Re: %destructor can leak the lookahead

[Paul Eggert]

Akim Demaille <address@hidden> writes:

> "what could it really mean, and how could a single typing fit a
> token such as error which is used with many different types".

Well, here we are veering off to talking about a possible extension,
but I can see where it might be useful for a rule to be able to
distinguish between error tokens that result from calls to a
YYERROR-like macro, and those resulting from errors generated by the
Bison-generated parser itself.  The former could specify arguments
giving the semantic value for the error token, and the latter could
generate a default semantic value.  That way, a rule could use the
semantic value of the error token to decide whether the error was
generated by Bison or by user code.  Admittedly this is of limited
utility, and it's probably not worth pursuing, but anyway that's what
I was thinking of.