Boost C++ Libraries

...one of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards

PrevUpHomeNext
About Tokens and Token Values

As already discussed, lexical scanning is the process of analyzing the stream of input characters and separating it into strings called tokens, most of the time separated by whitespace. The different token types recognized by a lexical analyzer often get assigned unique integer token identifiers (token ids). These token ids are normally used by the parser to identify the current token without having to look at the matched string again. The Spirit.Lex library is not different with respect to this, as it uses the token ids as the main means of identification of the different token types defined for a particular lexical analyzer. However, it is different from commonly used lexical analyzers in the sense that it returns (references to) instances of a (user defined) token class to the user. The only limitation of this token class is that it must carry at least the token id of the token it represents. For more information about the interface a user defined token type has to expose please look at the Token Class reference. The library provides a default token type based on the Lexertl library which should be sufficient in most cases: the lex::lexertl::token<> type. This section focusses on the description of general features a token class may implement and how this integrates with the other parts of the Spirit.Lex library.

The Anatomy of a Token

It is very important to understand the difference between a token definition (represented by the lex::token_def<> template) and a token itself (for instance represented by the lex::lexertl::token<> template).

The token definition is used to describe the main features of a particular token type, especially:

The token itself is a data structure returned by the lexer iterators. Dereferencing a lexer iterator returns a reference to the last matched token instance. It encapsulates the part of the underlying input sequence matched by the regular expression used during the definition of this token type. Incrementing the lexer iterator invokes the lexical analyzer to match the next token by advancing the underlying input stream. The token data structure contains at least the token id of the matched token type, allowing to identify the matched character sequence. Optionally, the token instance may contain a token value and/or the lexer state this token instance was matched in. The following figure shows the schematic structure of a token.

Figure 8. The structure of a token

The structure of a token


The token value and the lexer state the token has been recognized in may be omitted for optimization reasons, thus avoiding the need for the token to carry more data than actually required. This configuration can be achieved by supplying appropriate template parameters for the lex::lexertl::token<> template while defining the token type.

The lexer iterator returns the same token type for each of the different matched token definitions. To accommodate for the possible different token value types exposed by the various token types (token definitions), the general type of the token value is a Boost.Variant. At a minimum (for the default configuration) this token value variant will be configured to always hold a boost::iterator_range containing the pair of iterators pointing to the matched input sequence for this token instance.

[Note] Note

If the lexical analyzer is used in conjunction with a Spirit.Qi parser, the stored boost::iterator_range token value will be converted to the requested token type (parser attribute) exactly once. This happens at the time of the first access to the token value requiring the corresponding type conversion. The converted token value will be stored in the Boost.Variant replacing the initially stored iterator range. This avoids having to convert the input sequence to the token value more than once, thus optimizing the integration of the lexer with Spirit.Qi, even during parser backtracking.

Here is the template prototype of the lex::lexertl::token<> template:

template <
    typename Iterator = char const*,
    typename AttributeTypes = mpl::vector0<>,
    typename HasState = mpl::true_
>
struct lexertl_token;

where:

Iterator

This is the type of the iterator used to access the underlying input stream. It defaults to a plain char const*.

AttributeTypes

This is either a mpl sequence containing all attribute types used for the token definitions or the type omit. If the mpl sequence is empty (which is the default), all token instances will store a boost::iterator_range<Iterator> pointing to the start and the end of the matched section in the input stream. If the type is omit, the generated tokens will contain no token value (attribute) at all.

HasState

This is either mpl::true_ or mpl::false_, allowing control as to whether the generated token instances will contain the lexer state they were generated in. The default is mpl::true_, so all token instances will contain the lexer state.

Normally, during construction, a token instance always holds the boost::iterator_range as its token value, unless it has been defined using the omit token value type. This iterator range then is converted in place to the requested token value type (attribute) when it is requested for the first time.

The Physiognomy of a Token Definition

The token definitions (represented by the lex::token_def<> template) are normally used as part of the definition of the lexical analyzer. At the same time a token definition instance may be used as a parser component in Spirit.Qi.

The template prototype of this class is shown here:

template<
    typename Attribute = unused_type,
    typename Char = char
>
class token_def;

where:

Attribute

This is the type of the token value (attribute) supported by token instances representing this token type. This attribute type is exposed to the Spirit.Qi library, whenever this token definition is used as a parser component. The default attribute type is unused_type, which means the token instance holds a boost::iterator_range pointing to the start and the end of the matched section in the input stream. If the attribute is omit the token instance will expose no token type at all. Any other type will be used directly as the token value type.

Char

This is the value type of the iterator for the underlying input sequence. It defaults to char.

The semantics of the template parameters for the token type and the token definition type are very similar and interdependent. As a rule of thumb you can think of the token definition type as the means of specifying everything related to a single specific token type (such as identifier or integer). On the other hand the token type is used to define the general properties of all token instances generated by the Spirit.Lex library.

[Important] Important

If you don't list any token value types in the token type definition declaration (resulting in the usage of the default boost::iterator_range token type) everything will compile and work just fine, just a bit less efficient. This is because the token value will be converted from the matched input sequence every time it is requested.

But as soon as you specify at least one token value type while defining the token type you'll have to list all value types used for lex::token_def<> declarations in the token definition class, otherwise compilation errors will occur.

Examples of using lex::lexertl::token<>

Let's start with some examples. We refer to one of the Spirit.Lex examples (for the full source code of this example please see example4.cpp).

The first code snippet shows an excerpt of the token definition class, the definition of a couple of token types. Some of the token types do not expose a special token value (if_, else_, and while_). Their token value will always hold the iterator range of the matched input sequence. The token definitions for the identifier and the integer constant are specialized to expose an explicit token type each: std::string and unsigned int.

// these tokens expose the iterator_range of the matched input sequence
lex::token_def<> if_, else_, while_;

// The following two tokens have an associated attribute type, 'identifier'
// carries a string (the identifier name) and 'constant' carries the 
// matched integer value.
//
// Note: any token attribute type explicitly specified in a token_def<>
//       declaration needs to be listed during token type definition as 
//       well (see the typedef for the token_type below).
//
// The conversion of the matched input to an instance of this type occurs
// once (on first access), which makes token attributes as efficient as 
// possible. Moreover, token instances are constructed once by the lexer
// library. From this point on tokens are passed by reference only, 
// avoiding them being copied around.
lex::token_def<std::string> identifier;
lex::token_def<unsigned int> constant;

As the parsers generated by Spirit.Qi are fully attributed, any Spirit.Qi parser component needs to expose a certain type as its parser attribute. Naturally, the lex::token_def<> exposes the token value type as its parser attribute, enabling a smooth integration with Spirit.Qi.

The next code snippet demonstrates how the required token value types are specified while defining the token type to use. All of the token value types used for at least one of the token definitions have to be re-iterated for the token definition as well.

// This is the lexer token type to use. The second template parameter lists 
// all attribute types used for token_def's during token definition (see 
// calculator_tokens<> above). Here we use the predefined lexertl token 
// type, but any compatible token type may be used instead.
//
// If you don't list any token attribute types in the following declaration 
// (or just use the default token type: lexertl_token<base_iterator_type>)  
// it will compile and work just fine, just a bit less efficient. This is  
// because the token attribute will be generated from the matched input  
// sequence every time it is requested. But as soon as you specify at 
// least one token attribute type you'll have to list all attribute types 
// used for token_def<> declarations in the token definition class above, 
// otherwise compilation errors will occur.
typedef lex::lexertl::token<
    base_iterator_type, boost::mpl::vector<unsigned int, std::string>
> token_type;

To avoid the token to have a token value at all, the special tag omit can be used: token_def<omit> and lexertl_token<base_iterator_type, omit>.


PrevUpHomeNext