Re: Why put type information into syntax?

> > {
> > typedef int foo;
> > foo(x); /* This is a variable declaration. */
> > bar(y); /* This is a function call.
> > }
>
> Would it be possible to deal with this by having the parser build
> trees that retain the ambiguity, and then resolve it during a separate
> semantic-analysis tree walk? ...
> [Sure, that's what Earley parsers do, but the performance hit can be
> phenomenal. -John]


Actually, GLR (generalized LR) parsing due to Tomita does just that.
The performance hit is minimal: this is achieved through optimal
sharing of the parse trees and other tricks.


See "Current Parsing Techniques in Software Renovation Considered
Harmful" by Mark van den Brand et. al
http://www.tinaa.com/harmful/parsing.html


This paper offers some performance numbers and includes a host of great
references (including a reference to a comparison with Earley's
algorithms).


I have plenty of other GLR references if you are interesting in the
implementation details, but this paper offers a great overview.


Marat.

Return to the comp.compilers page.
Search the comp.compilers archives again.