"Tokenizing" is the process of scanning a string of characters, such as a line of text that the user types at a command prompt, and converting the character string into a list of words and punctuation marks. Each item in this list is called a "token." During parsing, we wish to deal with tokens, not directly with the original character string; it's much easier and faster to work with tokens. To parse a string, we must find word boundaries, skip whitespace, and find matching delimiters (such as quotes and parentheses); we do all of this work in advance, when we tokenize the string, so that we don't have to do it repeatedly while analyzing the syntax of the command.
TADS 3 has no built-in tokenizer. Instead, the system library provides a class called Tokenizer that does this job. You can create a custom tokenizer, if desired, by subclassing Tokenizer. The adv3 library does this to provide a more elaborate set of lexical rules for the English parser.
Calling the Tokenizer
To use the Tokenizer class, #include <tok.h> in your source code. (You'll also have to include the library module tok.t in your build, but you're probably already doing this indirectly by including the standard system library file, system.tl, in your build.)
To use the default rules defined in the class, simply use the class directly; to tokenize a string, make a call like this:
local str, tokList; str = inputLine(); tokList = Tokenizer.tokenize(str);
The tokenize() method scans the string and converts it into a list of tokens. The return value is a list consisting of one element per token. Information about each token can be obtained using the following macros:
- getTokVal(tok) returns the parsed value of the token. This is usually a string corresponding to the text matched by the regular expression, but can be another type if the token rule generated it with some other value type.
- getTokType(tok) returns the type of the token. This is a token type enumerator value assigned by the rule that matched the token. The default Tokenizer rules produce tokens of type tokPunct (punctuation marks), tokWord (words), tokString (strings), and tokInt (integer numbers).
- getTokOrig(tok) returns the original source text the token matched. This information is included because some token rules perform conversions on the value; for example, dictionary word tokens usually have their values converted to all upper-case or all lower-case for convenience in string comparisons. The original text is included so that the exact original token input can be easily reconstructed if needed.
The following code displays the parsed value of each token in a string:
for (local i = 1, local cnt = tokList.length() ; i <= cnt ; ++i) "[<<i>>] = <<getTokVal(tokList[i])>>\n";
(For the curious, the actual representation of a token is a list containing three elements: element 1 is the token value, element 2 is the token type, and element 3 is the original matched text. For more readable code and greater flexibility in case of future changes to this format, you should always use the getTokXxx() macros rather than referring to the list elements directly.)
Customizing the Tokenizer
You can customize the rules the Tokenizer class uses. To do this, subclass Tokenizer and override the rules_ property. This property's value must be a list of lists. Each sublist consists of five elements: a string giving the name of the rule; a regular expression specifying a pattern to match; a token type, which is the enum token value to assign to tokens matching the regular expression; a conversion rule, specifying how the token text to be stored in the result list is obtained; and a match method property pointer, which allows a programmatic check to determine whether or not the token matches the rule.
The name is for your use in identifying the rule. You can give a rule whatever name you like. The tokenizer class has methods that manipulate the rule set at run-time; rule names are used to identify which rules to operate on with these methods.
The conversion rule can be nil, a string, or a property pointer. If the conversion rule is nil, then the token text stored in the result list will simply be the exact text of the input string that matches the regular expression. If the rule is a string, it specifies a replacement string, using the same rules as rexReplace(), that's applied to the matching text; the result of the replacement is stored in the result list. If the conversion rule is a property pointer, it specifies a property (of self, which is the Tokenizer object which is doing the work) to be evaluated to yield the value to be stored in the result list; this property is called as follows:
self.(prop)(txt, typ, results)
In this argument list, txt is a string giving the text that was matched for the token; typ is the token type enum value from the rule list; and results is a Vector containing the token output list under construction. This method simply adds any number of token entries to the results list by calling results.append(). The method need not add any tokens; the default tokenizer rule for whitespace, for example, uses a processor method called tokCvtSkip(), which doesn't do anything at all, which means that whitespace characters in the input result in no tokens in the results list.
The match method can be nil or a property pointer. If it's nil, the regular expression solely determines what text matches the rule. If the match method is a property pointer, though, the tokenizer calls the property (on self, the Tokenizer object which is doing the work) as follows:
This method can examine the text to determine if it's really a match for the rule; the method returns true if the text matches the rule, nil if not. The match method can be used for more complex checks that cannot be performed with the regular expression pattern; for example, a match method can check to see if the token is a known dictionary word, so that a rule only matches known dictionary words.
Rule Processing Order
The rules are specified in order of priority. The tokenizer starts with the first rule; if the first rule's regular expression matches (and the rule's match method, if present, returns true), the tokenizer uses the match and ignores all of the remaining rules. If the first rule's regular expression does not match (or its match method returns nil), the tokenizer tries the second rule, and so on until it runs out of rules.
Each time the tokenizer finds a matching rule, it adds the result of applying the conversion rule to the result list, along with the token type specified by the rule. The tokenizer then removes the matching text from the input string. If that leaves the input string empty, the tokenizer returns the result list to the caller. If the input string is not yet empty, the tokenizer starts over, searching from the first rule to find a match to the remainder of the string. The tokenizer repeats this process until the input string is empty.
If the tokenizer exhausts its list of rules, it throws a TokErrorNoMatch exception. This exception object has a property, remainingStr_, which gives the text of the remainder of the string at the point at which the tokenizer could find no matching rule.
Suppose we wished to build a simple four-function calculator, which reads arithmetic expressions typed by the user and displays the results. For this calculator, we'd need to recognize two types of tokens: operators, and numbers. There's already a tokInt type defined by the Tokenizer class, but we'd have to define our own token type for operators:
enum token tokOp;
The default tokenizer rules won't work for the calculator because they don't accept all of the punctuation marks we'd need to use for operators (and besides, the default rules classify the punctuation marks they do recognize as type tokPunct, when we want tokOp tokens).
We'll need the following token rules:
- Whitespace, which we want to ignore so that the user can use spaces freely to format expressions.
- Integers, which consist of a series of one or more digits.
- Operators, which are the special punctuation marks that indicate arithmetic operations.
Here's how our subclass would look to implement these rules:
CalcTokenizer: Tokenizer rules_ = [ /* skip whitespace */ ['whitespace', R'[ \t]+', nil, &tokCvtSkip, nil], /* integer numbers */ ['integer', R'[0-9]+', tokInt, nil, nil], /* operators */ ['operator', R'[()+*-/]', tokOp, nil, nil] ] ;
To tokenize using our customized rules, we'd simply call our subclasses tokenizer rather than the default tokenizer:
tokList = CalcTokenizer.tokenize(str);