Rewriting

Changes to ASTs can be accomplished by both procedural methods coded in PARLANSE and source-to-source tree transformations coded as rewrite rules using surface-syntax conditioned by any extracted program facts, using DMS's Rule Specification Language (RSL). The rewrite rule engine supporting RSL handles associative and commutative rules. A rewrite rule for C to replace a complex condition by the ?: operator be written as:

   rule simplify_conditional_assignment(v:left_hand_side,e1:expression,e2:expression,e3:expression)
        :statement->statement
   =  " if (\e1) \v=\e2; else \v=e3; " 
   -> " \v=\e1?\e2:\e3; "
   if no_side_effects(v);

Rewrite rules have names, e.g. simplify_conditional_assignment. Each rule has a "match this" and "replace by that" pattern pair separated by ->, in our example, on separate lines for readability. The patterns must correspond to language syntax categories; in this case, both patterns must be of syntax category statement also separated in sympathy with the patterns by ->. Target language (e.g., C) surface syntax is coded inside meta-quotes ", to separate rewrite-rule syntax from that of the target language. Backslashes inside meta-quotes represent domain escapes, to indicate pattern meta variables (e.g., \v, \e1, \e2) that match any language construct corresponding to the metavariable declaration in the signature line, e.g., e1 must be of syntactic category: (any) expression. If a metavariable is mentioned multiple times in the match pattern, it must match to identical subtrees; the same identically shaped v must occur in both assignments in the match pattern in this example. Metavariables in the replace pattern are replaced by the corresponding matches from the left side. A conditional clause if provides an additional condition that must be met for the rule to apply, e.g., that the matched metavariable v, being an arbitrary left-hand side, must not have a side effect (e.g., cannot be of the form of a[i++]; the no_side_effects predicate is defined by an analyzer built with other DMS mechanisms).

Achieving a complex transformation on code is accomplished by providing a number of rules that cooperate to achieve the desired effect. The ruleset is focused on portions of the program by metaprograms coded in PARLANSE.

A complete example of a language definition and source-to-source transformation rules defined and applied is shown using high school algebra and a bit of calculus as a domain-specific language.