\documentclass[A4paper]{article}

\usepackage[T1]{fontenc}
\usepackage[latin1]{inputenc}
\usepackage{latexsym}
\usepackage{amsmath}
%Police pour generer des pdf sans defauts
\renewcommand{\sfdefault}{phv}
\renewcommand{\rmdefault}{ptm}
\renewcommand{\ttdefault}{pcr}

\title{How to use C code in Free Pascal projects}
\author{G. Marcou, E. Engler, A. Varnek}

\bibliographystyle{plain}
\begin{document}

\maketitle\footnote{Universit\'e Louis Pasteur,\\ Institut de Chimie de Strasbourg,\\ 4 rue Blaise Pascal,\\ 67000 Strasbourg,\\ FRANCE}

\section{Introduction}

Free Pascal provides a robust, portable and powerful compiler\cite{FPDoc}. On
the other hand, the C language benefits of a very long life time and
therefore of millions of lines of codes. Hopefully, the Free Pascal
compiler supports linking with objects and shared objects compiled in
C. Reusing this material is therefore often desirable in Free Pascal projects.

Furthermore, the C++ provides also a very large number of very useful
libraries. But it is not possible yet to access them from Free Pascal
directly\cite{FPDoc,FPFor}. Therefore, it is necessary to write in C code,
procedures and functions to access the C++ ones. Linking Free Pascal to C code
is therefore even more important since C will act as a ``glue'' between
Free Pascal code and C++ -or other languages.

This article summerizes our experience about this field
It will be extended, I hope, by ourself as we continue to gain
experience with this problem, and by others.

\section{Dangers of mixing code}

The support of other languages is not a characteristic of any
language. It is a property of compilers. Therefore, instructions for
doing so are dependent of the runtime environment. This includes the system
and, sometimes, the compiler chosen. Characteristics of the definitions of the
languages also plays a role \cite{C++}.

In the following, the environment used is the Free Pascal Compiler (fpc
2.0.4) \cite{FPDoc} and the GNU Compiler Collection (gcc 4.1.2) \cite{gcc}.

Incompatibilities will play a role at nearly every stage of the -executable-
binary generation. The less dangerous are those that will prevent compilation
or linking. The most delicate ones will generate unpredictable behavior. For
example, it is up to the user to correctly convert from C or C++ types to Free
Pascal. An error in this conversion will not necessarily generate compilation
or linking errors, neither the executable will crash immediately. Such bug can
become extremly difficult to spot.

Free Pascal compiler is meant to produce as straight as possible, binaries and
executable code \cite{FPDoc}. This leads to some differences compared to GCC
compilers suite. The main difference is that Free Pascal users should not
have to deal with a Makefile. Therefore, they should not have to compile the
program by modules and then design a linking section.

This explains why the pre-compiler will have to deal with commands dedicated to
the linker and it helps understand the logic with the \verb|Unit| key word of
Free Pascal.

\section{Hello in C}

This section is a tutorial showing, in a simple case, how it is possible to
use C written objects in a Free Pascal project.

\subsection{Hello object}

The first thing to do is to produce an C written library to be compiled as a
static object. Static objects, for the Linux world, are a legacy of the
\verb|a.out| binary format. Objects are pieces of code already compiled that
can be reused in other programs. Linking with a program which calls a function
or procedure in an object file must be complete before execution time
\cite{Drep2006}.

Here is proposed a small C code as a classical ``hello'' example. First a
small header (file \verb|chello.h|) defines what is the library composed of.

\begin{verbatim}
#ifndef CHELLO_H
#define CHELLO_H
#include <stdio.h>
void PrintHello();
void PrintHelloS(int);
#endif
\end{verbatim}

This is not compulsory for C, but header files are highly encouraged since
they simplify interface and helps linking the objects. Here are defined two
procedures that write a ``hello'' message the second passing an integer to be
written. The body of the program follows (file verb|chello.c|).

\begin{verbatim}
#include "chello.h"
void PrintHello(){
  printf("Hello\n");
  return;
};
void PrintHelloS(int nme){
  printf("Hello\n");
  printf("%i",nme);
  return;
};

\end{verbatim}

This library is compiled using the command \verb|gcc -c chello.c|. The result
of the compilation is a binary file \verb|chello.o|.

\subsection{Free Pascal wrapping}

In a C compatible world, it should be possible to pass the C header and let
the compiler link the library definition of the procedures in the object files
to the main program. Free Pascal does not understand C, so the header has to
be rewritten in standard Free Pascal \cite{FPDoc,FPFor}. This is the purpose
of the following Free Pascal unit (file helloU.pas), wrapping the C written
object in a standard Free Pascal code:

\begin{verbatim}
unit helloU;

{$link chello.o}
{$linklib c}

interface

uses CTypes;

procedure PrintHello; cdecl; external;
procedure PrintHelloS(nme : ctypes.cint32); cdecl; external;

implementation

end.
\end{verbatim}

Several parts shall be noted. First there is only definition of the
procedures. The implementation is empty. Even more, the definition is followed
by the \verb|external| keyword. This means the Free Pascal compiler should not
wait for the body of the procedure since it shall find it at linking
time \cite{FPDoc,FPFor}.

Second, the CTypes unit is called by the command \verb|uses CTypes|. This unit
is poorly documented, but its purpose is precisely to help for correct
conversion between Free Pascal native types and C native types. In fact, it is
quite difficult: the numerical limits of types like integers or floats are not
defined by a language but by the compilers. As pointed by
D. Mantione\footnote{a regular contributor to Free Pascal}, with GCC, an
\verb|int| is \$80000000..\$7fffffff (32 bits), while a Pascal integer is
defined as \$8000..\$7fff (16 bits). Thus, it would be required to use
\verb|longint| Free Pascal type to match it. Here we call the
\verb|ctypes.cint32| type converter. This method shall be prefered.

For type conversion of C types it is usefull to take a look at the
C/C++ header \verb|limits.h|. On Linux systems it is usually located in
\verb|/lib/include/|. For Free Pascal conversion, it shall be safer to take a
look to the documentation \cite{FPDoc,FPFor}.

Then, there is two instructions for the linker. The first one
\hspace{0cm}\verb|{$link chello.o}| %$
 means that the linker should link the program with the \verb|chello.o|
file \cite{FPDoc,FPFor}. So the Free Pascal compiler will never know about
what are truly those procedures, it will be the job of the linker to find
them in the object file.

Doing so, a lot of references will appear to some functions typical of C. For
example, it will find references to the \verb|print| function and others
defined in the \verb|<stdio.h>| included. Those are defined in a shared
library file, \verb|libc.so|.

Shared libraries are libraries that are not linked at linking time. So the
linking of the program is not complete. The program will not run until it
will have found the definition of, at least, the standard C functions. In
fact, at execution time, a special program, \verb|ld-linux.so| is called to
charge in memory the binary definition of the missing procedures
\cite{Drep2006}.

The linker needs to know at compile time, where to find not yet defined
functions and procedure. First, to be able to build in the program binary
instructions relative to which shared library has to be charged in
memory. Second to test that at least, at linking time, every single procedure
and function will be defined.

Here we tell the linker to search in the file \verb|libc.so| for
missing definitions. It is the purpose of the instruction \verb|{$link c}|.%$
The linker will add a prefix \verb|lib| and a suffix \verb|.so|, in the Linux
world to find the correct shared object file \cite{FPDoc,FPFor}.

It is usually necessary to use also the \verb|{$link gcc}|%$
to link with \verb|libgcc.so| if the C objects were created with
GCC. Nonetheless this library might not have been produced
during GCC compilation and therefore can be missing. Therefore, linking errors
might occurs if this linking statement is present as well as if this statement
is missing.

One last thing, is that the Free Pascal definition has a different protocol as
how to pass arguments in functions and procedures compared to C. This is what
the keyword \verb|cdecl| does. The \verb|cdecl| warns the compiler and
the linker that the arguments are to be understood as if they were in
C\cite{FPDoc,FPFor}. If
this keyword is omitted it is possible that the program will compile and link
but as soon as procedure will be called, it will crash. It might as well not
compile any more if one of the wrapped procedures or function is called.

\subsection{Free Pascal program}

Now it is time to get a program that will call our C written functions. This
is done in the file \verb|hello.pas|:

\begin{verbatim}
uses
  helloU;

begin
   PrintHello;
end.

\end{verbatim}

It is compiled using the command \verb|fpc hello.pas|. This Free Pascal code
only knows about the Free Pascal wrapper of the C object. So the compiler will
compile first the \verb|helloU| unit, then the main program. This part will
produce \verb|.ppu| and \verb|.o| files. The first ones are Free Pascal
binaries and the second ones, a translation in the \verb|a.out| Linux binary
format.

The linker is called that will link statically the \verb|.o| it knows the
existence of. It will then generate an executable binary with all informations
about the shared libraries to be dynamically loaded in memory at execution
time.

\section{Hello in C++}

This section is a tutorial showing how, in a simple case, it is possible to
use C++ written objects in a Free Pascal project.

\subsection{Hello object, again...}

In this hello project in C++, a class hello is defined. This class
contains a private structure and several public methods to access it or to
perform some actions. Here is the header, in the file \verb|hello.hpp|:

\begin{verbatim}
#ifndef HELLO_HPP
#define HELLO_HPP
#include<iostream>
#include<string>
using namespace std;
class hello{
public:
  hello(){
    string vide="";
    hello::SetName(vide);
  };
  hello(string &name){
    hello::SetName(name);
  };
  void HelloPrint();
  void SetName(string&);
  string GetName();
private:
  string name;
};
#endif

\end{verbatim}

The private structure is a C++ string containing a name. There is a
constructor to initialize this string to the null string and an overloaded
constructor to initialize it to an arbitrary value.

The other public methods include setting or retrieving the value of the
private structure. The last method just prints in the current console a
friendly message.

The body of the methods defined in the header is in the following file,
\verb|hello.cpp|:

\begin{verbatim}
#include "hello.hpp"
void hello::HelloPrint(){
  cout << "Hello " << GetName() << endl;
  return;
};
void hello::SetName(string &str){
  hello::name=str;
  return;
};
string hello::GetName(){
  return(hello::name);
};

\end{verbatim}

This set of file, the header and the body of the methods, is written in pure
C++. It is compiled using the command \verb|g++ -c hello.cc|. The compilation
provide an object file \verb|hello.o| in which there is many references to C
and C++ libraries, in particular standard template libraries -for the C++
string.

As stated before, Free Pascal cannot handle C++ objects
\cite{FPDoc,FPFor}. They have a very particular structure since they are
linked to C++ shared libraries and here, have class structure. It is therefore
necessary to build a C interface between this C++ code and Free Pascal.

\subsection{C wrapping}

As the C language does not know about C++ object oriented structure it is not
possible to keep it. This means that the main job of the C wrapper will be to
flatten the C++ class structure to a set of C functions
\cite{Rudy2006}. 

Another important point is that C does not know about C++ standard template
libraries and evolved structures like C++ strings, vectors, maps, etc. So
another function of the wrapper will be to cast those C++ classes in standard
C types \cite{Cline2006}.

To do this, it is useful to create a special header, as shown in the file
\verb|chello.h|:

\begin{verbatim}
#ifndef FHELLO_H
#define FHELLO_H
#ifdef __cplusplus
#include "hello.hpp"
#define EXPORTCALL __attribute__((stdcall))
typedef hello *helloHandle;
#else
typedef struct hello *helloHandle;
#define EXPORTCALL
#endif

#ifdef __cplusplus
extern "C"
{
#endif
  extern helloHandle EXPORTCALL NewHello();
  extern helloHandle EXPORTCALL NewHelloC(char*);
  extern void EXPORTCALL DeleteHello(helloHandle);
  extern void EXPORTCALL HelloPrint(helloHandle);
  extern void EXPORTCALL SetName(helloHandle, char*);
  extern const char* EXPORTCALL GetName(helloHandle);
  //
  extern helloHandle EXPORTCALL cNewHello();
  extern helloHandle EXPORTCALL cNewHelloC(char*);
  extern void EXPORTCALL cDeleteHello(helloHandle);
  extern void EXPORTCALL cHelloPrint(helloHandle);
  extern void EXPORTCALL cSetName(helloHandle, char*);
  extern const char* EXPORTCALL cGetName(helloHandle);
#ifdef __cplusplus
}
#endif
#endif/*FHELLO_H*/
\end{verbatim}

This header can be interpreted both by a C and a C++ compiler. And two body
files will be implemented, one in C++ code and the other one in standard C
code.

There is several instruction to help to deal with this. The first one is to
use the macro definition \verb|#ifdef __cplusplus| and the corresponding
\verb|#endif|. It is true if the compiler analyzing the code is a C++ compiler
and false otherwise \cite{C++,gcc}. This macro definition is widely spread
among compilers but it is not part of any standard. Therefore, it might be
necessary to define it manually through an option of the compiler.

This macro definition first protects the C++ header that needs to be included
to define the hello object. If it is not protected, a C compiler will refuse the
code since it would try to analyze the C++ code of the \verb|hello.hpp|
header.

The second statement that is protected so that only a C++ compiler can see it,
is the definition of the \verb|EXPORTCALL| to
\verb|__attribute__((stdcall))|. This value is interpreted by GCC and ensures
the argument passing of the function to be compatible with C definition. This
is why it is not set for the C compiler \cite{gcc,Rudy2006}.

The third protected statement only to be processed by the C++ compiler just
defines a new type \verb|helloHandle| as a pointer the \verb|hello|
class.

For the C compiler, \verb|helloHandle| will still be a pointer to a structure
\verb|hello| that is not yet defined. This definition is of course protected
only to be seen only by the C compiler so that it does not interfere with the
C++ instructions \cite{Rudy2006}.

Another important difference of behavior whether the header has to be
processed by the C or the C++ compiler comes form the instruction 
\verb|extern ''C''{}|. This is a C++ instruction and as such needs to be
hidden for the C compiler. This instruction specifies a linking convention:
how to access memory, integrated type formats, etc. This command does not
target specifically the C language: it is also used to link Fortran to C++ for
example \cite{C++}.

In the following, it shall be noted that each method of the \verb|hello| class
is replaced by a function. To allow the function to call a method of the class
for a given instance of the class, at least a pointer to this instance shall
be passed. Constructors are an exception since they only receive this pointer.

The last unusual point of this header is the double definition of all the
methods. It is so to cast C++ types in C standard types. In our example, the
methods of the \verb|hello| class uses C++ strings that cannot be understood
by C. As the functions definitions has to be understood by both C and C++
compiler, we had to pass only C types as argument of the functions -a
\verb|char*|. 

Functions written with a \verb|c| prefix will be fully
written in C. Therefore, they can only make calls to C functions. Functions
that has not the prefix will have a body in a separate file, written in C++.

The C body of the functions is the file \verb|fhello.c|:

\begin{verbatim}
#include "fhello.h"
helloHandle cNewHello(){
  return(NewHello());
};
helloHandle cNewHelloC(char* nme){
  return(NewHelloC(nme));
};
void cDeleteHello(helloHandle h){
  return(DeleteHello(h));
};
void cHelloPrint(helloHandle h){
  return(HelloPrint(h));
};
void cSetName(helloHandle h, char* nme){
  return(SetName(h,nme));
};
const char* cGetName(helloHandle h){
  return(GetName(h));
};

\end{verbatim}

It can be compiled with the command \verb|gcc -c fhello.c -o fhello_c.o|. The
C++ body of the functions is in the file \verb|fhello.cc|:

\begin{verbatim}
#include "fhello.h"
extern "C"{
helloHandle EXPORTCALL NewHello(){
  return new hello;
};
helloHandle EXPORTCALL NewHelloC(char *nme){
  string str(nme);
  return new hello(str);
};
void EXPORTCALL DeleteHello(helloHandle handle){
  delete handle;
};
void EXPORTCALL HelloPrint(helloHandle handle){
  handle->HelloPrint();
};
void EXPORTCALL SetName(helloHandle handle, char *nme){
  string str(nme);
  handle->SetName(str);
};
\end{verbatim}

Here again the \verb|extern "C"{}| statement specifies the linking
convention \cite{C++}. The body of the functions is obviously C++ so to
cast C++ types and classes into standard C types. In
this example, there is commands like \verb|string str(nme)| while \verb|nme|
is a \verb|char*|.

This code can be compiled using the command
\\\verb|g++ -c fhello.cc -o fhello_cc.o|.

Following this procedure, it shall be noted that on contrary of what is
sometime said \cite{Cline2006}, it is not necessary to compile the C wrapper
with the C++ compiler. The final program can be linked using the two separate
objects created.

\subsection{linking}

The task of integrating those methods in Free Pascal is not over
yet. Compilation of C++ objects is another compiler dependent task. In this
tutorial, \verb|gcc| is used as compiler. During compilation of this set of
compiler, a shared version of the \verb|libgcc| might not be compiled
\cite{gcc}. Therefore, it might not be possible to find such version to link
the Free Pascal code with the C++ object. In brief, this task might be
possible but can be more difficult than the one of linking with pure C written
objects.

Another solution is to link the C++ and the C objects in advance into a shared
library. This can be done with the command
\\\verb|gcc -shared -o libhello.so hello.o fhello_cc.o|.
Here, it is the \verb|gcc| compiler tools that
will find and set the correct dependencies of the libraries.

The shared library can be used as before in the Free Pascal Code.

\subsection{Free Pascal wrapping}

The Free Pascal wrapper is a Free Pascal unit defining the functions and
procedures to be called in a Free Pascal project.

\begin{verbatim}
unit helloFLAT;

interface

uses
sysutils;

type
   helloHandle = type pointer;

function cNewHello:helloHandle; cdecl;
function cNewHelloC(nme :PChar):helloHandle; cdecl;
procedure cDeleteHello(handle : helloHandle); cdecl;
procedure cHelloPrint(handle : helloHandle); cdecl;
procedure cSetName(handle : helloHandle;nme : PChar); cdecl;
function cGetName(handle : helloHandle):PChar ; cdecl;

implementation

{$link fhello_c.o}
{$linklib hello.so}
{$linklib c}
{$linklib stdc++}

function cNewHello:helloHandle; cdecl; external;
function cNewHelloC(nme :PChar):helloHandle; cdecl; external;
procedure cDeleteHello(handle : helloHandle); cdecl; external;
procedure cHelloPrint(handle : helloHandle); cdecl; external;
procedure cSetName(handle : helloHandle;nme : PChar); cdecl; external;
function cGetName(handle : helloHandle):PChar ; cdecl; external;

end.

\end{verbatim}

It shall be noted that the functions and procedures have to be declared with
the \verb|cdecl| key word. It ensures compatibility between C and Free Pascal
declarations of functions and procedures arguments \cite{FPDoc,FPFor}. Only
the C functions are defined.

Those declarations need also a new type to be defined:
\verb|helloHandle|. Free Pascal does not really need to know about the true
nature of this type: it will be linked to the C pointer \cite{Rudy2006}. As
stated by D. Mantione, with this definition, \verb|helloHandle| is not
compatible to a pointer, but still is a pointer, which is exactly what we want.

Implementation of the functions and procedures are referred as
\verb|external|. This means that the Free Pascal compiler knows it has to wait
to find them: they will be defined only at linking time or later
\cite{FPDoc,FPFor}.

The \verb|cdecl| instruction guarantees the argument passing protocol to be
compatible with C style. It should also be noted that to correctly map the
C style \verb|char*| type, the Free Pascal \verb|PChar| type was
used. Failing to map correctly the C type can be not detected at compilation
or linking time, leading the generated binary to crash.

At last, as explained before, two sets of pre-processor instructions are
defined. The first \verb|{$link fhello_c.o}|,%$
 link the C definition and the
Free Pascal definition of the procedures and functions with a \verb|c| prefix
\cite{FPDoc,FPFor}.

The second set of pre-processor instructions, \verb|{$linklib hello.so}|,%$
\\\verb|{$linklib c}|,%$
 \verb|{$linklib stdc++}|,%$
 pass instructions to the
linker so that is can link the C functions with the C++ implementations of
functions and class \cite{FPFor,FPDoc}. It also performs the link to the
standard \verb|libc| and \verb|libc++| libraries.

\section{Free Pascal program}

An application resulting of this work can be found in the file
\verb|helloP.pas|:

\begin{verbatim}
uses
  helloFLAT;

var
   h : helloHandle;

begin
   h:=cNewHelloC('Gilles');
   cHelloPrint(h);
   cDeleteHello(h);
end.
\end{verbatim}

It uses the \verb|helloFLAT| unit. Thus it uses the newly defined type
\verb|helloHandle|. It will use it to construct an instance of the
\verb|hello| class for which the private string structure is immediately
set. It will use it to print a message and clean the memory before leaving.

\subsection{Conclusion}

This small tutorial will help new comers to Free Pascal to build applications
that are able to take advantage of the huge libraries available in C and
C++. This shall help saving time as first not having to write again what was
already written in a quite optimal manner. Besides, reusing code prevent the
need of testing it. In fact, the Borland's Kylix project (that is by now, cold
dead) was build on the TrollTech's QT libraries \cite{Delphi}. Those libraries
were originally written in C for use in C++ code.

There exists some Free Pascal utilities developed for helping designing
correct wrappers. This is precisely the goal of the utility \verb|h2pas|
\cite{FPDoc,FPFor}. It is nonetheless essential to get knowledge of the
details of the protocol to be able to use it correctly.

This tutorial is not complete. For example, there shall be a way to use the
C++ class structure into Free Pascal without flattening it. It has been
reported previously for Delphi7 and Windows \cite{Rudy2006}.

Some work also has to be done to check how exceptions can be handled through
those mixed code.

\bibliography{CinFreePasca}
\end{document}