|2 Yacc Qs firstname.lastname@example.org (1997-03-22)|
|Re: 2 Yacc Qs Bronikov@ic.net (Dmitri Bronnikov) (1997-03-23)|
|Re: 2 Yacc Qs email@example.com (Brian Bliss) (1997-03-27)|
|From:||Dmitri Bronnikov <Bronikov@ic.net>|
|Date:||23 Mar 1997 23:25:14 -0500|
Barton Jaques wrote:
> I am building an expression grammar and having problems with associativity
> and shift/reduce conflicts. The language I am emulating permits a statement
> like this:
> set y to items 1 thru 2 of x as string
> which should be read as: "set y to (items 1 thru 2 of x) as string"
> and not as: "set y to items 1 thru 2 of (x as string)"
> My grammar always picks the 2nd path. The relevant rules look like this:
> SET IDENT TO expression
> | IDENT
> | expression AS class
> | object_container of expression
> | object_container of object_specifier
> | class expression THRU expression
Your problem reminds me the if-then-else shift/reduce conflict, but
unlike the latter, shift is not the action you'd like the parser to take
if such conflict occurred.
It is not quite clear from the example whether you want leftmost
derivation, rightmost derivation or something in between.
Consider this example:
"SomeContainer OF SomeIdentifier AS class1 AS class2 AS class4 ... AS
It's valid to associate any number (0..N) of AS qualifiers with
SomeIdentifier and leave the rest as qualifiers of the expression
beginning with SomeContainer.
Things are getting more complicated if we replace SomeIndetifier with a
more sophisticated expression.
If expression following either "OF" or "THRU" can be legitimately
qualified by an "AS class" in your language, then it is a real ambiguity
and it might be reasonable to require programmer to use parenthesis to
If your language rules out "AS class" qualifiers in association with
expressions following "OF"/"THRU", then the conflict can be resolved as
| expression AS class
| object_container of as_less_expression
| object_container of object_specifier
| class expression THRU as_less_expression
Return to the
Search the comp.compilers archives again.