PAST - Parrot abstract syntax tree


This file implements the various abstract syntax tree nodes for compiling programs in Parrot.

PAST Node types ^

PAST::Node ^

PAST::Node is the base class for all PAST nodes, and is derived from PCT::Node. A node has an array component to hold its children, and a hash component for its attributes. However, we tend to use accessor methods for accessing the node's attributes instead of accessing the hash directly.

Every PAST node inherits name, source, and pos attributes from PCT::Node. The name attribute is the node's name, if any, while source and pos are used to identify the location in the original source code for the node. The source and pos values are generally set by the node method inherited from PCT::Node.

Other node attributes are generally defined by subclasses of PAST::Node.


Accessor method -- sets/returns the return type for the invocant.


Accessor method -- sets/returns the arity (number of expected arguments) for the node.


Accessor method -- for named arguments, sets/returns the name to be associated with the argument.


Accessor method -- sets/returns the "flatten" flag on arguments.

PAST::Val ^

PAST::Val nodes represent constant values in the abstract syntax tree. The name attribute represents the value of the node.


Get/set the constant value for this node.

PAST::Var ^

PAST::Var nodes represent variables within the abstract syntax tree. The variable name (if any) is given as the node's name attribute.


Get/set the PAST::Var node's "scope" (i.e., how the variable is accessed or set). Allowable values include "package", "lexical", "parameter", and "keyed", representing HLL global, lexical, block parameter, and array/hash variables respectively.


Get/set the node's isdecl attribute (for lexical variables) to flag. A true value of isdecl indicates that the variable given by this node is to be created within the current lexical scope. Otherwise, the node refers to a lexical variable from an outer scope.


Get/set the lvalue attribute, which indicates whether this variable is being used in an lvalue context.


Get/set the variable's namespace attribute to the array of strings given by namespace. Useful only for variables with a scope of 'package'.


Get/set the node's slurpy attribute (for parameter variables) to flag. A true value of slurpy indicates that the parameter variable given by this node is to be created as a slurpy parameter (consuming all remaining arguments passed in).


If the variable needs to be instantiated, then type indicates either the type of the value to create for the node or (future implementation) a PAST tree to create the value.


For keyed nodes, type indicates the type of aggregate to create for the base if the base doesn't specify its own 'viviself' attribute.

PAST::Op ^

PAST::Op nodes represent the operations in an abstract syntax tree. The primary function of the node is given by its pasttype, secondary functions may be given by the node's name, pirop, or other attributes.


A PAST::Op node's pasttype determines the type of operation to be performed. Predefined values of pasttype are:

assign - Copy the value of the node's second child into the variable expression given by its first child.

bind - Bind the variable given by the node's first child to the value given by its second child.

if - Evaluate the first child; if the first child is true then evaluate the second child (if any) otherwise evaluate the third child (if any). If either the second or third child are missing, then they evaluate as the result of the first child.

unless - Same as 'if' above, but reverse the evaluation of the second and third children nodes.

while - Evaluate the first child, if the result is true then evaluate the second child and repeat.

until - Evaluate the first child, if the result is false then evaluate the second child and repeat.

for - Iterate over the first child. For each element, invoke the sub in the second child, passing the element as the only parameter.

call - Call a subroutine, passing the results of any child nodes as arguments. The subroutine to be called is given by the node's name attribute, if the node has no name attribute then the first child is assumed to evaluate to a callable sub.

pirop - Execute the named PIR opcode, passing the results of any children nodes as arguments.

inline - Execute the sequence of PIR statements given by the node's inline attribute (a string). See the inline method below for details.

callmethod - Invokes a method on an object, using children nodes as arguments. If the node has a name attribute, then the first child is the invocant and any remaining children are arguments. If the node doesn't have a name attribute, then the first child evaluates to the method to be called, the second child is the invocant, and the remaining children are arguments to the method call.

try - (preliminary) Execute the code given by the first child, and if any exceptions occur then handle them using the code given by the second child.

If a node doesn't have a value set for pasttype, then it assumes "pirop" if its pirop attribute is set, otherwise it assumes "call".


Get/set the PIR opcode to be executed for this node. The PAST implementation knows about the argument types for many of the PIR opcodes, and will try to automatically convert the results of any children nodes into the correct types if needed. (In general, the implementation tries to convert things to PMCs whenever it isn't certain what else to do.) The table of PIR opcodes that PAST "knows" about is in POST.pir.


Get/set whether this node is an lvalue, or treats its first child as an lvalue (e.g., for assignment).

inline([STRING code])

Get/set the code to be used for inline PIR when pasttype is "inline". The code argument is PIR text to be inserted in the final generated code sequence. Sequences of "%0", "%1", "%2", ... "%9" in code are replaced with the evaluated results of the first, second, third, ..., tenth children nodes. (If you need more than ten arguments to your inline PIR, consider making it a subroutine call instead.)

The register to hold the result of the inline PIR operation is given by "%r", "%t", or "%u" in the code string:

  %r   - Generate a unique PMC register for the result.
  %t   - Generate a unique PMC register for the result,
         and initialize it with an object of type C<returns>
         before the execution of the inline PIR.
  %u   - Re-use the first child's PMC (%0) if it's a temporary
         result, otherwise same as %t above.
  %v   - (void) Re-use the first child's PMC (%0) as the result
         of this operation.

Set a variety of PAST::Op attributes based on entries in hash. Typically hash is an entry in the operator precedence table, and the attributes being set correspond to traits in the grammar.

PAST::Stmts ^

PAST::Stmts is a container of PAST::Node without any specific methods.

PAST::Block ^

PAST::Block nodes represent lexical scopes within an abstract syntax tree, and roughly translate to individual Parrot subroutines. A PAST::Block node nested within another PAST::Block node acts like a nested lexical scope.

If the block has a name attribute, that becomes the name of the resulting Parrot sub, otherwise a unique name is automatically generated for the block.

blocktype([STRING type])

Get/set the type of the block. The currently understood values are 'declaration', 'immediate', and 'method'. 'Declaration' indicates that a block is simply being defined at this point, while 'immediate' indicates a block that is to be immediately executed when it is evaluated in the AST (e.g., the immediate blocks in Perl6 if, while, and other similar statements).


Get/set the control exception handler for this block to value. The exception handler can be any PAST tree. The special (string) value "return_pir" generates code to handle CONTROL_RETURN exceptions.


Get/set the "load initializer" for this block to past. The load initializer is a set of operations to be performed as soon as the block is compiled/loaded. For convenience, requests to loadinit autovivify an empty PAST::Stmts node if one does not already exist.

Within the load initializer, the block PMC register is automatically initialized to refer to the block itself (to enable attaching properties, adding the block as a method, storing in a symbol table, etc.).


Get/set the namespace for this block. The namespace argument can be either a string or an array of strings.

symbol(name, [attr1 => val1, attr2 => val2, ...])

If called with named arguments, sets the symbol hash corresponding to name in the current block. The HLL is free to select any symbol attributes desired, although the 'scope' attribute is typically used to assist with lexical scoping of PAST::Var nodes.

If no named arguments are given, returns the current attribute hash for symbol name.

symbol_defaults([attr1 => val1, attr2 => val2, ... ])

Set default attributes for non-existent symbols in the symbol hash (see symbol above). If no named arguments are given, returns the default attribute hash itself.

Currently we just use the '' entry of the symbol hash to store the default attributes, but it's probably not safe to rely on this behavior in the future.


Get/set the symbol table for the block. May be deprecated in favor of the symbol method above.


Get/set whether the block is a lexical block. A block with this attribute set to false is not lexically scoped inside of its parent, and will not act as an outer lexical scope for any nested blocks within it.


Indicate that the children nodes of this block are to be compiled using compiler name instead of the standard PAST compiler.


Specify named arguments to be passed to the compiler set through the compiler attribute. Not used if compiler is not set.


Get/set any pirflags for this block.


Patrick Michaud <> is the author and maintainer. Please send patches and suggestions to the Parrot porters or Perl 6 compilers mailing lists.


2006-11-20 Patrick Michaud added first draft of POD documentation. 2007-11-21 Re-implementation with pdd26 compliance, compiler toolkit


Copyright (C) 2006-2008, The Perl Foundation.