old-idea query: data sharing between parser and nodes

[Paul Madden]

I've tried to do my homework, and I see that Clifford has discussed this several times, most notably here

  https://groups.google.com/forum/?fromgroups#!topic/treetop-dev/ipn6AwTblnQ

where he mentions my exact need: attaching symbol-table references to tree nodes. My parser has a symbol-table stack, which I consult to make decisions in semantic predicates. But I need (or believe I need) to access the same symbol table during a post-parse translation task, so I'm currently (ab)using semantic predicates to attach the current symbol table to certain nodes I know will need it later. It'd be easier and cleaner to just do this for all nodes as they're constructed, but I haven't thought of a way to allow the node's constructor to access the parser's data structures.

I read with interest Clifford's comment in this thread:

  https://groups.google.com/forum/?fromgroups#!topic/treetop-dev/FJkgreiy4IY

But does this require modifying the Treetop code itself, to use InputProxy in place of the input string? I'd rather not do that, but maybe I don't understand the suggestion. Like the OP in the second thread there, I was previously using a class variable for data sharing, but don't want to do that, either.

I welcome any suggestions. I'm soooo late to the party, but I'd have voted for Clifford's proposed change if I'd been around. Any chance it might be revived?

thanks,

paul

[Clifford Heath]

On 22/05/2013, at 9:17 AM, Paul Madden <pma...@gmail.com> wrote:
> I read with interest Clifford's comment in this thread:
>
> https://groups.google.com/forum/?fromgroups#!topic/treetop-dev/FJkgreiy4IY
>
> But does this require modifying the Treetop code itself, to use InputProxy in place of the input string? I'd rather not do that, but maybe I don't understand the suggestion. Like the OP in the second thread there, I was previously using a class variable for data sharing, but don't want to do that, either.

You can see how I use it here:
<https://github.com/cjheath/activefacts/blob/master/lib/activefacts/cql/parser.rb>

Basically it's this:

class MyParser < TreetopGeneratedParser
class Context
...
end
class InputProxy
…
end
def parse(input, options = {})
input = InputProxy.new(input, context, self) unless input.respond_to?(:context)
super(input, options)
end
end

The "unless" is there in case a caller wanted to pass in a different kind of context.
as far as I know, that hasn't happened yet.

This makes it possible from anywhere in the parser (including in a sem-pred)
to say "input.context.whatever".

Clifford Heath.

[Paul Madden]

Thanks, Clifford -- I've applied this mechanism to my parser and it works perfectly.

paul