues/11,

foo = @bar _ @baz
bar = $"bar"i
baz = $"baz"i
_ = " "*

parse('barbaz') // returns [ 'bar', 'baz' ]

I use this syntax frequently when it’s available.

[hildjj]

PR #89 is pretty close at this point. Now it’s time to decide if we like it enough to land it.

I do like it; I use @ enough that i’ve gone out of my way in a couple of non-production projects to rely on odd versions of pegjs in the past. The implementation is adequate (if a little confusing because of the way the stack manipulation has to happen). Tests gave adequate coverage. I’ve prototyped ahead the few other places I might like us to use @ and am not concerned about overlaps from a grammar perspective.

Note that the example I added to the end of the documentation is:

list = head:word tail:(_ "," _ @word)* { return [head].concat(tail); }
word = $[a-z]i+
_ = [ \t]*

which shows off how @ can pluck from parens, which is its killer feature, imo.

[Mingun]

Another possible way of evolution – change how the rule rule itself is defined and thus make it impossible to create clash.

At least two options can be suggested:

• change the rule definition to something like this (the https://github.com/kevinmehall/rust-peg way):

  rule rule_name() = ...
  

  The empty brackets are not allowed in the current grammar, so no clash.
• the other way is just make the optional semicolon in the end of the rule definition required. Personally I think that is easest and right way to solve this problem, because modern JS rules also requires semicolons to make syntax consistent.