RE: 90/10 rule... source?

Me:


> I've wondered if there are any other* parser generators out there that can
> profile at the production level.
> * I say "other" because Meta-S does this.


John:


> [I think I've seen a profiling version of yacc, but it was a long time
> ago. Unless a compiler does a great deal of analysis and
> optimization, the lexer is usually the part of the program that eats
> up the most time. I don't ever recall a parser that took much of the
> overall runtime. -John]


I added profiling into Meta-S because the adaptive(k) algorithm used
to match an $-grammar against input can make the grammar behave
somewhat like a "program", and certain productions can be optimized if
one knows they are the productions that take the most cycles to accept
or fail to accept input. I haven't done any studies on whether or not
the 90/10 rule applies to sophisticated grammars, though.


It might be interesting to run the C++, Perl, and C# grammars (that
is, the "larger" grammars) against test input and determine if the
grammars tend to spend most of their time in so few of the rules.


--
Quinn Tyler Jackson


http://members.shaw.ca/qjackson/

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