Tutorial on CALLS BETWEEN LANGUAGES
Here is a tutorial on inter-calling between all sorts of different languages
in Unix, courtesy of Chris Torek, U. of Maryland:
>What needs to be done (other than ensuring consistent data types
>for args) in order to link C and/or Fortran main programs and
>routines with each other?
I am going to attempt to produce the definitive article on mixed language
linking under 4BSD Vax Unix. Stand back!
First, there is the matter of names: The symbols in the object files must
match, that the linker may resolve the right references. Yet each compiler
has its own methods for mapping from source to object. Within one language we
may safely ignore this mapping; but when mixing tongues, it becomes important
indeed.
The C compiler takes any global symbol and prepends an underscore character,
`_'. Names are not limited in length---though in fact there is a limit of
about a thousand characters, no one seems to run into it. Thus
int global_var;
char *
somefunc()
{
...
generates the symbols `_global_var' and `_somefunc'.
The F77 compiler limits names to six characters, then prepends and appends an
underscore:
subroutine sub
integer var
common /com/ var
...
names the subroutine `_sub_' and creates a global `variable' containing one
integer. The `variable' is called `_com_'. Variables that are not part of a
common block do not have global names. F77 does not allow underscores in
source-level names: `subroutine sub_1' is illegal. The compiler also ignores
any PROGRAM name:
program prog
...
creates the symbol `_MAIN_'.
The Berkeley Pascal compiler strings together the names of all nested
procedures to concoct unique global names. Only variables defined in the
`program' part are global (no surprise here), and these names are constructed
in the same way as C's globals. However, the program name is ignored, and the
compiler uses the name `_program':
program foo; { symbol _program }
var v: integer; { symbol _v }
procedure proc; { symbol _proc }
function func; { symbol _proc_func }
begin func := 0 end; { end proc's func }
begin end; { end proc }
begin end. { end program }
generates the symbols `_program', `_v', `_proc', and `_proc_func'. (It also
generates the names `___proc_func' and `___proc', but we shall ignore these
for the moment.) The Pascal compiler does not permit source-level names to
contain `_': i.e., `procedure proc_a' is illegal.
It should be clear at this point that C programs can call any F77 or Pascal
subroutines (procedures) or functions, and that Pascal can call many C
routines---not all, for names with underscores are out of reach---while F77
routines can call only specially-named C routines, namely those that end with
an underscore, are less than seven other characters, and contain no internal
underscores. F77 and Pascal routines can never call each other directly.
Even with a compatible set of names, the task is not yet done. There remain
two problems, each bound up with the other. Every program must have an entry
point (`main'); and every language has its libraries. C's is the simplest of
the three, for its main looks like every other C routine and needs no
libraries not used by both F77 and Pascal as well. F77's main is a C routine
that initialises its I/O system, traps signals, and calls the program's
_MAIN_. Pascal's main is similar to F77's, but does not trap signals and
calls _program, not _MAIN_. Both F77's and Pascal's mains also save argc and
argv, F77's in `_xargc' and `_xargv' and Pascal's in __argc and __argv.
Now if you intend to call C routines from F77 or Pascal, and assuming that
these routines are entirely self-contained, all you need do is compile the C
code to object, and mention the `.o' file in the linking command. Of course,
you must also use the proper parameter passing conventions---but I anticipate.
Calling F77 or Pascal routines from C, however, is a bit more difficult. If
the routines will do no I/O, you can simply compile the routines to object,
and mention them in the linking command. If they may do I/O, you will need
not only to initialise the I/O system, but also to clean up afterward. This
becomes quite tricky and is best avoided whenever possible.
F77's I/O system is initialised by the C routine `f_init' (F77's support
library is written almost entirely in C) and torn down by the routine
`f_exit'. Both take no parameters. Indeed, the F77 main program consists
mainly of the three lines
f_init();
MAIN_(); /* recall that C prepends an underscore */
f_exit();
though there is much other code dealing with signals, and of course with argc
and argv.
Pascal's I/O system is initialised by the C routine `PCSTART'. (Yes, Pascal's
support library too is written in C. I find it amusing to note that other
language libraries can be written in C, but C's language libraries cannot, for
the most part, be written in the other languages.) Pascal's main can be
written in C as
extern int _argc;
extern char **_argv;
main(argc, argv)
int argc;
char **argv;
{
PCSTART(0);
_argc = argc;
_argv = argv;
program();
PCEXIT(0);
/*NOTREACHED*/
}
---though the compiler in fact generates this directly, eliminating an
unnecessary return instruction. PCEXIT, unfortunately, terminates the program
as well as flushing any pending output.
As to the various libraries themselves, there are many:
Library Used by
------- -------
-lF77 F77
-lI77 F77
-lU77 F77
-lpc Pascal
-lm F77, Pascal
-lc C, F77, Pascal
In other words, all the linking commands pass `-lc' to the linker `ld'; the
others depend on the command. `f77' calls ld with all except `-lpc'; `pc'
calls ld with `-lpc -lm -lc'. `cc' calls ld with only `-lc', so to use an F77
routine with a C main, one must link with
cc main.o f77sub.o -lF77 -lI77 -lU77 -lm
Moreover, the order of the libraries specified is also important. `-lF77'
builds on `-lI77', and `-lI77' builds on `-lU77'; all build on `-lm' and
`-lc'. `-lpc' builds on `-lm' and `-lc'. Thus `-lpc' may be put anywhere
with respect to `-lI77', for example; but both must appear before `-lm'.
You should now be able to compile and link mixed language sources. But this
is not the whole story: There is still the matter of parameter passing. The
F77 compiler uses call by reference; the Pascal compiler uses call by value or
call by reference, depending on the declaration of the called routine. The C
compiler invariably uses call by value, but the language is powerful enough to
simulate other parameter mechanisms using only call by value. One thing that
can be done in Pascal but not C is to pass arrays by value. (This can be
simulated in C using structures.)
Pretty words, those: but what do they mean? For a strict definition I will
tell you only to consult any good compiler book; but here are some examples:
[f77sub.f]
SUBROUTINE SUB (A)
INTEGER A
DOUBLE PRECISION D
C Mixed mode arithmetic is legal in Unix F77
D = A + 2.0
CALL CSUB(D)
RETURN
END
[psub.p]
{ declare external C subroutine }
procedure csub2(i: integer); external;
procedure psub(var i: integer);
begin i := 3 end;
function pfunc(i: integer): integer;
begin
pfunc := i + 2;
csub2(i)
end;
[cmain.c]
/*ARGSUSED*/
main(argc, argv)
int argc;
char **argv;
{
int i;
psub(&i); /* call Pascal subroutine with var parameter */
sub_(&i); /* call F77 subroutine: call by reference */
i = pfunc(7); /* call Pascal function with value parameter */
exit(0);
}
/* called from F77: call by reference */
csub_(d) double *d; { printf("%g\n", *d); }
/* called from Pascal by value */
csub2(i) int i; { printf("%d\n", i); }
Fortunately, function return values are all done the same way for
simple-valued functions. Structure-valued functions should simply be avoided.
Since the above example does no I/O in its F77 and Pascal routines, and in
fact calls no F77 or Pascal intrinsics, this can be compiled with the
commands:
f77 -c f77sub.f
pc -c psub.p
cc -c cmain.c
cc -o example cmain.o psub.o f77sub.o
Appending `-lF77 -lI77 -lU77 -lpc -lm' to the last command would not hurt, and
might be required in more complex cases.
There is one remaining trick in linking Pascal and C/F77 routines, and that
has to do with nested procedures and functions, and nonlocal variable access.
Neither C nor F77 have these, and there is no provision in the runtime
environment for them. Pascal, however, uses something called a `display' to
be able to get at nonlocal variables. The display manipulation is normally
compiled in-line; for procedure parameters, the compiler uses those `extra'
names. In the earlier example, these were `___proc' and `___proc_func'.
These routines do display winding for entry to _proc and _proc_func. The
unwinding after procedure parameter calls is generated in-line. If you never
use nested procedures, or nonlocal variables, you can safely ignore this. If
you do, but do not know what a display is all about, again I will tell you
only to consult a good compiler book. Look at the assembly code generated by
`pc -S' for details on the display format. Indeed, looking at the assembly
code is a good way to determine just what the compiler is really doing for all
three of these compilers.
last updated: 4/21/98
|
