04-Nov-2013 12:28, Robert Schadek пишет:

On 11/04/2013 06:48 AM, Philippe Sigaud wrote:

On Sun, Nov 3, 2013 at 7:08 PM, Timothee Cour

<thelastmammoth@gmail.com <mailto:thelastmammoth@gmail.com>> wrote:


    On Sun, Nov 3, 2013 at 1:13 AM, Philippe Sigaud

    <philippe.sigaud@gmail.com <mailto:philippe.sigaud@gmail.com>>wrote:

        My current plan is to write different engines, and letting
        either the user select them at compile-time, or to have the
        parser decide which one to use, depending on the grammar. I'm
        pretty sure the 'Type 3' parts of a grammar (regular
        expressions) could be bone by using std.regex, for example.


    even lexing can't be done with regex, eg nesting comments : /+ ... +/
    Also, although it may seem cleaner at first to combine lexing and
    parsing in 1 big grammar (as done in pegged), it usually is faster
    do feed a (separate) lexer output into parser.


Lexing, yes. I was imprecise: even in a context-free grammar, some
rules are regular and could use std.regex (the ct part) as the
underlying engine, just for that rule.

Lexing can not be done with regex. Think myArray[1. ] ! What is next a
dot or a number.

You could use lookahead extension ;)

In general I would not recommend ATM to use std.regex for performance-critical lexer if only because it wasn't tuned for such a use case yet.

I have plans for extending std.regex capabilities in many directions, lexing being one important use case.

--
Dmitry Olshansky