Dan Sugalski wrote: > > I'm not so sure about that. (Not that it won't be replaced, but that > it needs the grammar engine) I'm pretty sure that grammars as Larry's > defined 'em are recursive-descent, and if that's true then I've this > really nasty feeling we're going to find it insufficiently fast to > parse perl. Could be wrong, but even if we pick up two orders of > magnitude in speed over Parse::RecDescent things'll still be > significantly more sluggish than the current perl 5 parser. > > (OTOH, if the grammar engine's not required to be recursive-descent > that makes things rather faster, which is fine with me :)
That's a question with multiple layers of answers. :) The abstraction we need to meet is: a) everything that can be done in the parser that ships with Perl 6 can be done in Perl 6; and b) it is possible to change the way Perl 6 is parsed within Perl 6 code (by loading a custom grammar module, for example). But, there's some flexibility in how we satisfy that abstraction. It might be reasonable to have "fast parsing" that merely comes close to simulating recursive-descent, as long as we also provide a (slower) full recursive-descent parser that produces the same PIR output. Certainly some things won't be recursive-descent anyway (operator precedence). Also, it may be a little too early to decide that one method of parsing will be fast or slow. IIRC we nearly ditched CPS because it was too slow, then found a way to speed it up. Allison
