Writing and Generating Documentation

Documentation for the GNU C++ Library is created from three independent sources: a manual, a FAQ, and an API reference.

The sub-directory doc within the main source directory contains Makefile.am and Makefile.in, which provide rules for generating documentation, described in excruciating detail below. The doc sub-directory also contains three directories: doxygen, which contains scripts and fragments for doxygen, html, which contains an html version of the manual, and xml, which contains an xml version of the manual.

Diverging from established documentation conventions in the rest of the GCC project, libstdc++ does not use Texinfo as a markup language. Instead, Docbook is used to create the manual and the FAQ, and Doxygen is used to construct the API reference. Although divergent, this conforms to the GNU Project recommendations as long as the output is of sufficient quality, as per GNU Manuals.

Certain Makefile rules are required by the GNU Coding Standards. These standard rules generate HTML, PDF, XML, or man files. For each of the generative rules, there is an additional install rule that is used to install any generated documentation files into the prescribed installation directory. Files are installed into share/doc or share/man directories.

The standard Makefile rules are conditionally supported, based on the results of examining the host environment for prerequisites at configuration time. If requirements are not found, the rule is aliased to a dummy rule that does nothing, and produces no documentation. If the requirements are found, the rule forwards to a private rule that produces the requested documentation.

For more details on what prerequisites were found and where, please consult the file config.log in the libstdc++ build directory. Compare this log to what is expected for the relevant Makefile conditionals: BUILD_INFO, BUILD_XML, BUILD_HTML, BUILD_MAN, BUILD_PDF, and BUILD_EPUB.

Supported Makefile rules:

Makefile rules for several other formats are explicitly not supported, and are always aliased to dummy rules. These unsupported formats are: info, ps, and dvi.


Prerequisite tools are Bash 2.0 or later, Doxygen, and the GNU coreutils. (GNU versions of find, xargs, and possibly sed and grep are used, just because the GNU versions make things very easy.)

To generate the pretty pictures and hierarchy graphs, the Graphviz package will need to be installed. For PDF output, pdflatex is required.

Be warned the PDF file generated via doxygen is extremely large. At last count, the PDF file is over three thousand pages. Generating this document taxes the underlying TeX formatting system, and will require the expansion of TeX's memory capacity. Specifically, the pool_size variable in the configuration file texmf.cnf may need to be increased by a minimum factor of two.

In general, libstdc++ files should be formatted according to the rules found in the Coding Standard. Before any doxygen-specific formatting tweaks are made, please try to make sure that the initial formatting is sound.

Adding Doxygen markup to a file (informally called doxygenating) is very simple. The Doxygen manual can be found here. We try to use a very-recent version of Doxygen.

For classes, use deque/vector/list and std::pair as examples. For functions, see their member functions, and the free functions in stl_algobase.h. Member functions of other container-like types should read similarly to these member functions.

Some commentary to accompany the first list in the Special Documentation Blocks section of the Doxygen manual:

  1. For longer comments, use the Javadoc style...

  2. ...not the Qt style. The intermediate *'s are preferred.

  3. Use the triple-slash style only for one-line comments (the brief mode).

  4. This is disgusting. Don't do this.

Some specific guidelines:

Use the @-style of commands, not the !-style. Please be careful about whitespace in your markup comments. Most of the time it doesn't matter; doxygen absorbs most whitespace, and both HTML and *roff are agnostic about whitespace. However, in <pre> blocks and @code/@endcode sections, spacing can have interesting effects.

Use either kind of grouping, as appropriate. doxygroups.cc exists for this purpose. See stl_iterator.h for a good example of the other kind of grouping.

Please use markup tags like @p and @a when referring to things such as the names of function parameters. Use @e for emphasis when necessary. Use @c to refer to other standard names. (Examples of all these abound in the present code.)

Complicated math functions should use the multi-line format. An example from random.h:


/**
 * @brief A model of a linear congruential random number generator.
 *
 * @f[
 *     x_{i+1}\leftarrow(ax_{i} + c) \bmod m
 * @f]
 */

One area of note is the markup required for @file markup in header files. Two details are important: for filenames that have the same name in multiple directories, include part of the installed path to disambiguate. For example:


/** @file debug/vector
 *  This file is a GNU debug extension to the Standard C++ Library.
 */

The other relevant detail for header files is the use of a libstdc++-specific doxygen alias that helps distinguish between public header files (like random) from implementation or private header files (like bits/c++config.h.) This alias is spelled @headername and can take one or two arguments that detail the public header file or files that should be included to use the contents of the file. All header files that are not intended for direct inclusion must use headername in the file block. An example:


/** @file bits/basic_string.h
 *  This is an internal header file, included by other library headers.
 *  Do not attempt to use it directly. @headername{string}
 */

Be careful about using certain, special characters when writing Doxygen comments. Single and double quotes, and separators in filenames are two common trouble spots. When in doubt, consult the following table.



Editing the DocBook sources requires an XML editor. Many exist: some notable options include emacs, Kate, or Conglomerate.

Some editors support special XML Validation modes that can validate the file as it is produced. Recommended is the nXML Mode for emacs.

Besides an editor, additional DocBook files and XML tools are also required.

Access to the DocBook 5.0 stylesheets and schema is required. The stylesheets are usually packaged by vendor, in something like docbook5-style-xsl. To exactly match generated output, please use a version of the stylesheets equivalent to docbook5-style-xsl-1.75.2-3. The installation directory for this package corresponds to the XSL_STYLE_DIR in doc/Makefile.am and defaults to /usr/share/sgml/docbook/xsl-ns-stylesheets.

For processing XML, an XSLT processor and some style sheets are necessary. Defaults are xsltproc provided by libxslt.

For validating the XML document, you'll need something like xmllint and access to the relevant DocBook schema. These are provided by a vendor package like libxml2 and docbook5-schemas-5.0-4

For PDF output, something that transforms valid Docbook XML to PDF is required. Possible solutions include dblatex, xmlto, or prince. Of these, dblatex is the default. Other options are listed on the DocBook web pages. Please consult the list when preparing printed manuals for current best practice and suggestions.

For Texinfo output, something that transforms valid Docbook XML to Texinfo is required. The default choice is docbook2X.

For epub output, the stylesheets for EPUB3 are required. These stylesheets are still in development. To validate the created file, epubcheck is necessary.


      Which files are important

      All Docbook files are in the directory
      libstdc++-v3/doc/xml

      Inside this directory, the files of importance:
      spine.xml   - index to documentation set
      manual/spine.xml  - index to manual
      manual/*.xml   - individual chapters and sections of the manual
      faq.xml   - index to FAQ
      api.xml   - index to source level / API

      All *.txml files are template xml files, i.e., otherwise empty files with
      the correct structure, suitable for filling in with new information.

      Canonical Writing Style

      class template
      function template
      member function template
      (via C++ Templates, Vandevoorde)

      class in namespace std: allocator, not std::allocator

      header file: iostream, not <iostream>


      General structure

      <set>
      <book>
      </book>

      <book>
      <chapter>
      </chapter>
      </book>

      <book>
      <part>
      <chapter>
      <section>
      </section>

      <sect1>
      </sect1>

      <sect1>
      <sect2>
      </sect2>
      </sect1>
      </chapter>

      <chapter>
      </chapter>
      </part>
      </book>

      </set>