Allow for rules that don't create separate nodes

[victorandree]

It's often useful to define some basic rules that can be composited to form larger expressions:

digit = ~r"[0-9]"
number = digit+

However, there is not always any reason why each separate digit needs its own node, even if it's useful as a "building block" in the grammar. As such, it would be nice to be able to mark a rule as "non-node creating". The library simpleparse supports this by creating 'unreported productions' that would translate into something like this for parsimonious:

<digit> = ~r"[0-9]"
number = digit+

Then, any number node could simply be the number, and we don't get a bunch of (unneeded) nodes for each component – because the angle brackets signify that this rule should not be "reported" separately.

I suppose this could either be implemented while parsing the grammar – replacing any reference to such a rule with simply its contents (at a preprocessing stage), or as part of the parsing of the text...

[erikrose]

But then wouldn't you have to repeat the visitation code (e.g. int(...)) in the visit_* method for each rule that calls digit? Maybe there's a more compelling example I'm not forseeing; I could use some help.

If we do this, here's a spelling idea:

>thing< = some nodeless rule  # or think of it as a dumb, C-style "macro"

That's consistent with, or at least reminiscent, of the term-silencing syntax proposed in #29 (comment).

[erikrose]

I suppose suppressing node formation might not be as disastrous if you aren't using NodeVisitor. But I can't help thinking that, whatever you write to process the tree, you'll end up repeating yourself if you don't let digit be a node.

[keleshev]

If you take number from JSON grammar:

number = int frac? exp?
int = "-"? ((digit1to9 digits) / digit)
frac = "." digits
exp = e digits
digits = digit+
e = "e+" / "e-" / "e" / "E+" / "E-" / "E"

Here, you don't really care about int or frac or exp, you just want to call float(number), you don't need the rest of nodes.