public Class RuleBasedNumberFormat

A class that formats numbers according to a set of rules. This number formatter is typically used for spelling out numeric values in words (e.g., 25,3476 as "twenty-five thousand three hundred seventy-six" or "vingt-cinq mille trois cents soixante-seize" or "funfundzwanzigtausenddreihundertsechsundsiebzig"), but can also be used for other complicated formatting tasks, such as formatting a number of seconds as hours, minutes and seconds (e.g., 3,730 as "1:02:10").

The resources contain three predefined formatters for each locale: spellout, which spells out a value in words (123 is "one hundred twenty-three"); ordinal, which appends an ordinal suffix to the end of a numeral (123 is "123rd"); and duration, which shows a duration in seconds as hours, minutes, and seconds (123 is "2:03").  The client can also define more specialized RuleBasedNumberFormats by supplying programmer-defined rule sets.

The behavior of a RuleBasedNumberFormat is specified by a textual description that is either passed to the constructor as a String or loaded from a resource bundle. In its simplest form, the description consists of a semicolon-delimited list of rules. Each rule has a string of output text and a value or range of values it is applicable to. In a typical spellout rule set, the first twenty rules are the words for the numbers from 0 to 19:

zero; one; two; three; four; five; six; seven; eight; nine;
ten; eleven; twelve; thirteen; fourteen; fifteen; sixteen; seventeen; eighteen; nineteen;

For larger numbers, we can use the preceding set of rules to format the ones place, and we only have to supply the words for the multiples of 10:

20: twenty[->>];
30: thirty{->>];
40: forty[->>];
50: fifty[->>];
60: sixty[->>];
70: seventy[->>];
80: eighty[->>];
90: ninety[->>];

In these rules, the base value is spelled out explicitly and set off from the rule's output text with a colon. The rules are in a sorted list, and a rule is applicable to all numbers from its own base value to one less than the next rule's base value. The ">>" token is called a substitution and tells the formatter to isolate the number's ones digit, format it using this same set of rules, and place the result at the position of the ">>" token. Text in brackets is omitted if the number being formatted is an even multiple of 10 (the hyphen is a literal hyphen; 24 is "twenty-four," not "twenty four").

For even larger numbers, we can actually look up several parts of the number in the list:

100: << hundred[ >>];

The "<<" represents a new kind of substitution. The << isolates the hundreds digit (and any digits to its left), formats it using this same rule set, and places the result where the "<<" was. Notice also that the meaning of >> has changed: it now refers to both the tens and the ones digits. The meaning of both substitutions depends on the rule's base value. The base value determines the rule's divisor, which is the highest power of 10 that is less than or equal to the base value (the user can change this). To fill in the substitutions, the formatter divides the number being formatted by the divisor. The integral quotient is used to fill in the << substitution, and the remainder is used to fill in the >> substitution. The meaning of the brackets changes similarly: text in brackets is omitted if the value being formatted is an even multiple of the rule's divisor. The rules are applied recursively, so if a substitution is filled in with text that includes another substitution, that substitution is also filled in.

This rule covers values up to 999, at which point we add another rule:

1000: << thousand[ >>];

Again, the meanings of the brackets and substitution tokens shift because the rule's base value is a higher power of 10, changing the rule's divisor. This rule can actually be used all the way up to 999,999. This allows us to finish out the rules as follows:

1,000,000: << million[ >>];
1,000,000,000: << billion[ >>];
1,000,000,000,000: << trillion[ >>];
1,000,000,000,000,000: OUT OF RANGE!;

Commas, periods, and spaces can be used in the base values to improve legibility and are ignored by the rule parser. The last rule in the list is customarily treated as an "overflow rule," applying to everything from its base value on up, and often (as in this example) being used to print out an error message or default representation. Notice also that the size of the major groupings in large numbers is controlled by the spacing of the rules: because in English we group numbers by thousand, the higher rules are separated from each other by a factor of 1,000.

To see how these rules actually work in practice, consider the following example: Formatting 25,430 with this rule set would work like this:

The above syntax suffices only to format positive integers. To format negative numbers, we add a special rule:

-x: minus >>;

This is called a negative-number rule, and is identified by "-x" where the base value would be. This rule is used to format all negative numbers. the >> token here means "find the number's absolute value, format it with these rules, and put the result here."

We also add a special rule called a fraction rule for numbers with fractional parts:

x.x: << point >>;

This rule is used for all positive non-integers (negative non-integers pass through the negative-number rule first and then through this rule). Here, the << token refers to the number's integral part, and the >> to the number's fractional part. The fractional part is formatted as a series of single-digit numbers (e.g., 123.456 would be formatted as "one hundred twenty-three point four five six").

To see how this rule syntax is applied to various languages, examine the resource data.

There is actually much more flexibility built into the rule language than the description above shows. A formatter may own multiple rule sets, which can be selected by the caller, and which can use each other to fill in their substitutions. Substitutions can also be filled in with digits, using a DecimalFormat object. There is syntax that can be used to alter a rule's divisor in various ways. And there is provision for much more flexible fraction handling. A complete description of the rule syntax follows:

The description of a RuleBasedNumberFormat's behavior consists of one or more rule sets. Each rule set consists of a name, a colon, and a list of rules. A rule set name must begin with a % sign. Rule sets with names that begin with a single % sign are public: the caller can specify that they be used to format and parse numbers. Rule sets with names that begin with %% are private: they exist only for the use of other rule sets. If a formatter only has one rule set, the name may be omitted.

The user can also specify a special "rule set" named %%lenient-parse. The body of %%lenient-parse isn't a set of number-formatting rules, but a RuleBasedCollator description which is used to define equivalences for lenient parsing. For more information on the syntax, see RuleBasedCollator. For more information on lenient parsing, see setLenientParse()

The body of a rule set consists of an ordered, semicolon-delimited list of rules. Internally, every rule has a base value, a divisor, rule text, and zero, one, or two substitutions. These parameters are controlled by the description syntax, which consists of a rule descriptor, a colon, and a rule body.

A rule descriptor can take one of the following forms (text in italics is the name of a token):

A rule set may be either a regular rule set or a fraction rule set, depending on whether it is used to format a number's integral part (or the whole number) or a number's fractional part. Using a rule set to format a rule's fractional part makes it a fraction rule set.

Which rule is used to format a number is defined according to one of the following algorithms: If the rule set is a regular rule set, do the following:

• If the rule set includes a default rule (and the number was passed in as a double), use the default rule.  (If the number being formatted was passed in as a long, the default rule is ignored.)
• If the number is negative, use the negative-number rule.
• If the number has a fractional part and is greater than 1, use the improper fraction rule.
• If the number has a fractional part and is between 0 and 1, use the proper fraction rule.
• Binary-search the rule list for the rule with the highest base value less than or equal to the number. If that rule has two substitutions, its base value is not an even multiple of its divisor, and the number is an even multiple of the rule's divisor, use the rule that precedes it in the rule list. Otherwise, use the rule itself.

If the rule set is a fraction rule set, do the following:

• Ignore negative-number and fraction rules.
• For each rule in the list, multiply the number being formatted (which will always be between 0 and 1) by the rule's base value. Keep track of the distance between the result the nearest integer.
• Use the rule that produced the result closest to zero in the above calculation. In the event of a tie or a direct hit, use the first matching rule encountered. (The idea here is to try each rule's base value as a possible denominator of a fraction. Whichever denominator produces the fraction closest in value to the number being formatted wins.) If the rule following the matching rule has the same base value, use it if the numerator of the fraction is anything other than 1; if the numerator is 1, use the original matching rule. (This is to allow singular and plural forms of the rule text without a lot of extra hassle.)

A rule's body consists of a string of characters terminated by a semicolon. The rule may include zero, one, or two substitution tokens, and a range of text in brackets. The brackets denote optional text (and may also include one or both substitutions). The exact meanings of the substitution tokens, and under what conditions optional text is omitted, depend on the syntax of the substitution token and the context. The rest of the text in a rule body is literal text that is output when the rule matches the number being formatted.

A substitution token begins and ends with a token character. The token character and the context together specify a mathematical operation to be performed on the number being formatted. An optional substitution descriptor specifies how the value resulting from that operation is used to fill in the substitution. The position of the substitution token in the rule body specifies the location of the resultant text in the original rule text.

The meanings of the substitution token characters are as follows:

The substitution descriptor (i.e., the text between the token characters) may take one of three forms:

Whitespace is ignored between a rule set name and a rule set body, between a rule descriptor and a rule body, or between rules. If a rule body begins with an apostrophe, the apostrophe is ignored, but all text after it becomes significant (this is how you can have a rule's rule text begin with whitespace). There is no escape function: the semicolon is not allowed in rule set names or in rule text, and the colon is not allowed in rule set names. The characters beginning a substitution token are always treated as the beginning of a substitution token.

See the resource data and the demo program for annotated examples of real rule sets using these features.