All of those strange functions with sv in their names help convert Perl scalars to C types. They're described in perlguts and perlapi.

If you compile and run string.c, you'll see the results of using SvIV() to create an int, SvNV() to create a float, and SvPV() to create a string:

   a = 9
   a = 9.859600
   a = Just Another Perl Hacker

In the example above, we've created a global variable to temporarily store the computed value of our eval'd expression. It is also possible and in most cases a better strategy to fetch the return value from eval_pv() instead. Example:

   ...
   STRLEN n_a;
   SV *val = eval_pv("reverse 'rekcaH lreP rehtonA tsuJ'", TRUE);
   printf("%s\n", SvPV(val,n_a));
   ...

This way, we avoid namespace pollution by not creating global variables and we've simplified our code as well.

Performing Perl pattern matches and substitutions from your C program

The eval_sv() function lets us evaluate strings of Perl code, so we can define some functions that use it to ``specialize'' in matches and substitutions: match(), substitute(), and matches().

   I32 match(SV *string, char *pattern);

Given a string and a pattern (e.g., m/clasp/ or /\b\w*\b/, which in your C program might appear as ``/\\b\\w*\\b/''), match() returns 1 if the string matches the pattern and 0 otherwise.

   int substitute(SV **string, char *pattern);

Given a pointer to an SV and an =~ operation (e.g., s/bob/robert/g or tr[A-Z][a-z]), substitute() modifies the string within the SV as according to the operation, returning the number of substitutions made.

   int matches(SV *string, char *pattern, AV **matches);

Given an SV, a pattern, and a pointer to an empty AV, matches() evaluates $string =~ $pattern in a list context, and fills in matches with the array elements, returning the number of matches found.

Here's a sample program, match.c, that uses all three (long lines have been wrapped here):

 #include <EXTERN.h>
 #include <perl.h>

 static PerlInterpreter *my_perl;

 /** my_eval_sv(code, error_check)
 ** kinda like eval_sv(), 
 ** but we pop the return value off the stack 
 **/
 SV* my_eval_sv(SV *sv, I32 croak_on_error)
 {
     dSP;
     SV* retval;
     STRLEN n_a;

     PUSHMARK(SP);
     eval_sv(sv, G_SCALAR);

     SPAGAIN;
     retval = POPs;
     PUTBACK;

     if (croak_on_error && SvTRUE(ERRSV))
        croak(SvPVx(ERRSV, n_a));

     return retval;
 }

 /** match(string, pattern)
 **
 ** Used for matches in a scalar context.
 **
 ** Returns 1 if the match was successful; 0 otherwise.
 **/

 I32 match(SV *string, char *pattern)
 {
     SV *command = NEWSV(1099, 0), *retval;
     STRLEN n_a;

     sv_setpvf(command, "my $string = '%s'; $string =~ %s",
              SvPV(string,n_a), pattern);

     retval = my_eval_sv(command, TRUE);
     SvREFCNT_dec(command);

     return SvIV(retval);
 }

 /** substitute(string, pattern)
 **
 ** Used for =~ operations that modify their left-hand side (s/// and tr///)
 **
 ** Returns the number of successful matches, and
 ** modifies the input string if there were any.
 **/

 I32 substitute(SV **string, char *pattern)
 {
     SV *command = NEWSV(1099, 0), *retval;
     STRLEN n_a;

     sv_setpvf(command, "$string = '%s'; ($string =~ %s)",
              SvPV(*string,n_a), pattern);

     retval = my_eval_sv(command, TRUE);
     SvREFCNT_dec(command);

     *string = get_sv("string", FALSE);
     return SvIV(retval);
 }

 /** matches(string, pattern, matches)
 **
 ** Used for matches in a list context.
 **
 ** Returns the number of matches,
 ** and fills in **matches with the matching substrings
 **/

 I32 matches(SV *string, char *pattern, AV **match_list)
 {
     SV *command = NEWSV(1099, 0);
     I32 num_matches;
     STRLEN n_a;

     sv_setpvf(command, "my $string = '%s'; @array = ($string =~ %s)",
              SvPV(string,n_a), pattern);

     my_eval_sv(command, TRUE);
     SvREFCNT_dec(command);

     *match_list = get_av("array", FALSE);
     num_matches = av_len(*match_list) + 1; /** assume $[ is 0 **/

     return num_matches;
 }

 main (int argc, char **argv, char **env)
 {
     char *embedding[] = { "", "-e", "0" };
     AV *match_list;
     I32 num_matches, i;
     SV *text;
     STRLEN n_a;

     PERL_SYS_INIT3(&argc,&argv,&env);
     my_perl = perl_alloc();
     perl_construct(my_perl);
     perl_parse(my_perl, NULL, 3, embedding, NULL);
     PL_exit_flags |= PERL_EXIT_DESTRUCT_END;

     text = NEWSV(1099,0);
     sv_setpv(text, "When he is at a convenience store and the "
        "bill comes to some amount like 76 cents, Maynard is "
        "aware that there is something he *should* do, something "
        "that will enable him to get back a quarter, but he has "
        "no idea *what*.  He fumbles through his red squeezey "
        "changepurse and gives the boy three extra pennies with "
        "his dollar, hoping that he might luck into the correct "
        "amount.  The boy gives him back two of his own pennies "
        "and then the big shiny quarter that is his prize. "
        "-RICHH");

     if (match(text, "m/quarter/")) /** Does text contain 'quarter'? **/
        printf("match: Text contains the word 'quarter'.\n\n");
     else
        printf("match: Text doesn't contain the word 'quarter'.\n\n");

     if (match(text, "m/eighth/")) /** Does text contain 'eighth'? **/
        printf("match: Text contains the word 'eighth'.\n\n");
     else
        printf("match: Text doesn't contain the word 'eighth'.\n\n");

     /** Match all occurrences of /wi../ **/
     num_matches = matches(text, "m/(wi..)/g", &match_list);
     printf("matches: m/(wi..)/g found %d matches...\n", num_matches);

     for (i = 0; i < num_matches; i++)
        printf("match: %s\n", SvPV(*av_fetch(match_list, i, FALSE),n_a));
     printf("\n");

     /** Remove all vowels from text **/
     num_matches = substitute(&text, "s/[aeiou]//gi");
     if (num_matches) {
        printf("substitute: s/[aeiou]//gi...%d substitutions made.\n",
               num_matches);
        printf("Now text is: %s\n\n", SvPV(text,n_a));
     }

     /** Attempt a substitution **/
     if (!substitute(&text, "s/Perl/C/")) {
        printf("substitute: s/Perl/C...No substitution made.\n\n");
     }

     SvREFCNT_dec(text);
     PL_perl_destruct_level = 1;
     perl_destruct(my_perl);
     perl_free(my_perl);
     PERL_SYS_TERM();
 }

which produces the output (again, long lines have been wrapped here)

   match: Text contains the word 'quarter'.

   match: Text doesn't contain the word 'eighth'.

   matches: m/(wi..)/g found 2 matches...
   match: will
   match: with

   substitute: s/[aeiou]//gi...139 substitutions made.
   Now text is: Whn h s t  cnvnnc str nd th bll cms t sm mnt lk 76 cnts,
   Mynrd s wr tht thr s smthng h *shld* d, smthng tht wll nbl hm t gt bck
   qrtr, bt h hs n d *wht*.  H fmbls thrgh hs rd sqzy chngprs nd gvs th by
   thr xtr pnns wth hs dllr, hpng tht h mght lck nt th crrct mnt.  Th by gvs
   hm bck tw f hs wn pnns nd thn th bg shny qrtr tht s hs prz. -RCHH

   substitute: s/Perl/C...No substitution made.

Fiddling with the Perl stack from your C program

When trying to explain stacks, most computer science textbooks mumble something about spring-loaded columns of cafeteria plates: the last thing you pushed on the stack is the first thing you pop off. That'll do for our purposes: your C program will push some arguments onto ``the Perl stack'', shut its eyes while some magic happens, and then pop the results--the return value of your Perl subroutine--off the stack.

perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 168. perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 168.

First you'll need to know how to convert between C types and Perl types, with newSViv() and sv_setnv() and newAV() and all their friends. They're described in perlguts and perlapi.

perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 169.

Then you'll need to know how to manipulate the Perl stack. That's described in perlcall.

Once you've understood those, embedding Perl in C is easy.

Because C has no builtin function for integer exponentiation, let's make Perl's ** operator available to it (this is less useful than it sounds, because Perl implements ** with C's pow() function). First I'll create a stub exponentiation function in power.pl:

    sub expo {
        my ($a, $b) = @_;
        return $a ** $b;
    }

Now I'll create a C program, power.c, with a function PerlPower() that contains all the perlguts necessary to push the two arguments into expo() and to pop the return value out. Take a deep breath...

    #include <EXTERN.h>
    #include <perl.h>

    static PerlInterpreter *my_perl;

    static void
    PerlPower(int a, int b)
    {
      dSP;                            /* initialize stack pointer      */
      ENTER;                          /* everything created after here */
      SAVETMPS;                       /* ...is a temporary variable.   */
      PUSHMARK(SP);                   /* remember the stack pointer    */
      XPUSHs(sv_2mortal(newSViv(a))); /* push the base onto the stack  */
      XPUSHs(sv_2mortal(newSViv(b))); /* push the exponent onto stack  */
      PUTBACK;                      /* make local stack pointer global */
      call_pv("expo", G_SCALAR);      /* call the function             */
      SPAGAIN;                        /* refresh stack pointer         */
                                    /* pop the return value from stack */
      printf ("%d to the %dth power is %d.\n", a, b, POPi);
      PUTBACK;
      FREETMPS;                       /* free that return value        */
      LEAVE;                       /* ...and the XPUSHed "mortal" args.*/
    }

    int main (int argc, char **argv, char **env)
    {
      char *my_argv[] = { "", "power.pl" };

      PERL_SYS_INIT3(&argc,&argv,&env);
      my_perl = perl_alloc();
      perl_construct( my_perl );

      perl_parse(my_perl, NULL, 2, my_argv, (char **)NULL);
      PL_exit_flags |= PERL_EXIT_DESTRUCT_END;
      perl_run(my_perl);

      PerlPower(3, 4);                      /*** Compute 3 ** 4 ***/

      perl_destruct(my_perl);
      perl_free(my_perl);
      PERL_SYS_TERM();
    }

Compile and run:

    % cc -o power power.c `perl -MExtUtils::Embed -e ccopts -e ldopts`

    % power
    3 to the 4th power is 81.

Maintaining a persistent interpreter

When developing interactive and/or potentially long-running applications, it's a good idea to maintain a persistent interpreter rather than allocating and constructing a new interpreter multiple times. The major reason is speed: since Perl will only be loaded into memory once.

However, you have to be more cautious with namespace and variable scoping when using a persistent interpreter. In previous examples we've been using global variables in the default package main. We knew exactly what code would be run, and assumed we could avoid variable collisions and outrageous symbol table growth.

Let's say your application is a server that will occasionally run Perl code from some arbitrary file. Your server has no way of knowing what code it's going to run. Very dangerous.

If the file is pulled in by perl_parse(), compiled into a newly constructed interpreter, and subsequently cleaned out with perl_destruct() afterwards, you're shielded from most namespace troubles.

perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 190. perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 190.

One way to avoid namespace collisions in this scenario is to translate the filename into a guaranteed-unique package name, and then compile the code into that package using perlfunc/eval. In the example below, each file will only be compiled once. Or, the application might choose to clean out the symbol table associated with the file after it's no longer needed. Using perlapi/call_argv, We'll call the subroutine Embed::Persistent::eval_file which lives in the file persistent.pl and pass the filename and boolean cleanup/cache flag as arguments.

perldoc2tree.cgi: /usr/lib/perl5/5.8.8/pod/perlembed.pod: cannot resolve L in paragraph 191.

Note that the process will continue to grow for each file that it uses. In addition, there might be AUTOLOADed subroutines and other conditions that cause Perl's symbol table to grow. You might want to add some logic that keeps track of the process size, or restarts itself after a certain number of requests, to ensure that memory consumption is minimized. You'll also want to scope your variables with perlfunc/my whenever possible.

 package Embed::Persistent;
 #persistent.pl

 use strict;
 our %Cache;
 use Symbol qw(delete_package);

 sub valid_package_name {
     my($string) = @_;
     $string =~ s/([^A-Za-z0-9\/])/sprintf("_%2x",unpack("C",$1))/eg;
     # second pass only for words starting with a digit
     $string =~ s|/(\d)|sprintf("/_%2x",unpack("C",$1))|eg;

     # Dress it up as a real package name
     $string =~ s|/|::|g;
     return "Embed" . $string;
 }

 sub eval_file {
     my($filename, $delete) = @_;
     my $package = valid_package_name($filename);
     my $mtime = -M $filename;
     if(defined $Cache{$package}{mtime}
        &&
        $Cache{$package}{mtime} <= $mtime)
     {
        # we have compiled this subroutine already,
        # it has not been updated on disk, nothing left to do
        print STDERR "already compiled $package->handler\n";
     }
     else {
        local *FH;
        open FH, $filename or die "open '$filename' $!";
        local($/) = undef;
        my $sub = <FH>;
        close FH;

        #wrap the code into a subroutine inside our unique package
        my $eval = qq{package $package; sub handler { $sub; }};
        {
            # hide our variables within this block
            my($filename,$mtime,$package,$sub);
            eval $eval;
        }
        die $@ if $@;

        #cache it unless we're cleaning out each time
        $Cache{$package}{mtime} = $mtime unless $delete;
     }

     eval {$package->handler;};
     die $@ if $@;

     delete_package($package) if $delete;

     #take a look if you want
     #print Devel::Symdump->rnew($package)->as_string, $/;
 }

 1;

 __END__

 /* persistent.c */
 #include <EXTERN.h>
 #include <perl.h>

 /* 1 = clean out filename's symbol table after each request, 0 = don't */
 #ifndef DO_CLEAN
 #define DO_CLEAN 0
 #endif

 #define BUFFER_SIZE 1024

 static PerlInterpreter *my_perl = NULL;

 int
 main(int argc, char **argv, char **env)
 {
     char *embedding[] = { "", "persistent.pl" };
     char *args[] = { "", DO_CLEAN, NULL };
     char filename[BUFFER_SIZE];
     int exitstatus = 0;
     STRLEN n_a;

     PERL_SYS_INIT3(&argc,&argv,&env);
     if((my_perl = perl_alloc()) == NULL) {
        fprintf(stderr, "no memory!");
        exit(1);
     }
     perl_construct(my_perl);

     exitstatus = perl_parse(my_perl, NULL, 2, embedding, NULL);
     PL_exit_flags |= PERL_EXIT_DESTRUCT_END;
     if(!exitstatus) {
        exitstatus = perl_run(my_perl);

        while(printf("Enter file name: ") &&
              fgets(filename, BUFFER_SIZE, stdin)) {

            filename[strlen(filename)-1] = '\0'; /* strip \n */
            /* call the subroutine, passing it the filename as an argument */
            args[0] = filename;
            call_argv("Embed::Persistent::eval_file",
                           G_DISCARD | G_EVAL, args);

            /* check $@ */
            if(SvTRUE(ERRSV))
                fprintf(stderr, "eval error: %s\n", SvPV(ERRSV,n_a));
        }
     }

     PL_perl_destruct_level = 0;
     perl_destruct(my_perl);
     perl_free(my_perl);
     PERL_SYS_TERM();
     exit(exitstatus);
 }

Now compile:

 % cc -o persistent persistent.c `perl -MExtUtils::Embed -e ccopts -e ldopts`

Here's an example script file:

 #test.pl
 my $string = "hello";
 foo($string);

 sub foo {
     print "foo says: @_\n";
 }

Now run:

 % persistent
 Enter file name: test.pl
 foo says: hello
 Enter file name: test.pl
 already compiled Embed::test_2epl->handler
 foo says: hello
 Enter file name: ^C

Execution of END blocks

Traditionally END blocks have been executed at the end of the perl_run. This causes problems for applications that never call perl_run. Since perl 5.7.2 you can specify PL_exit_flags |= PERL_EXIT_DESTRUCT_END to get the new behaviour. This also enables the running of END blocks if the perl_parse fails and perl_destruct will return the exit value.

Maintaining multiple interpreter instances

Some rare applications will need to create more than one interpreter during a session. Such an application might sporadically decide to release any resources associated with the interpreter.

The program must take care to ensure that this takes place before the next interpreter is constructed. By default, when perl is not built with any special options, the global variable PL_perl_destruct_level is set to 0, since extra cleaning isn't usually needed when a program only ever creates a single interpreter in its entire lifetime.

Setting PL_perl_destruct_level to 1 makes everything squeaky clean:

 while(1) {
     ...
     /* reset global variables here with PL_perl_destruct_level = 1 */
     PL_perl_destruct_level = 1;
     perl_construct(my_perl);
     ...
     /* clean and reset _everything_ during perl_destruct */
     PL_perl_destruct_level = 1;
     perl_destruct(my_perl);
     perl_free(my_perl);
     ...
     /* let's go do it again! */
 }

When perl_destruct() is called, the interpreter's syntax parse tree and symbol tables are cleaned up, and global variables are reset. The second assignment to PL_perl_destruct_level is needed because perl_construct resets it to 0.

Now suppose we have more than one interpreter instance running at the same time. This is feasible, but only if you used the Configure option -Dusemultiplicity or the options -Dusethreads -Duseithreads when building perl. By default, enabling one of these Configure options sets the per-interpreter global variable PL_perl_destruct_level to 1, so that thorough cleaning is automatic and interpreter variables are initialized correctly. Even if you don't intend to run two or more interpreters at the same time, but to run them sequentially, like in the above example, it is recommended to build perl with the -Dusemultiplicity option otherwise some interpreter variables may not be initialized correctly between consecutive runs and your application may crash.

Using -Dusethreads -Duseithreads rather than -Dusemultiplicity is more appropriate if you intend to run multiple interpreters concurrently in different threads, because it enables support for linking in the thread libraries of your system with the interpreter.

Let's give it a try:

 #include <EXTERN.h>
 #include <perl.h>

 /* we're going to embed two interpreters */
 /* we're going to embed two interpreters */

 #define SAY_HELLO "-e", "print qq(Hi, I'm $^X\n)"

 int main(int argc, char **argv, char **env)
 {
     PerlInterpreter *one_perl, *two_perl;
     char *one_args[] = { "one_perl", SAY_HELLO };
     char *two_args[] = { "two_perl", SAY_HELLO };

     PERL_SYS_INIT3(&argc,&argv,&env);
     one_perl = perl_alloc();
     two_perl = perl_alloc();

     PERL_SET_CONTEXT(one_perl);
     perl_construct(one_perl);
     PERL_SET_CONTEXT(two_perl);
     perl_construct(two_perl);

     PERL_SET_CONTEXT(one_perl);
     perl_parse(one_perl, NULL, 3, one_args, (char **)NULL);
     PERL_SET_CONTEXT(two_perl);
     perl_parse(two_perl, NULL, 3, two_args, (char **)NULL);

     PERL_SET_CONTEXT(one_perl);
     perl_run(one_perl);
     PERL_SET_CONTEXT(two_perl);
     perl_run(two_perl);

     PERL_SET_CONTEXT(one_perl);
     perl_destruct(one_perl);
     PERL_SET_CONTEXT(two_perl);
     perl_destruct(two_perl);

     PERL_SET_CONTEXT(one_perl);
     perl_free(one_perl);
     PERL_SET_CONTEXT(two_perl);
     perl_free(two_perl);
     PERL_SYS_TERM();
 }

Note the calls to PERL_SET_CONTEXT(). These are necessary to initialize the global state that tracks which interpreter is the ``current'' one on the particular process or thread that may be running it. It should always be used if you have more than one interpreter and are making perl API calls on both interpreters in an interleaved fashion.

Compile as usual:

 % cc -o multiplicity multiplicity.c `perl -MExtUtils::Embed -e ccopts -e ldopts`

Run it, Run it:

 % multiplicity
 Hi, I'm one_perl
 Hi, I'm two_perl

Using Perl modules, which themselves use C libraries, from your C program

If you've played with the examples above and tried to embed a script that use()s a Perl module (such as Socket) which itself uses a C or C++ library, this probably happened:

 Can't load module Socket, dynamic loading not available in this perl.
  (You may need to build a new perl executable which either supports
  dynamic loading or has the Socket module statically linked into it.)

What's wrong?

Your interpreter doesn't know how to communicate with these extensions on its own. A little glue will help. Up until now you've been calling perl_parse(), handing it NULL for the second argument:

 perl_parse(my_perl, NULL, argc, my_argv, NULL);

That's where the glue code can be inserted to create the initial contact between Perl and linked C/C++ routines. Let's take a look some pieces of perlmain.c to see how Perl does this:

 static void xs_init (pTHX);

 EXTERN_C void boot_DynaLoader (pTHX_ CV* cv);
 EXTERN_C void boot_Socket (pTHX_ CV* cv);

 EXTERN_C void
 xs_init(pTHX)
 {
        char *file = __FILE__;
        /* DynaLoader is a special case */
        newXS("DynaLoader::boot_DynaLoader", boot_DynaLoader, file);
        newXS("Socket::bootstrap", boot_Socket, file);
 }

Simply put: for each extension linked with your Perl executable (determined during its initial configuration on your computer or when adding a new extension), a Perl subroutine is created to incorporate the extension's routines. Normally, that subroutine is named Module::bootstrap() and is invoked when you say use Module. In turn, this hooks into an XSUB, boot_Module, which creates a Perl counterpart for each of the extension's XSUBs. Don't worry about this part; leave that to the xsubpp and extension authors. If your extension is dynamically loaded, DynaLoader creates Module::bootstrap() for you on the fly. In fact, if you have a working DynaLoader then there is rarely any need to link in any other extensions statically.

Once you have this code, slap it into the second argument of perl_parse():

 perl_parse(my_perl, xs_init, argc, my_argv, NULL);

Then compile:

 % cc -o interp interp.c `perl -MExtUtils::Embed -e ccopts -e ldopts`

 % interp
   use Socket;
   use SomeDynamicallyLoadedModule;

   print "Now I can use extensions!\n"'

ExtUtils::Embed can also automate writing the xs_init glue code.

 % perl -MExtUtils::Embed -e xsinit -- -o perlxsi.c
 % cc -c perlxsi.c `perl -MExtUtils::Embed -e ccopts`
 % cc -c interp.c  `perl -MExtUtils::Embed -e ccopts`
 % cc -o interp perlxsi.o interp.o `perl -MExtUtils::Embed -e ldopts`

 C:\ExtUtils-Embed\eg> perl genmake interp.c
 C:\ExtUtils-Embed\eg> nmake
 C:\ExtUtils-Embed\eg> interp -e "print qq{I'm embedded in Win32!\n}"

 perl -MExtUtils::Embed -e xsinit

    warn("%d bottles of beer on the wall", bottlecount);

    Perl_warn(aTHX_ "%d bottles of beer on the wall", bottlecount);