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

Click here to view the latest version of this page.

Warming up

Learning how to use Spirit.Karma is really simple. We will start from trivial examples, ramping up as we go.

Trivial Example #1 Generating a number

Let's create a generator that will output a floating-point number:


Easy huh? The above code actually instantiates a Spirit floating point generator (a built-in generator). Spirit has many pre-defined generators and consistent naming conventions will help you finding your way through the maze. Especially important to note is that things related to identical entities (as in this case, floating point numbers) are named identically in Spirit.Karma and in Spirit.Qi. Actually, both libraries are using the very same variable instance to refer to a floating point generator or parser: double_.

Trivial Example #2 Generating two numbers

Now, let's create a generator that will output a line consisting of two floating-point numbers.

double_ << double_

Here you see the familiar floating-point numeric generator double_ used twice, once for each number. If you are used to see the '>>' operator for concatenating two parsers in Spirit.Qi you might wonder, what's that '<<' operator doing in there? We decided to distinguish generating and parsing of sequences the same way as the std::stream libraries do: we use operator '>>' for input (parsing), and operator '<<' for output (generating). Other than that there is no significant difference. The above program creates a generator from two simpler generators, glueing them together with the sequence operator. The result is a generator that is a composition of smaller generators. Whitespace between numbers can implicitly be inserted depending on how the generator is invoked (see below).

[Note] Note

When we combine generators, we end up with a "bigger" generator, but it's still a generator. Generators can get bigger and bigger, nesting more and more, but whenever you glue two generators together, you end up with one bigger generator. This is an important concept.

Trivial Example #3 Generating one or more numbers

Now, creating output for two numbers is not too interesting. Let's create a generator that will output zero or more floating-point numbers in a row.


This is like a regular-expression Kleene Star. We moved the * to the front for the same reasons we did in Spirit.Qi: we must work with the syntax rules of C++. But if you know regular expressions (and for sure you remember those C++ syntax rules) it will start to look very familiar in a matter of a very short time.

Any expression that evaluates to a generator may be used with the Kleene Star. Keep in mind, though, that due to C++ operator precedence rules you may need to put the expression in parentheses for complex expressions. As above, whitespace can be inserted implicitly in between the generated numbers, if needed.

Trivial Example #4 Generating a comma-delimited list of numbers

We follow the lead of Spirit.Qi's warming up section and will create a generator that produces a comma-delimited list of numbers.

double_ << *(lit(',') << double_)

Notice lit(','). It is a literal character generator that simply generates the comma ','. In this case, the Kleene Star is modifying a more complex generator, namely, the one generated by the expression:

(lit(',') << double_)

Note that this is a case where the parentheses are necessary. The Kleene Star encloses the complete expression above, repeating the whole pattern in the generated output zero or more times.

Let's Generate!

We're done with defining the generator. All that's left is to invoke the generator to do its work. For now, we will use the generate_delimited function. One overload of this function accepts four arguments:

  1. An output iterator accepting the generated characters
  2. The generator expression
  3. Another generator called the delimiting generator
  4. The data to format and output

While comparing this minimal example with an equivalent parser example we notice a significant difference. It is possible (and actually, it makes a lot of sense) to use a parser without creating any internal representation of the parsed input (i.e. without 'producing' any data from the parsed input). Using a parser in this mode checks the provided input against the given parser expression allowing to verify whether the input is parsable. For generators this mode doesn't make any sense. What is output generation without generating any output? So we always will have to supply the data the output should be generated from. In our example we supply a list of double numbers as the last parameter to the function generate_delimited (see code below).

In this example, we wish to delimit the generated numbers by spaces. Another generator named space is included in Spirit's repertoire of predefined generators. It is a very trivial generator that simply produces spaces. It is the equivalent to writing lit(' '), or simply ' '. It has been implemented for similarity with the corresponding predefined space parser. We will use space as our delimiter. The delimiter is the one responsible for inserting characters in between generator elements such as the double_ and lit.

Ok, so now let's generate (for the complete source code of this example please refer to num_list1.cpp).

template <typename OutputIterator>
bool generate_numbers(OutputIterator& sink, std::list<double> const& v)
    using karma::double_;
    using karma::generate_delimited;
    using ascii::space;

    bool r = generate_delimited(
        sink,                           // destination: output iterator
        double_ << *(',' << double_),   // the generator
        space,                          // the delimiter-generator
        v                               // the data to output 
    return r;

[Note] Note

You might wonder how a vector<double>, which is actually a single data structure, can be used as an argument (we call it attribute) to a sequence of generators. This seems to be counter intuitive and doesn't match with your experience of using printf, where each formatting placeholder has to be matched with a corresponding argument. Well, we will explain this behavior in more detail later in this tutorial. For now just consider this to be a special case, implemented on purpose to allow more flexible output formatting of STL containers: sequences accept a single container attribute if all elements of this sequence accept attributes compatible with the elements held by this container.

The generate function returns true or false depending on the result of the output generation. As outlined in different places of this documentation, a generator may fail for different reasons. One of the possible reasons is an error in the underlying output iterator (memory exhausted or disk full, etc.). Another reason might be that the data doesn't match the requirements of a particular generator.

[Note] Note

char and wchar_t operands

The careful reader may notice that the generator expression has ',' instead of lit(',') as the previous examples did. This is ok due to C++ syntax rules of conversion. Spirit provides << operators that are overloaded to accept a char or wchar_t argument on its left or right (but not both). An operator may be overloaded if at least one of its parameters is a user-defined type. In this case, the double_ is the 2nd argument to operator<<, and so the proper overload of << is used, converting ',' into a character literal generator.

The problem with omitting the lit should be obvious: 'a' << 'b' is not a spirit generator, it is a numeric expression, left-shifting the ASCII (or another encoding) value of 'a' by the ASCII value of 'b'. However, both lit('a') << 'b' and 'a' << lit('b') are Spirit sequence generators for the letter 'a' followed by 'b'. You'll get used to it, sooner or later.

Note that we inlined the generator directly in the call to generate_delimited. Upon calling this function, the expression evaluates into a temporary, unnamed generator which is passed into the generate_delimited function, used, and then destroyed.

Here, we chose to make the generate function generic by making it a template, parameterized by the output iterator type. By doing so, it can put the generated data into any STL conforming output iterator.