On Oct 30, 2007, at 9:18 AM, Oliver Bandel wrote:
Hi all,
I have some problem with precedence declaration in OCaml parser.
If I what to say exponential(ATOB) is prior to *(STAR) and / (DIVIDE),
* and / are prior to +(PLUS) and -(MINUS),
I wrote the following in the parser:
/***** Precedence Rules *****/
...
%left PLUS MINUS
%left STAR DIVIDE
%left ATOB
...
But I still have the following problems:
(1) It appears that the parser
reads "test = 2^2 + 7;" as "test = 2^9" instead of "test = 4+7", which
would follow the conventional order of operations.
(2)It also interprets "test = (1^2)/3 + 1;" as "test = (1 ^ 2
/ (3 + 1));"
Can any one help me to see why it happens? Why the precedence rules
doesn't work?
[...]
Precedences also can be created by sophisticated
organization of the grammar rules.
But I want to avoid this.
So, if your grammar rules may have a contradictory
meaning, then your parser works not as expected.
In general I would use the precedence-declarations only,
when you run into parser conflicts, if you don't use them.
When developing a grammr, I would recommend, first to start
with the grammar rules, and add precedence-/associatitivity-
declarations, at the end, if really necessary.
What is the rest of your mly-file?
A complete example would be helpful.
Here is part of my .mly file:
Beside the precedence issue, everything works fine.
%{
open Past
open Parsing
open ParseError
let pi = 4.0 *. atan 1.0;;
let get_range n = {
pos_start = Parsing.rhs_start_pos n;
pos_end = Parsing.rhs_end_pos n;
}
let unclosed opening_name opening_num closing_name closing_num =
raise(Error(Unclosed(get_range opening_num, opening_name,
get_range closing_num, closing_name)))
%}
/* List of all tokens the lexer can output */
...
%token PLUS
%token STAR
%token MINUS
%token DIVIDE
%token AND
%token OR
...
%token ATOB /* A^B: exponential */
...
/***** Precedence Rules *****/
%left PLUS MINUS
%left STAR DIVIDE
%left ATOB
%nonassoc prec_unary_minus
%start prog
%type <Past.pprog> prog
%%
/* Rules for parsing. The parsing rules should generally be in a */
/* one-to-one correspondence with the BNF */
/* type prog = Prog of consDeclare list * varDeclare list * inpDeclare list * sysDeclare list */
prog:
exp:
LPAREN exp RPAREN { $2 }
| LPAREN exp error { unclosed "(" 1 ")" 3 }
| exp PLUS exp { Add($1, $3) }
| MINUS exp { Sub(Value(VFloat(0.0)), $2) }
| exp MINUS exp { Sub($1, $3) }
| exp DIVIDE exp { Divide($1, $3) }
| exp STAR exp { Mult($1, $3) }
| exp ATOB exp { Atob($1, $3) }
| value PLUS exp { Add(Value($1), $3) }
| value MINUS exp { Sub(Value($1), $3) }
| value DIVIDE exp { Divide(Value($1), $3) }
| value STAR exp { Mult(Value($1), $3) }
| value ATOB exp { Atob(Value($1), $3) }
...
| IDENT { Id($1) }
| value { Value($1) }
;
...