Design overview

pygccxml has 4 packages:

  • declarations

    This package defines classes that describe C++ declarations and types.

  • parser

    This package defines classes that parse GCC-XML or CastXML generated files. It also defines a few classes that will help you unnecessary parsing of C++ source files.

  • utils

    This package defines a few functions useful for the whole project, but which are mainly used internally by pygccxml.

declarations package

Please take a look on the UML diagram. This UML diagram describes almost all classes defined in the package and their relationship. declarations package defines two hierarchies of class:

  1. types hierarchy - used to represent a C++ type
  2. declarations hierarchy - used to represent a C++ declaration

Types hierarchy

Types hierarchy is used to represent an arbitrary type in C++. class type_t is the base class.


Are you aware of boost::type_traits library? The boost::type_traits library contains a set of very specific traits classes, each of which encapsulate a single trait from the C++ type system; for example, is a type a pointer or a reference? Or does a type have a trivial constructor, or a const-qualifier?

pygccxml implements a lot of functionality from the library:

  • a lot of algorithms were implemented

    • is_same
    • is_enum
    • is_void
    • is_const
    • is_array
    • is_pointer
    • is_volatile
    • is_integral
    • is_reference
    • is_arithmetic
    • is_convertible
    • is_fundamental
    • is_floating_point
    • is_base_and_derived
    • is_unary_operator
    • is_binary_operator
    • remove_cv
    • remove_const
    • remove_alias
    • remove_pointer
    • remove_volatile
    • remove_reference
    • has_trivial_copy
    • has_trivial_constructor
    • has_any_non_copyconstructor

    For a full list of implemented algorithms, please consult API documentation.

  • a lot of unit tests has been written base on unit tests from the boost::type_traits library.

If you are going to build code generator, you will find type_traits very handy.

Declarations hierarchy

A declaration hierarchy is used to represent an arbitrary C++ declaration. Basically, most of the classes defined in this package are just “set of properties”.

declaration_t is the base class of the declaration hierarchy. Every declaration has parent property. This property keeps a reference to the scope declaration instance, in which this declaration is defined.

The scopedef_t class derives from declaration_t. This class is used to say - “I may have other declarations inside”. The “composite” design pattern is used here. class_t and namespace_t declaration classes derive from the scopedef_t class.

parser package

Please take a look on parser package UML diagram . Classes defined in this package, implement parsing and linking functionality. There are few kind of classes defined by the package:

  • classes, that implements parsing algorithms of GCC-XML generated XML file
  • parser configuration classes
  • cache - classes, those one will help you to eliminate unnecessary parsing
  • patchers - classes, which fix GCC-XML generated declarations. ( Yes, sometimes GCC-XML generates wrong description of C++ declaration. )

Parser classes

source_reader_t - the only class that have a detailed knowledge about GCC-XML. It has only one responsibility: it calls GCC-XML with a source file specified by user and creates declarations tree. The implementation of this class is split to 2 classes:

  1. scanner_t - this class scans the “XML” file, generated by GCC-XML and creates pygccxml declarations and types classes. After the xml file has been processed declarations and type class instances keeps references to each other using GCC-XML generated ids.
  2. linker_t - this class contains logic for replacing GCC-XML generated ids with references to declarations or type class instances.

Both those classes are implementation details and should not be used by user. Performance note: scanner_t class uses Python xml.sax package in order to parse XML. As a result, scanner_t class is able to parse even big XML files pretty quick.

project_reader_t - think about this class as a linker. In most cases you work with few source files. GCC-XML does not supports this mode of work. So, pygccxml implements all functionality needed to parse few source files at once. project_reader_t implements 2 different algorithms, that solves the problem:

  1. project_reader_t creates temporal source file, which includes all the source files.
  2. project_reader_t parse separately every source file, using source_reader_t class and then joins the resulting declarations tree into single declarations tree.

Both approaches have different trades-off. The first approach does not allow you to reuse information from already parsed source files. While the second one allows you to setup cache.

Parser configuration classes

gccxml_configuration_t - a class, that accumulates all the settings needed to invoke GCC-XML:

file_configuration_t - a class, that contains some data and description how to treat the data. file_configuration_t can contain reference to the the following types of data:

  1. path to C++ source file

  2. path to GCC-XML generated XML file

  3. path to C++ source file and path to GCC-XML generated XML file

    In this case, if XML file does not exists, it will be created. Next time you will ask to parse the source file, the XML file will be used instead.

    Small tip: you can setup your makefile to delete XML files every time, the relevant source file has changed.

  4. Python string, that contains valid C++ code

There are few functions that will help you to construct file_configuration_t object:

  • def create_source_fc( header )

    header contains path to C++ source file

  • def create_gccxml_fc( xml_file )

    xml_file contains path to GCC-XML generated XML file

  • def create_cached_source_fc( header, cached_source_file )

    • header contains path to C++ source file
    • xml_file contains path to GCC-XML generated XML file
  • def create_text_fc( text )

    text - Python string, that contains valid C++ code

Cache classes

There are few cache classes, which implements different cache strategies.

  1. file_configuration_t class, that keeps path to C++ source file and path to GCC-XML generated XML file.
  2. file_cache_t class, will save all declarations from all files within single binary file.
  3. directory_cache_t class will store one index file called “index.dat” which is always read by the cache when the cache object is created. Each header file will have its corresponding *.cache file that stores the declarations found in the header file. The index file is used to determine whether a *.cache file is still valid or not (by checking if one of the dependent files (i.e. the header file itself and all included files) have been modified since the last run).

In some cases, directory_cache_t class gives much better performance, than file_cache_t. Many thanks to Matthias Baas for its implementation.

Warning: when pygccxml writes information to files, using cache classes, it does not write any version information. It means, that when you upgrade pygccxml you have to delete all your cache files. Otherwise you will get very strange errors. For example: missing attribute.


Well, GCC-XML has few bugs, which could not be fixed from it. For example

namespace ns1{ namespace ns2{
    enum fruit{ apple, orange };
} }
void fix_enum( ns1::ns2::fruit arg=ns1::ns2::apple );

GCC-XML will report the default value of arg as apple. Obviously this in an error. pygccxml knows how to fix this bug.

This is not the only bug, which could be fixed, there are few of them. pygccxml introduces few classes, which knows how to deal with specific bug. More over, those bugs are fixed, only if I am 101% sure, that this is the right thing to do.

utils package

Use internally by pygccxml. Some methods/classes may be still usefull: loggers, find_xml_generator


That’s all. I hope I was clear, at least I tried. Any way, pygccxml is an open source project. You always can take a look on the source code. If you need more information please read API documentation.