>> I wish it had something like ml-yacc's automatic handling of syntax-errors
>> (which tries to insert/delete tokens in the input stream until the parse
>> succeeds).

> Do you mean instead of or in addition to the current mechanism? (I assume
> instead of.  I don't know if the two can be made to coexist.)

I don't know enough of the details to tell you, but I'd guess it'd have to
be "instead of", indeed.

> I am not particularly happy with Menhir's current error mechanism, which
> attempts to follow yacc's.  If you could explain to us why ML-Yacc's
> approach is superior, that would be great.

ML-Yacc's error recovery if fully automatic: you write your grammar without
worrying about error handling and the generated parser automatically signals
syntax errors (note the "s" at the end: it doesn't just stop at the first
syntax error) by mentioning that it had to insert a ";" at line L1 and
remove an ID at line L2.

For toy projects it's amazing: you get good error messages with 0 effort.

The main downside of ML-Yacc IIRC has to do with the fact that it requires
a fairly strong decoupling between the lexer and parser (you can't do tricks
like change the lexer state from a parser semantic action IIRC), but it may
be just a limitation of ML-Yacc rather than of the underlying technique.

See http://www.smlnj.org/doc/ML-Yacc/mlyacc001.html#toc3


        Stefan