Re: Problems with bison creating a GLR parser

[Paul Hilfinger]

OK.  It's clear what needs to be done here: embedded actions make what
are in effect references of the form $0, $-1, ..., since they are
translated rules with empty RHSs.  The working test case does not 
go into GLR mode, and so does things exactly like lalr.c---those
"non-positive" semantic values are right there on the stack in the
right place.  In GLR mode, the semantic values are available through
what is essentially a linked list, whose members I copy as needed into
an array (so that the action code can do the same thing regardless of
the parser's current mode).  I will have to arrange to keep track of
the number of symbols that need to be copied by scanning the actions,
or something like that.  I'll let you know.

Paul Hilfinger