Re: Bison-like source locations in LALR-parser

[Ludovic Courtès]

Hi,

Jan Nieuwenhuizen <address@hidden> skribis:

> Ludovic Courtès writes:
>
>> I don’t think this is needed.  Lexers are expected to use
>> ‘make-lexical-token’ and ‘make-source-location’ from (system base lalr)
>> to preserve source location information.
>
> I hope you're right...and that's what I tried, but I didn't get it
> working.  Possibly I need to cook-up a small example.
>
> What I found was that, yes the LEXER has all it needs, but once you
> get to construction rules in the parser; say
>
>    (interface-spec
>     (interface Identifier lbrace type-list event-list optional-behaviour 
> rbrace) : `(,$1 ,$2 ,$4 ,$5 ,$6))
>
> all it sees is the values, i.e. strings of the tokens that the lexer
> sees.  The lexer cannot hand tokens to the parser, AFAICS.

Here’s an example:

http://git.savannah.gnu.org/cgit/guile-rpc.git/tree/modules/rpc/compiler/lexer.l
http://git.savannah.gnu.org/cgit/guile-rpc.git/tree/modules/Makefile.am#n58
http://git.savannah.gnu.org/cgit/guile-rpc.git/tree/modules/rpc/compiler/parser.scm#n101

Note that actually, the ‘location’ produced by Makefile.am does:

  (make-lexical-token type (make-source-location ...)
                      (append value (list (vector line column))))

So the line/column info is also stored alongside the token’s value, and
then access from parser.scm with those cadr calls.

>> Besides, note that lalr.upstream.scm is a copy of the upstream lalr-scm,
>> hosted at <http://code.google.com/p/lalr-scm/>.
>
> Yes...that's looks quit dead.  Isn't it?  I can try though...

Well, that’s because it’s “finished.”  :-)

But yeah, it’s worth trying.  Dominique has been responsive and helpful
in the past.

Thanks,
Ludo’.