Re: Flex speedup

My experience with scanner generates is mixed. In general, I have
found that for production work a hand written scanner will run
significantly faster.


If you are using flex and you want better performance I would
try to merge rules. Let me illustrate what I mean. Often you might
have one rule for identifiers, and one rule for each reserved word.
The assumption here, is that identifiers can not be reserved words. I
would merge those rules into one rule, and use some type of hashing
scheme. By the way, since the number of reserved words is fixed, I
have found that closed hashing will work better then open hashing in
this case.


I might be able to provide additional information, if I knew
more about the language you are tring to process. I hope this helps.




Robert Sherry




Alan Morgan wrote:
>
> I'm trying to squeeze all the performance I can out of flex (without
> resorting to the "don't use flex" strategy").

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