Up: Notes

Rewriting of precedenced rules to embed the precedence into the grammar

Concrete Implementation

Precedences

• Precedences are integers <= 0.
• The tighest precedence is 0.
• The loosest precedence is some K <= 0 (dependent on the LHS)

Rules:

looser(x)  = x-1 for x > K,	x in [K+1, 0]
           = K   for x = K

tighter(x) = x+1 for x < 0,	x in [K, -1]
           = 0   for x = 0

tighest    = 0
loosest    = K

Rewriting

In what follows, E will represent the LHS symbol of the precedenced rule, i.e. the operator. We will need to have a unique symbol for E at each level of precedence. We will represent the symbol for E at level x as E[x].

Top level

The top level rule for the precedenced rule will be a rule whose LHS is the precedenced rule's LHS, and whose RHS is the symbol for E at the lowest precedence (K). That is, the top level rule will be

E ::= E[K]

The "spine"

In many expressions, there is no logic at a given precedence level, so that we need to simply "fall through" to the next-tighter level of precedence. Therefore, for every precedence x, such that x is less than tightest, we add the rule

E[x] ::= E[tighter(x)], x < 0

For example, if the precedence ran from loosest == -4 to tightest == 0, we would add the BNF rules

E[-4] ::= E[-3]
E[-3] ::= E[-2]
E[-2] ::= E[-1]
E[-1] ::= E[0]

Left association

We now can deal with the RHS alternatives in the work list.

Let curr be the precedence level of the RHS alternative.

If a RHS alternative has left association (the default) we rewrite the RHS, replacing all occurrences of E.

We replace the leftmost occurrence of E with E[curr], and the others with E[tighter(curr)]. We then add a rule with the rewritten RHS, and E[curr] as the LHS.

For example, if the RHS alternative is

E + E

we add the BNF rule

E[curr] ::= E[curr] + E[tighter(curr)]

Note that E may not occur on the RHS, in which case no RHS replacements will be necessary.

Right association

Right association is handled in a way that is symmetric with left association.

Again, let curr be the precedence of the RHS alternative.

If a RHS alternative has right association, we rewrite the RHS, replacing all occurrences of E.

We replace the rightmost occurrence of E with E[curr], and the others with E[tighter(curr)]. We then add a rule with the rewritten RHS, and E[curr] as the LHS.

For example, if the RHS alternative is

E ** E

we add the BNF rule

E[curr] ::= E[tighter(curr)] ** E[curr]

Group association

The archetypal case for group association is the parenthesis operator.

Intuitively, group association allows any expression to occur inside another "surrounding" expression. In the case of the parentheses, of course, they actually surround their contents lexically.

Let curr be the precedence of the RHS alternative.

If a RHS alternative has group association, we rewrite the RHS, replacing all occurrences of E with E[K], We then add a rule with the rewritten RHS, and E[curr] as the LHS.

For example, if the RHS alternative is

( E )

we add the BNF rule

E[curr] ::= ( E[K] )