james - A simple XML schema compiler for C++, written in C++

Purpose/goals
=============

The purpose of this program is to transform a subset of XSL into C++ code for marshalling and unmarshalling documents conforming to one or more schemas.
The generated code should not be needlessly complicated (no getters or setters).
The number of dependencies should be kept to a minimum.

Compilation
===========

james makes use of CMake.
This allows project files and Makefiles to be generated for a variety of system without having to maintain said files.
The program is known to compile in *NIX like system (GNU/Linux, Mac OS X) via Makefiles and Windows via Visual Studio 2008.

Dependencies
------------

james depends on Xerces-C++ 2.8.x or 3.1.x.

Building
--------

The program is built like a typical CMake project.
A normal build and install will look something like this (output omitted):

 ~/james$ mkdir build
 ~/james$ cd build
 ~/james/build$ cmake ..
 ~/james/build$ make
 ~/james/build$ sudo make install

Note that this also installs libjames, which is required by the generated code.
This requirement may disappear in the future, at which time the only install will be the program.

A short guide to usage
======================

Running the program without arguments produces the following usage information:

 USAGE: james [-v] [-d] [-nr] [-nv] [-a] [-cmake targetname] [--dry-run] output-dir list-of-XSL-documents
  -v         Verbose mode
  -d         Generate default constructors
  -nr        Don't generate constructors taking required elements
  -nv        Don't generate constructors taking required elements and vectors
  -a         Generate constructors taking all elements
  -cmake     Generate CMakeLists.txt with all generated .cpp files as part of a library with the specified target name
  --dry-run  Perform generation but don't write anything to disk - instead does exit(1) if any file changes

By default james generates at most two constructors for the generated classes.
The first constructor takes all required elements. The intent is to not allow the user to forget setting required elements.
The second constructor includes all vector element in addition to the required elements. This is meant more as a convenience.
No default constructors are generated, except for types lacking members - for those there is only one constructor, which is the default cosntructor.
The -d switch is used to tell james to generate defalult constructors for all classes.
The -a switch also adds constructors taking optional elements. It is not very useful though, hence why they're disabled by default.

The -cmake switch is used to generate a CMakeLists.txt file along with the .h and .cpp files.
This CMake file turns the loose collection of files into a static library, suitable for inclusion in a CMake project via add_subdirectory().

Finally, the --dry-run switch can be used to test whether running james would change any files on disk.
This includes modifying existing files or creating new ones, but not no longer generating files for types that no longer exist.
The latter is however detected if --dry-run is used together with the -cmake switch, since removing a type would change CMakeLists.txt.

Non-CMake build example
-----------------------

For a small project with a single schema, generating and compiling a library for handling said schema looks like this:

 ~/foo$ mkdir generated
 ~/foo$ james generated schema.xsd

Then make sure generated/*.cpp is somehow included in your build, and possibly that generated/ is part of your include path.
Your program also needs to link with libjames, since the generated code currently depends on it.
Note that outputing the files in the current directory is not recommended since that makes cleaning up difficult.

CMake build example
-------------------

If foo is a CMake project, then schema.xsd can be turned into a library.
This library, called libbar in this example, can be put in a subdirectory and add_subdirectory():ed once generated.
CMake does not yet seem able to depend on generated subdirectories, so libbar would have to be generated prior to running CMake for foo.
In other words, something like this:

 ~/foo$ cat CMakeLists.txt
 ...
 add_subdirectory(libbar)
 add_executable(foo main.cpp [...])
 target_link_libraries(foo bar libjames)
 ...
 ~/foo$ mkdir libbar
 ~/foo$ james -cmake bar libbar/ schema.xsd
 ~/foo$ mkdir build
 ~/foo$ cd build
 ~/foo/build$ cmake ..
 ~/foo/build$ make

After this ~/foo/build contains the binary foo, and ~/foo/build/libbar contains libbar.a.
A better solution would be to somehow have add_subdirectory() depend on a custom target that runs james and generated ~/bar/libfoo.
It is possible to hack something similar to this using the CMake module ExternalProject.

Using the generated code
========================

Here is a small example schema and code that makes use of the generated classes:

example.xsd
-----------

<?xml version="1.0" encoding="UTF-8"?>
<xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema" targetNamespace="http://example.com/" elementFormDefault="qualified" xmlns="http://example.com/">
  <xs:element name="PersonDocument" type="tns:PersonType"/>
  <xs:element name="PersonListDocument">
    <xs:complexType>
      <xs:sequence>
        <xs:element name="person" type="tns:PersonType" minOccurs="0" maxOccurs="unbounded"/>
      </xs:sequence>
    </xs:complexType>
  </xs:element>
  <xs:complexType name="PersonType">
    <xs:attribute name="name"      type="xs:string" use="required"/>
    <xs:attribute name="address"   type="xs:string" use="required"/>
    <xs:attribute name="birthYear" type="xs:int"    use="required"/>
  </xs:complexType>
</xs:schema>

person.xml
----------
<?xml version="1.0" encoding="UTF-8"?>
<PersonDocument name="Foo Oldperson" address="Bar Street 123" birthYear="1900"/>

example.cpp
-----------

#include <iostream>
#include <xercesc/util/PlatformUtils.hpp>
#include "generated/PersonDocument.h"
#include "generated/PersonListDocument.h"

int main() {
    //initialize Xerces-C++
    xercesc::XMLPlatformUtils::Initialize();

    //unmarshal personIn from stdin
    PersonDocument personIn(std::cin);

    //create a list containing personIn and some other person
    PersonListDocument list;
    list.person.push_back(personIn);
    list.person.push_back(PersonType("Some Otherguy", "Somewhere 999", 1985));

    //finally, marshal the list to stdout
    std::cout << list;

    //terminate Xerces-C++
    xercesc::XMLPlatformUtils::Terminate();

    return 0;
}

Compiling
---------

~/example$ mkdir generated
~/example$ james generated example.xsd
A generated/PersonDocument.cpp
A generated/PersonDocument.h
A generated/PersonListDocument.cpp
A generated/PersonListDocument.h
A generated/PersonListDocumentType.cpp
A generated/PersonListDocumentType.h
A generated/PersonType.cpp
A generated/PersonType.h
~/example$ g++ example.cpp generated/*.cpp -ljames -lxerces-c -o example

Running
-------

$ ./example < person.xml | xmllint --format -
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<PersonListDocument xmlns="http://example.com/">
  <person address="Bar Street 123" birthYear="1900" name="Foo Oldperson"/>
  <person address="Somewhere 999" birthYear="1985" name="Some Otherguy"/>
</PersonListDocument>

Limitations
===========

Some schemas, like that of XHTML or XSL itself, will never be supported by james.
The intent is to only support schemas that can be compiled into classes that consist of plain C++ objects.
Any schema that allows arbitrary order of elements in any type is explicitly not supported.

Motivation
==========

This program came about as an alternative to using xmlbeansxx in a project I was working on.
Using xmlbeansxx eventually became intolerable enough that it was easier to write a new tool than deal with its problems.
James is an attempt to address these problems, which at the time (possibly to this day) included;

* Complicated build system
 - Requires the Java SDK, Maven and Ant
 - Lacks of out-of-source build
* All code being put into a pair of giant .h/.cpp files (~1 MLOC each for a typical schema)
* scompxx always touching said file pair, regardless of whether their content would change or not
 - This leads difficulties using scompxx in a pre-build script
* Converting namespace URIs into a complicated structure of namespaces, similar to Java packages
 - The generated classes are put in said namespace, making for needlessly verbose code
* Didn't compile in Visual Studio
* Barely compiled in Cygwin (after patching)
* Producing very Java-like classes, like using getters, setters and factories without good reason

TODO
====

* Support more XSL features (within reason)
* Make switching between Xerces-C++ 2.8.x and 3.1.x easier
* Create output directory if it doesn't exist
* Some way of specifying one or namespaces into which the generated classes can be put
* Put libjames/* among generated output, removing the need to install or link with it