Package TWiki::Infix::Parser

A simple stack-based parser that parses infix expressions with nonary, unary and binary operators specified using an operator table.

Escapes are supported in strings, using backslash.

new($client_class, \%options) -> parser object

Creates a new infix parser. Operators must be added for it to be useful.

The tokeniser matches tokens in the following order: operators, quotes (" and '), numbers, words, brackets. If you have any overlaps (e.g. an operator '<' and a bracket operator '<<') then the first choice will match.

$client_class needs to be the name of a package that supports the following two functions:

• newLeaf($val, $type) - create a terminal. $type will be:
  1. if the terminal matched the words specification (see below).
  2. if it is a number matched the numbers specification (see below)
  3. if it is a quoted string
• =newNode($op, @params) - create a new operator node. @params is a variable-length list of parameters, left to right. $op is a reference to the operator hash in the \@opers list.

These functions should throw Error::Simple in the event of errors. TWiki::Infix::Node is such a class, ripe for subclassing.

The remaining parameters are named, and specify options that affect the behaviour of the parser:

1. words=>qr// - should be an RE specifying legal words (unquoted terminals that are not operators i.e. names and numbers). By default this is \w+. It's ok if operator names match this RE; operators always have precedence over atoms.
2. numbers=>qr// - should be an RE specifying legal numbers (unquoted terminals that are not operators or words). By default this is qr/[+-]?(?:\d+\.\d+|\d+\.|\.\d+|\d+)(?:[eE][+-]?\d+)?/, which matches integers and floating-point numbers. Number matching always takes precedence over word matching (i.e. "1xy" will be parsed as a number followed by a word. A typical usage of this option is when you only want to recognise integers, in which case you would set this to numbers => qr/\d+/.

ObjectMethod addOperator (%oper)

Add an operator to the parser.

%oper is a hash, containing the following fields:

• name - operator string
• prec - operator precedence, positive non-zero integer. Larger number => higher precedence.
• arity - set to 1 if this operator is unary, 2 for binary. Arity 0 is legal, should you ever need it.
• close - used with bracket operators. name should be the open bracket string, and close the close bracket. The existance of close marks this as a bracket operator.
• casematters= - indicates that the parser should check case in the operator name (i.e. treat 'AND' and 'and' as different). By default operators are case insensitive. Note that operator names must be caselessly unique i.e. you can't define 'AND' and 'and' as different operators in the same parser. Does not affect the interpretation of non-operator terminals (names).

Other fields in the hash can be used for other purposes; the parse tree generated by this parser will point to the hashes passed to this function.

Field names in the hash starting with InfixParser_ are reserved for use by the parser.

ObjectMethod parse ($string) -> $parseTree

Parses $string, calling newLeaf and newNode in the client class as necessary to create a parse tree. Returns the result of calling newNode on the root of the parse.

Throws TWiki::Infix::Error in the event of parse errors.