Directives

Parser directives have the form: directive[expression]

A directive modifies the behavior of its enclosed expression, essentially decorating it. The framework pre-defines a few directives. Clients of the framework are free to define their own directives as needed. Information on how this is done will be provided later. For now, we shall deal only with predefined directives.

lexeme_d

Turns off white space skipping. At the phrase level, the parser ignores white spaces, possibly including comments. Use lexeme_d in situations where we want to work at the character level instead of the phrase level. Parsers can be made to work at the character level by enclosing the pertinent parts inside the lexeme_d directive. For example, let us complete the example presented in the Introduction. There, we skipped the definition of the integer rule. Here's how it is actually defined:

    integer = lexeme_d[ !(ch_p('+') | '-') >> +digit ];

The lexeme_d directive instructs the parser to work on the character level. Without it, the integer rule would have allowed erroneous embedded white spaces in inputs such as "1 2 345" which will be parsed as "12345".

as_lower_d

There are times when we want to inhibit case sensitivity. The as_lower_d directive converts all characters from the input to lower-case.

as_lower_d behavior

It is important to note that only the input is converted to lower case. Parsers enclosed inside the as_lower_d expecting upper case characters will fail to parse. Example: as_lower_d['X'] will never succeed because it expects an upper case 'X' that the as_lower_d directive will never supply.

For example, in Pascal, keywords and identifiers are case insensitive. Pascal ignores the case of letters in identifiers and keywords. Identifiers Id, ID and id are indistinguishable in Pascal. Without the as_lower_d directive, it would be awkward to define a rule that recognizes this. Here's a possibility:

    r = str_p("id") | "Id" | "iD" | "ID";

Now, try doing that with the case insensitive Pascal keyword "BEGIN". The as_lower_d directive makes this simple:

    r = as_lower_d["begin"];
Primitive arguments

The astute reader will notice that we did not explicitly wrap "begin" inside an str_p. Whenever appropriate, directives should be able to allow primitive types such as char, int, wchar_t, char const*, wchar_t const* and so on. Examples:

as_lower_d["hello"] // same as as_lower_d[str_p("hello")]
as_lower_d
['x'] // same as as_lower_d[ch_p('x')]

no_actions_d

There are cases where you want semantic actions not to be triggered. By enclosing a parser in the no_actions_d directive, all semantic actions directly or indirectly attached to the parser will not fire.

    no_actions_d[expression]

Tweaking the Scanner Type

How does lexeme_d, as_lower_d and no_actions_d work? These directives do their magic by tweaking the scanner policies. Well, you don't need to know what that means for now. Scanner policies are discussed later. However, it is important to note that when the scanner policy is tweaked, the result is a different scanner. Why is this important to note? The rule is tied to a particular scanner (one or more scanners, to be precise). If you wrap a rule inside a lexeme_d, as_lower_d or no_actions_d,the compiler will complain about scanner mismatch unless you associate the required scanner with the rule.

lexeme_scanner, as_lower_scanner and no_actions_scanner are your friends if the need to wrap a rule inside these directives arise. Learn bout these beasts in the next chapter on The Scanner and Parsing.

longest_d

Alternatives in the Spirit parser compiler are short-circuited (see Operators). Sometimes, this is not what is desired. The longest_d directive instructs the parser not to short-circuit alternatives enclosed inside this directive, but instead makes the parser try all possible alternatives and choose the one matching the longest portion of the input stream.

Consider the parsing of integers and real numbers:

    number = real | integer;

A number can be a real or an integer. This grammar is ambiguous. An input "1234" should potentially match both real and integer. Recall though that alternatives are short-circuited . Thus, for inputs such as above, the real alternative always wins. However, if we swap the alternatives:

    number = integer | real;

we still have a problem. Now, an input "123.456" will be partially matched by integer until the decimal point. This is not what we want. The solution here is either to fix the ambiguity by factoring out the common prefixes of real and integer or, if that is not possible nor desired, use the longest_d directive:

    number = longest_d[ integer | real ];

shortest_d

Opposite of the longest_d directive.

Multiple alternatives

The longest_d and shortest_d directives can accept two or more alternatives. Examples:

longest[ a | b | c ];
shortest
[ a | b | c | d ];

limit_d

Ensures that the result of a parser is constrained to a given min..max range (inclusive). If not, then the parser fails and returns a no-match.

Usage:

    limit_d(min, max)[expression]

This directive is particularly useful in conjunction with parsers that parse specific scalar ranges (for example, numeric parsers). Here's a practical example. Although the numeric parsers can be configured to accept only a limited number of digits (say, 0..2), there is no way to limit the result to a range (say -1.0..1.0). This design is deliberate. Doing so would have undermined Spirit's design rule that "the client should not pay for features that she does not use". We would have stored the min, max values in the numeric parser itself, used or unused. Well, we could get by by using static constants configured by a non-type template parameter, but that is not acceptable because that way, we can only accommodate integers. What about real numbers or user defined numbers such as big-ints?

Example, parse time of the form HH:MM:SS:

    uint_parser<int, 10, 2, 2> uint2_p;

    r = lexeme_d
        [
                limit_d(0u, 23u)[uint2_p] >> ':'    //  Hours 00..23
            >>  limit_d(0u, 59u)[uint2_p] >> ':'    //  Minutes 00..59
            >>  limit_d(0u, 59u)[uint2_p]           //  Seconds 00..59
        ];

min_limit_d

Sometimes, it is useful to unconstrain just the maximum limit. This will allow for an interval that's unbounded in one direction. The directive min_limit_d ensures that the result of a parser is not less than minimum. If not, then the parser fails and returns a no-match.

Usage:

    min_limit_d(min)[expression]

Example, ensure that a date is not less than 1900

    min_limit_d(1900u)[uint_p]

max_limit_d

Opposite of min_limit_d. Take note that limit_d[p] is equivalent to:

    min_limit_d(min)[max_limit_d(max)[p]]