From: Clifford Heath <no@...>
Date: 2008-07-14T12:55:44+09:00
Subject: Re: Treetop positive lookahead problem

Tom Aadland wrote:
> Thank you for responding and I'm sorry I didn't provide more 
> information.

Not a problem, we're here to help :-)

> I thought that the 'foo' &'bar' rule meant that bar has to come togehter 
> with foo as in foobar and on top of that not really consume bar so that 
> the text_value returned would only be foo.

That's correct.

>  However the following test below runs fine with that in mind :)

Your test doesn't ask for that, it asks that foo should be followed
by a sequence of one or more of any character, starting with bar,
and it consumes all that input.

> What I want with this rule is: match foobar or even foobarsomething, but 
> return the textvalue without bar.

I believe the only problem with your original rule is that by default,
for a Treetop parse to succeed, all input must be consumed. If you
change to the following:

	parser = FooGrammarParser.new
	parser.consume_all_input = false
	res = parser.parse("foobar")

then the original rule ("foo" &"bar") will match and return "foo",
without the parse failing due to not having consumed all input.

You should always print "parser.failure_reason" when a parse fails,
it'll tell you how far the parser got and what input would have allowed
it to get further.

Also, a common trap for new Treetoppers is to forget that matching is
greedy. For that reason, you almost never want to say ".*" or ".+"
because those will consume all input and leave nothing for subsequent
rules to even look at. For for example, this rule will *always* fail:

	rule big_fail
	  'foo' .* 'bar'
	end

This must be rewritten, for example like this:

	rule ok
	  'foo (!'bar' .)* 'bar'
	end

Clifford Heath.