Nesc-internals/AST

Objects

The abstract syntax tree (AST) is defined using a "pure data" (no methods) object-oriented style:

• a node type is defined by its name, super-type, and fields; there is a root type called node
• the actual fields of a node type are the union of its own fields and of those of its super-type
• all node types have a kind field that identifies the nodes actual type, allowing for up- and down-casts
• fields are either data fields (some piece of information, e.g. the name of an identifier) or tree fields, representing a child of the current node in the AST; this child relationship is used in the tree walker (AST_walk.h) and a few other places

Because C is not an object-oriented language, these node types are defined in a file called nodetypes.def, and a series of emacs-lisp scripts generate C types, macros and functions for manipulating the AST nodes. Specifically, if X is a node definition with fields f1, ..., fn and super-type Y then

• X and Y are C pointer types to AST_X, AST_Y structures respectively
• AST_X, AST_Y have an initial integer field called kind
• the other fields of AST_X are: the fields of AST_Y, followed by f1, ..., fn
• new_X, new_Y allocate nodes of type X and Y respectively; these constructors take as arguments a subset of the fields of X, Y (the choice is made on a per-field basis in nodetypes.def)
• CAST(T, e) does a checked cast of expression e to type T; T must be a super- or sub-type of the actual (runtime) type of e (a pointer to some node)
• CASTPTR(T, e) does the same as CAST(T, e) except that e must evaluate to a pointer to a pointer to some node (CASTSRPTR(T, e) is the same as CASTPTR(T, e))
• is_X(e) is true if the node e points to is an X or a subtype of X

Lists

The node type has a next field which points to another node. This is used to represent homogeneous lists of AST nodes, which are rather common (e.g. lists of declarations, lists of arguments to a function, etc, etc). Thus in a node a whose dynamic type is X, the runtime type of a->next must be X even though its static type is node.

The following functions are defined to simplify handling these lists: for every node type X:

• last_X(a) returns the last element in the list rooted at a
• X_length(a) returns the length of the list rooted at a
• X_reverse(a) reverses the list rooted at a
• scan_X (p, a) s iterates p over the list a, executing statement s at each list element
• X_merge(a, b) concatenates lists a and b

Other operations

AST nodes have parent and parentptr (pointer to the pointer to me in my parent) fields. These are not set automatically, instead a call to AST_set_parents(a) sets these fields in an AST rooted at a. Individual parent and parentptr fields can be set with set_parent and set_parent_list.

AST_print(a) prints the abstract syntax tree rooted at a (useful for debugging).

The AST_walk.h implements a visitor pattern for abstract syntax trees. You define the operation to perform on some node types (the operation applies to a node type and all its subtype, unless you override it on specific subtypes). The default action is just to visit AST children recursively.