On Jun 9, 2007, at 1:28 AM, Roman Redziejowski wrote:
On Jun 8 13:44:15 EDT 2007 Terence Parr wrote:
Despite the disadvantages, I still favor the separate lexer as
it's faster.
I fully agree. As I pointed out in my paper, PEG is not good at
specifying
lexical structures. But, if I need an ad hoc parser, I would prefer
to skip
a separate lexer. So, my question is: how could one modify PEG
to facilitate this, and at the same time have a decent speed?
The approach I've taken in ANTLR is to optimize PEG with LL(*) to
reduce number of backtracking decisions. Analogous to GLR's
optimization over Earley's algorithm: only pursue multiple paths upon
LR nondeterminism.
ANTLR lexers can backtrack if they need to but generally use a DFA
predictor and normal LL lexing.
BTW: I've rewritten Jean Bovet's and my ANTLRWorks paper to include a
description of LL(*) and relationship to other strategies...will
resubmit to SP&E shortly. Anybody care to review a draft for me?
Ter
_______________________________________________
PEG mailing list
PEG@lists.csail.mit.edu
https://lists.csail.mit.edu/mailman/listinfo/peg
