[antlr-interest] single-pass pattern matching "for free"?

[Sohail Somani]

On Sun, 2006-01-15 at 11:39 -0800, Terence Parr wrote:
> On Jan 14, 2006, at 8:09 PM, Sohail Somani wrote:
> > After some more thought, doesn't it seem like just a more concise  
> > way to
> > specify tree parsing? What would the differences be?
> 
> And that is the most important question of all!  The answer is:
> 
> 1. just listing a bunch of patterns is easier until you get a huge  
> number of them, when it becomes hard to find bugs in rule applications.

Yes, I remember having to use Prolog. Which was the coolest thing, but
you really had to consciously keep the number of rules low.

> 2. patterns have no guarantee of coverage whereas a tree grammar does.

What does this mean?

> 3. patterns are floating in space w/o context; a grammar knows that  
> expressions cannot occur outside of a method or var  
> initialization...patterns could specify context, however, but it is  
> not as clear as a grammar.  want an action to execute when you see  
> the '}' of a function, just stick an action after the '}' in the  
> appropriate spot in the grammar.

This is probably the biggest reason. I guess rule-based translation
makes sense for simple translations but I can't see it working for full
translations.

> The translation of a pattern -> replacement system would be trivial  
> for a simple system; just translate to ANTLR.  Further, to separate  
> the view from the controller, one could use templates on the right:
> 
> "<a:expr>+<b:expr>" -> add(left=a,right=b)

or the equivalent in ST:

add(lexpr,rexpr)::<<
(<lexpr>) + (<rexpr>)
>>

Still haven't figured out how to get rid of 
((((((((5))))))) + ((((((3)))))))) but it would probably require a
change in the grammar rules, is all.

I think this whole thread will converge to say: "Hey wait, what we got
is the best way".

> or some such just like an antlr grammar.  The idea would be really  
> groovy.  You give me a grammar and then a list of patterns and I  
> generate a combined new grammar that does exactly what you want,  
> yanking in all the required tokens/rules.  Now where is a grad  
> student when I need some cheap smart labor! ;)  Actually, that would  
> make a damn fine paper and system.

Academia... shudder... (its ok, they feel the same way about me :D)