I have this grammar (I reduced it to 42 lines) that breaks some things in the parser.

module TestSdf

exports
context-free start-symbols
	Start
context-free syntax
	K -> Start {cons("Start")}

context-free priorities
{
	Int -> K
	Id -> K
} > {
	K -> Int
	K -> Id
}
context-free priorities
{
	K -> Int
	K -> Id
} > {
	Int -> K
	Id -> K
}
context-free syntax
	"r" Int -> K {cons("R")}
	"rw" -> K {cons("RW")}

	ID		-> Id	{cons("Id")}
lexical syntax
	[a-zA-Z][a-zA-Z0-9]*	-> ID
	"rw" -> ID {reject}
	"r" -> ID {reject}

lexical restrictions
	ID  -/- [a-zA-Z0-9]

context-free priorities
{
} > { non-assoc:
	K "=>" K	-> K {non-assoc, cons("Rewrite")}
}

Input: "rw"

Result: Start(amb([RW(), Id(“rw”)]))

I would expect to have only the RW() production, but I also get the Id. Somehow the parser is tricked by the poduction “r” Int -> K and the subsorts K->Int and K->Id…

What is even more weird is that if I remove the last context-free priorities block, it works again. Something is not quite correct somewhere.

I tested this example with the old sglr written in C, and it works fine. So it is only related to JSGLR.

Tested with Spoofax 1.0.9.0-r25288, and sglr 3.16

Submitted by Radu Mereuta on 18 April 2013 at 13:59