Re: recursive parsers and FOLLOW sets

KB Sriram <kbs@best.com> writes:
>When generating recursive descent parsers from LL(1) grammars, are
>FOLLOW sets needed for anything other than detecting syntax errors
>"quicker" and error repair/recovery?


I absolutely agree. Everyone who ever wrote a recursive descent parser
by hand probably didn't consider the follow set at all. I never saw
any reasons to use it.


In my opinion parser-generator theory is too heavily mixed with
grammar theory. Grammars are about GENERATING a text, while parsers
for RECOGNISING it. The goals are entirely different.


Advantages of leaving out FOLLOW sets are:
- the 'dangling-else' problem simply disappears
- tables become much shorter
- the grammar is much simpeler to write


Off course, there is one drawback: every grammar rule can have at most
one (possibly) empty production. But I don't see any problems in that.


By the way, for my own programming language I will build a custom
parser-generator shortly. It won't use FOLLOW sets. It will also allow
the programmer to extend the grammar in the programs' source. Is
anyone interested in co-operating on this?


Regards,


Mauk van der Laan
--

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