[ Note:  This directory was actually brought in to be able to use curseperl,
  but it's also a useful reference for general extension ]

This directory contains an example of how you might link in C subroutines
with perl to make your own special copy of perl.  In the perl distribution
directory, there will be (after make is run) a file called uperl.o, which
is all of perl except for a single undefined subroutine, named userinit().
See usersub.c.

The sole purpose of the userinit() routine is to call the initialization
routines for any modules that you want to link in.  In this example, we just
call init_curses(), which sets up to link in the System V curses routines.
You'll find this in the file curses.c, which is the processed output of
curses.mus.  (To get BSD curses, replace curses.mus with bsdcurses.mus.)

The magicname() routine adds variable names into the symbol table.  Along
with the name of the variable as Perl knows it, we pass a structure containing
an index identifying the variable, and the names of two C functions that
know how to set or evaluate a variable given the index of the variable.
Our example uses a macro to handle this conveniently.

The init routine calls make_usub() to add user-defined subroutine names
into the symbol table.  The arguments are

	make_usub(subname, subindex, subfunc, filename);
	char *subname;
	int subindex;
	int subfunc();
	char *filename;

The subname is the name that will be used in the Perl program.  The subindex
will be passed to subfunc() when it is called to tell it which C function
is desired.  subfunc() is a glue routine that translates the arguments
from Perl internal stack form to the form required by the routine in
question, calls the desired C function, and then translates any return
value back into the stack format.  The glue routine used by curses just
has a large switch statement, each branch of which does the processing
for a particular C function.  The subindex could, however, be used to look
up a function in a dynamically linked library.  No example of this is
provided.

As a help in producing the glue routine, a preprocessor called "mus" lets
you specify argument and return value types in a tabular format.  An entry
such as:

    CASE int waddstr
    I       WINDOW*         win
    I       char*           str
    END

indicates that waddstr takes two input arguments, the first of which is a
pointer to a window, and the second of which is an ordinary C string.  It
also indicates that an integer is returned.  The mus program turns this into:

    case US_waddstr:
        if (items != 2)
            fatal("Usage: &waddstr($win, $str)");
        else {
            int retval;
            WINDOW*     win =           *(WINDOW**)     str_get(st[1]);
            char*       str =           (char*)         str_get(st[2]);

            retval = waddstr(win, str);
            str_numset(st[0], (double) retval);
        }
        return sp;

It's also possible to have output parameters, indicated by O, and input/ouput
parameters indicated by IO.

The mus program isn't perfect.  You'll note that curses.mus has some
cases which are hand coded.  They'll be passed straight through unmodified.
You can produce similar cases by analogy to what's in curses.c, as well
as similar routines in the doarg.c, dolist.c and doio.c routines of Perl.
The mus program is only intended to get you about 90% there.  It's not clear,
for instance, how a given structure should be passed to Perl.  But that
shouldn't bother you--if you've gotten this far, it's already obvious
that you are totally mad.

Here's an example of how to return an array value:

    case US_appl_errlist:
	if (!wantarray) {
	    str_numset(st[0], (double) appl_nerr);
	    return sp;
	}
	astore(stack, sp + appl_nerr, Nullstr);		/* extend stack */
	st = stack->ary_array + sp;			/* possibly realloced */
	for (i = 0; i < appl_nerr; i++) {
	    tmps = appl_errlist[i];
	    st[i] = str_2mortal(str_make(tmps,strlen(tmps)));
	}
	return sp + appl_nerr - 1;


In addition, there is a program, man2mus, that will scan a man page for
function prototypes and attempt to construct a mus CASE entry for you.  It has
to guess about input/output parameters, so you'll have to tidy up after it.
But it can save you a lot of time if the man pages for a library are
reasonably well formed.

If you happen to have curses on your machine, you might try compiling
a copy of curseperl.  The "pager" program in this directory is a rudimentary
start on writing a pager--don't believe the help message, which is stolen
from the less program.

User-defined subroutines may not currently be called as a signal handler,
though a signal handler may itself call a user-defined subroutine.

There are now glue routines to call back from C into Perl.  In usersub.c
in this directory, you'll find callback() and callv().  The callback()
routine presumes that any arguments to pass to the Perl subroutine
have already been pushed onto the Perl stack.  The callv() routine
is a wrapper that pushes an argv-style array of strings onto the
stack for you, and then calls callback().  Be sure to recheck your
stack pointer after returning from these routine, since the Perl code
may have reallocated it.