Often an action accompanies a production rule. It contains bf(C++) code that is executed once that production rule is recognized. The task of most actions is to compute a semantic value for the nonterminal recognized by the production rule, often using the semantic values made available by components of the production rule. An action consists of bf(C++) statements surrounded by braces, much like a compound statement in bf(C++). Most rules have just one action at the end of the rule, following all the components. Actions in the middle of a rule are tricky and should be used only for special purposes (see section ref(MIDACTIONS)). The components of production rules are numbered, the first component having number 1. E.g., in a production rule verb( nonterm: first second third ) tt(first) is component #1, tt(second) is component #2, tt(third) is component #3. bf(C++) code in action blocks may refer to semantic values of these components using em(dollar-notations), where tt($i) refers to the semantic value of the i^sups(th) component. Likewise, the semantic value of the rule's nonterminal is represented by tt($$). Here is a typical example: verb( exp: ... | exp '+' exp { $$ = $1 + $3; } | ... ) This rule constructs tt(exp) from two exp nonterminals connected by a plus-sign token. In the action, tt($1) and tt($3) represent the semantic values of, respectively, the first (left-hand side) tt(exp) component, and the second (right-hand side) tt(exp) component. The sum is assigned to tt($$), which becomes the semantic value of the tt(exp) nonterminal represented by the production rule. Depending on the specification of the tt(default-actions) option/directive (cf. section ref(DEFAULTACTION)) b() may supply non-empty production rules with default action blocks containing the statement tt($$ = $1): the semantic value of the first component of the production rule is returned as the nonterminal's semantic value. Of course, the default action only is valid if the two data types match. Empty production rules are not provided with default action blocks. Negative dollar indices (e.g., $-1) are allowed, and refer to semantic values of elements in rules before the component naming the current rule's nonterminal. This is a risky practice, and you should use it only when you know what you're doing. Here is a situation where you can use this reliably: verb( vardef: type varlist ';' ; varlist: varlist ',' variable { defineVar($-1, $3); } | variable { defineVar($-1, $1); } ; ) As long as tt(varlist) is only used in the tt(vardef) rule tt(varlist) can be sure that the type of the variable is available as the semantic value of the component immediately before (hence: $-1) the tt(varlist) component in the tt(vardef) rule. See also section ref(SPECIAL). In addition to the dollar-notations shown here, b() supports several more dollar-notations. The next three subsections describe the dollar-notations that are available after specifying, respectively, the tt(%stype, %union) or tt(%polymorphic) directives.