Re: Supporting multiple input syntaxes

> Elijah Stone <elr...@elronnd.net> wrote:
> I did a C parser, it was not hard at all. I in C (like in standard
> Pascal) there are conflicts, but that conflicts can be resolved
> easily using semantic info. Alternativly, for C one can use 2
> token lookahead.


I'm not sure whether I captured the full context of your statement, but, if I
did, I don't think it's 100% correct:


- In regards to lookahead of 2:
This isn't enough to disambiguate, e.g., between a cast-expression in 6.5.4,
`( type-name ) cast-expression`, and a compound literal in 6.5.2, `( type-name
) { initializer-list }`.


- In regards to using semantic info:
Yes, with semantic info you can disambiguate things like `x * y;`, so I'd say
that, from a pragmatic/practical standpoint, this affirmation is right.
However, from a more theoretical perspective, a parser (thinking of it a
program that "simply" validates a sentence based on a grammar), isn't expected
— arguably — to rely on anything else other than syntax. Whether or not
the theoretical aspect of it is relevant, depends on the application of the
parser, I guess. For instance, for the implementation of static analysis tool,
not depending, as much as possible, on semantic information to guide parsing
is an advantage.


This is a table (only for expressions) that I recently put up when rewriting
the C parser of my project:
https://docs.google.com/spreadsheets/d/1oGjtFaqLzSoBEp2aGNgHrbEHxSi4Ijv57mXMP
ymZEcQ/edit?usp=sharing


--
Leandro T. C. Melo

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