libt2n: (tomj) fixed call of virtual function close() from destructor, fixed return...

[libt2n] / doc / index.doc

/*! \mainpage libt2n - (talk to neighbor)
 \section intro_sec Introduction
 \par
 libt2n (talk to neighbor) is a C++ library for inter-process communication (IPC \ref notes1 "[1]") with an additional code generator (libt2n-codegen) to make C++ remote procedure calls (RPC \ref notes2 "[2]")  simple.
 \par
Using libt2n and its code generator only the server procedure implementations have to be written manually. The stubs are generated by the code generator and the inter-process communication routines are provided by libt2n. libt2n provides an abstract interface for
	\ref libt2n::connection "connections"
 hiding the details of the concrete communication mechanism used. At the moment it is implemented for local sockets (a.k.a "Unix domain sockets") and TCP-sockets. (\ref libt2n::socket_client_connection, \ref libt2n::socket_server_connection)
 \par
 The client procedure stubs provide an interface to the server procedure implementations. To call the server procedure the client stub procedure is called. The client stub procedure transforms the procedure arguments and signature including the procedure "name" into a sequence of bytes using boost serialization \ref notes6 "[6]". This byte sequence represents a request to call a procedure with those arguments). Using the communication routines of libt2n the request is sent to the server. The server procedure stubs de-serialize the request and call the corresponding server procedure using the reconstructed arguments. The result of the server procedure is serialized and sent back to the client. Finally the result is de-serialized on client side again and returned to the caller.
 \par
 The input for the code generator is standard C++ code
	\ref notes4 "[4]"
 and you mark the procedures you want to expose to other processes. Because the interface is described in C++ directly there is no need for a interface description language (IDL, \ref notes5 "[5]").
 The code generator is used to create the client and server stubs for the marked procedures. The procedures can be grouped and each group maps to two classes (cmd_group_... and cmd_group_..._client). The cmd_group_..._client class provides the interface to call remote procedures. Each remote procedure maps to a method of this class and the constructor takes a connection object establishing the binding to the remote side. For each group the code generator is called and generates 6 output files: group_common.hxx, group_common.cpp, group_client.hxx, group_client.cpp, group_server.hxx, group_server.cpp. The _common files are used by client and server whereas the _client files contain the client stub code (the cmd_group_..._client class) and the _server files the server stub code. 
 \par
 To simplify the build process a Makefile snippet is provided that allows to create a server program and a client library (including a corresponding .pc file) using the autotools easily.
 \section License
 The libt2n libaray is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License version 2.1 as published by the Free Software Foundation. The code generator (libt2n-codegen) is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

 \section install_sec Installation

 \subsection requirements Requirements
 \par libt2n
 - boost <http://www.boost.org/> (serialization <http://www.boost.org/libs/serialization/>)
 \par libt2n-codegen
 - gccxml <http://www.gccxml.org>
 - libxmlpp <http://libxmlplusplus.sourceforge.net/>
 \par client-wrapper
 - boost <http://www.boost.org/> (bind <http://www.boost.org/libs/bind/>, function <http://www.boost.org/libs/function/>)
 - gcc 4.1 or newer <http://gcc.gnu.org/>

 \subsection recommended Recommended
 - pkg-config <http://pkgconfig.freedesktop.org/wiki/>
 - autotools (autoconf <http://www.gnu.org/software/autoconf/>, automake, libtool)

 \subsection Compilation
 \verbatim
 ./configure && make check && make install
 \endverbatim

 \subsection Usage
 Take a look at the \ref example.

 \subsection Homepage
 The libt2n homepage with downloads and further information can be found at <http://www.intra2net.com/de/produkte/opensource/libt2n/>.

 \subsection References
 - \anchor notes1
 [1] inter-process communication (IPC), http://en.wikipedia.org/wiki/Inter-process_communication
 - \anchor notes2
 [2] remote procedure call (RPC), http://en.wikipedia.org/wiki/Remote_procedure_call and "THE RPC MODEL" http://www.faqs.org/rfcs/rfc1050.html
 - \anchor notes3
 [3] Figure: Remote procedure call overview, http://jan.newmarch.name/webservices/rpc_stub.png , Jan Newmarch "Web services" http://jan.newmarch.name/webservices/tutorial.html
 - \anchor notes4
 [4] in fact gccxml is used to parse the C++ code and the XML output of gccxml is used as input for the code generator
 - \anchor notes5
 [5] interface description language (IDL), http://en.wikipedia.org/wiki/Interface_description_language
 - \anchor notes6
 [6] boost serialization, http://www.boost.org/libs/serialization/
*/

/*! \page example Usage example

 In this example we create two packages using the autotools:
 - server program and client library to connect to the server. The server exports a simple procedure using one group: "t2nexample"
 - client program using the library

 \section server Example server program and client library

 \par The procedure to export (input for the code generator - libt2n-codegen): t2nexample.cpp:
 First the procedure to export is defined. It is marked for export with the attribute macro \ref LIBT2N_EXPORT. In this example the procedure throws an exception if the input string is "throw". The exception is passed back to the client transparently. Otherwise some text is appended and returned.
 \include example1/t2nexample.cpp

 \par Required includes go into the group header file: t2nexample.hxx:
 All includes required to get the declarations of the types used by the procedures to export go into the group header file. libt2n uses boost for serialization. This means all types involved in a remote procedure call must be boost serializable. In this example we only use std::string provided by \<string\> and boost already provides serialization for std::string in the boost/serialization/string.hpp header file.
 \include example1/t2nexample.hxx

 \par The server program: server.cpp:
 We have to provide the main entry point for the server program. In this example we use a local socket and the server program simply waits until a request is received which then is handled by the generated code directly. We use a \ref libt2n::group_command_server to handle the requests. Each \ref libt2n::group_command_server is
tied to one command group. Which source file exports it's functions into which command group is defined in the Makefile (see below).
 \include example1/server.cpp

 \par Using autoconf and automake to build a example server program and a client library.
 In the configure.in(.ac) we check for libt2n using the LIBT2N_CHECK m4 macro provided by libt2n.
 \verbinclude example1/configure.in
 Writing the Makefile.am isn't difficult either. We have to list the command groups used. For each command group we have to list the C++ source files with the procedures to export. For each group we build a client library by listing the generated client code (group_client.cpp) in a corresponding libtool library target. The .pc file for the library is generated automatically. The sources of the server program must include the generated server code (group_server.cpp), the file with the main entry point and of course the procedure definition.
 \verbinclude example1/Makefile.am

 \par Build and install the package
 To build and install the package we first have to create the configure script and the other help scripts of the autotools by running autoreconf.
 \verbatim
 autoreconf -f -i && ./configure && make install
 \endverbatim

 \section client Client using the library
 Using the library is as simple as using any other library using pkg-config (the pkg-config .pc file is created automatically by the included Makefile snippet)
 \par We only have to check that the library is installed
 \verbinclude example1-client/configure.in
 \par The Makefile.am needs nothing special
 \verbinclude example1-client/Makefile.am
 \par The client program: client.cpp:
 The example client first connects to the local socket. The connection is passed to the constructor of the generated class. To call the remote procedure the "testfunc" method is called. The example first passes "throw" to the remote procedure which will result in a exception to be thrown which is passed back to the client and thrown on the client side again. In the example the exception is caught and it is checked whether the string returned by what() is correct. If so a second remote procedure call is made and its return value is checked. Only if both tests succeed the program will exit with a status value indicating success.
 \include example1-client/client.cpp

 \par Build and install the package
 \verbatim
 autoreconf -f -i && ./configure && make install
 \endverbatim

 \par Test
 To test whether it works we first start the server that creates a socket 'socket' in the current working directory. Then we run the client and print "ok" if it exited with a status value indicating success.
 \verbatim
$ cd /tmp
$ file socket
socket: cannot open `socket' (No such file or directory)
$ libt2n-example1-server &
[1] 7711
$ file socket
socket: socket
$ libt2n-example1-client && echo ok
ok
$ kill %1
$ rm socket
 \endverbatim

 \section wrapper The Client-Wrapper
 The interfaces can be called directly in the way outlined above. But this means you have to take care of connection errors between client and server
 at each call, possibly try to reconnect and so on. Libt2n provides the Client-Wrapper to ease this. It is a way to select a error handling strategy
 once and use it automatically for all calls invoked through the Wrapper. Tough error-handling is the common usecase, the Client-Wrapper could be used
 to execute any user-provided code before and after a call to the server is made.

 The other feature that the Client-Wrapper provides is a connection-singleton. T2n (currently) only offers single-threaded servers. So if you use methods of a T2n-server in a program, you usually only want to maintain one common connection to this server - even if it is accessed from different parts/modules/classes/... of your program. The Client-Wrapper is initialized with a \ref libt2n::ConnectionWrapper.

 This \ref libt2n::ConnectionWrapper takes the error-handling strategy (e.g. reconnect-then-throw) and everything needed to establish a connection (e.g. socket name or host and tcp-port) as parameters. A connection is established at the first actual request to the server and re-used for following requests. You don't need to pass around client-handles and the like to your classes or methods, finding the right wrapper is done via the global singleton created for each server-interface initialized for the wrapper.

 This example shows how to use the Client-Wrapper:

 \include example1-client-wrapper/client.cpp

The details of the Client-Wrapper can be found in the \ref libt2n::T2nSingletonWrapper, but beware, the code is full of ugly templates and template-construction-defines.

*/