FreeBSD manual
download PDF document: perlembed.1.pdf
PERLEMBED(1) Perl Programmers Reference Guide PERLEMBED(1)
NAME
perlembed - how to embed perl in your C program
DESCRIPTION
PREAMBLE
Do you want to:
Use C from Perl?
Read perlxstut, perlxs, h2xs, perlguts, and perlapi.
Use a Unix program from Perl?
Read about back-quotes and about "system" and "exec" in perlfunc.
Use Perl from Perl?
Read about "do" in perlfunc and "eval" in perlfunc and "require"
in perlfunc and "use" in perlfunc.
Use C from C?
Rethink your design.
Use Perl from C?
Read on...
ROADMAP
o Compiling your C program
o Adding a Perl interpreter to your C program
o Calling a Perl subroutine from your C program
o Evaluating a Perl statement from your C program
o Performing Perl pattern matches and substitutions from your C
program
o Fiddling with the Perl stack from your C program
o Maintaining a persistent interpreter
o Maintaining multiple interpreter instances
o Using Perl modules, which themselves use C libraries, from your C
program
o Embedding Perl under Win32
Compiling your C program
If you have trouble compiling the scripts in this documentation, you're
not alone. The cardinal rule: COMPILE THE PROGRAMS IN EXACTLY THE SAME
WAY THAT YOUR PERL WAS COMPILED. (Sorry for yelling.)
Also, every C program that uses Perl must link in the perl library.
What's that, you ask? Perl is itself written in C; the perl library is
the collection of compiled C programs that were used to create your
perl executable (/usr/bin/perl or equivalent). (Corollary: you can't
use Perl from your C program unless Perl has been compiled on your
If your copy of Perl is recent enough to contain this documentation
(version 5.002 or later), then the perl library (and EXTERN.h and
perl.h, which you'll also need) will reside in a directory that looks
like this:
/usr/local/lib/perl5/your_architecture_here/CORE
or perhaps just
/usr/local/lib/perl5/CORE
or maybe something like
/usr/opt/perl5/CORE
Execute this statement for a hint about where to find CORE:
perl -MConfig -e 'print $Config{archlib}'
Here's how you'd compile the example in the next section, "Adding a
Perl interpreter to your C program", on my Linux box:
% gcc -O2 -Dbool=char -DHAS_BOOL -I/usr/local/include
-I/usr/local/lib/perl5/i586-linux/5.003/CORE
-L/usr/local/lib/perl5/i586-linux/5.003/CORE
-o interp interp.c -lperl -lm
(That's all one line.) On my DEC Alpha running old 5.003_05, the
incantation is a bit different:
% cc -O2 -Olimit 2900 -I/usr/local/include
-I/usr/local/lib/perl5/alpha-dec_osf/5.00305/CORE
-L/usr/local/lib/perl5/alpha-dec_osf/5.00305/CORE -L/usr/local/lib
-D__LANGUAGE_C__ -D_NO_PROTO -o interp interp.c -lperl -lm
How can you figure out what to add? Assuming your Perl is post-5.001,
execute a "perl -V" command and pay special attention to the "cc" and
"ccflags" information.
You'll have to choose the appropriate compiler (cc, gcc, et al.) for
your machine: "perl -MConfig -e 'print $Config{cc}'" will tell you what
to use.
You'll also have to choose the appropriate library directory
(/usr/local/lib/...) for your machine. If your compiler complains that
certain functions are undefined, or that it can't locate -lperl, then
you need to change the path following the "-L". If it complains that
it can't find EXTERN.h and perl.h, you need to change the path
following the "-I".
You may have to add extra libraries as well. Which ones? Perhaps
those printed by
perl -MConfig -e 'print $Config{libs}'
Provided your perl binary was properly configured and installed the
ExtUtils::Embed module will determine all of this information for you:
The ExtUtils::Embed kit on CPAN also contains all source code for the
examples in this document, tests, additional examples and other
information you may find useful.
Adding a Perl interpreter to your C program
In a sense, perl (the C program) is a good example of embedding Perl
(the language), so I'll demonstrate embedding with miniperlmain.c,
included in the source distribution. Here's a bastardized, non-
portable version of miniperlmain.c containing the essentials of
embedding:
#include <EXTERN.h> /* from the Perl distribution */
#include <perl.h> /* from the Perl distribution */
static PerlInterpreter *my_perl; /*** The Perl interpreter ***/
int main(int argc, char **argv, char **env)
{
PERL_SYS_INIT3(&argc,&argv,&env);
my_perl = perl_alloc();
perl_construct(my_perl);
PL_exit_flags |= PERL_EXIT_DESTRUCT_END;
perl_parse(my_perl, NULL, argc, argv, (char **)NULL);
perl_run(my_perl);
perl_destruct(my_perl);
perl_free(my_perl);
PERL_SYS_TERM();
exit(EXIT_SUCCESS);
}
Notice that we don't use the "env" pointer. Normally handed to
"perl_parse" as its final argument, "env" here is replaced by "NULL",
which means that the current environment will be used.
The macros PERL_SYS_INIT3() and PERL_SYS_TERM() provide system-specific
tune up of the C runtime environment necessary to run Perl
interpreters; they should only be called once regardless of how many
interpreters you create or destroy. Call PERL_SYS_INIT3() before you
create your first interpreter, and PERL_SYS_TERM() after you free your
last interpreter.
Since PERL_SYS_INIT3() may change "env", it may be more appropriate to
provide "env" as an argument to perl_parse().
Also notice that no matter what arguments you pass to perl_parse(),
PERL_SYS_INIT3() must be invoked on the C main() argc, argv and env and
only once.
Mind that argv[argc] must be NULL, same as those passed to a main
function in C.
Now compile this program (I'll call it interp.c) into an executable:
% cc -o interp interp.c `perl -MExtUtils::Embed -e ccopts -e ldopts`
After a successful compilation, you'll be able to use interp just like
perl itself:
% interp
% interp -e 'printf("%x", 3735928559)'
deadbeef
You can also read and execute Perl statements from a file while in the
midst of your C program, by placing the filename in argv[1] before
calling perl_run.
Calling a Perl subroutine from your C program
To call individual Perl subroutines, you can use any of the call_*
functions documented in perlcall. In this example we'll use
"call_argv".
That's shown below, in a program I'll call showtime.c.
#include <EXTERN.h>
#include <perl.h>
static PerlInterpreter *my_perl;
int main(int argc, char **argv, char **env)
{
char *args[] = { NULL };
PERL_SYS_INIT3(&argc,&argv,&env);
my_perl = perl_alloc();
perl_construct(my_perl);
perl_parse(my_perl, NULL, argc, argv, NULL);
PL_exit_flags |= PERL_EXIT_DESTRUCT_END;
/*** skipping perl_run() ***/
call_argv("showtime", G_DISCARD | G_NOARGS, args);
perl_destruct(my_perl);
perl_free(my_perl);
PERL_SYS_TERM();
exit(EXIT_SUCCESS);
}
where showtime is a Perl subroutine that takes no arguments (that's the
G_NOARGS) and for which I'll ignore the return value (that's the
G_DISCARD). Those flags, and others, are discussed in perlcall.
I'll define the showtime subroutine in a file called showtime.pl:
print "I shan't be printed.";
sub showtime {
print time;
}
Simple enough. Now compile and run:
% cc -o showtime showtime.c \
`perl -MExtUtils::Embed -e ccopts -e ldopts`
% showtime showtime.pl
818284590
If you want to pass arguments to the Perl subroutine, you can add
strings to the "NULL"-terminated "args" list passed to call_argv. For
other data types, or to examine return values, you'll need to
manipulate the Perl stack. That's demonstrated in "Fiddling with the
Perl stack from your C program".
Evaluating a Perl statement from your C program
Perl provides two API functions to evaluate pieces of Perl code. These
are "eval_sv" in perlapi and "eval_pv" in perlapi.
Arguably, these are the only routines you'll ever need to execute
snippets of Perl code from within your C program. Your code can be as
long as you wish; it can contain multiple statements; it can employ
"use" in perlfunc, "require" in perlfunc, and "do" in perlfunc to
include external Perl files.
eval_pv lets us evaluate individual Perl strings, and then extract
variables for coercion into C types. The following program, string.c,
executes three Perl strings, extracting an "int" from the first, a
"float" from the second, and a "char *" from the third.
#include <EXTERN.h>
#include <perl.h>
static PerlInterpreter *my_perl;
main (int argc, char **argv, char **env)
{
char *embedding[] = { "", "-e", "0", NULL };
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;
perl_run(my_perl);
/** Treat $a as an integer **/
eval_pv("$a = 3; $a **= 2", TRUE);
printf("a = %d\n", SvIV(get_sv("a", 0)));
/** Treat $a as a float **/
eval_pv("$a = 3.14; $a **= 2", TRUE);
printf("a = %f\n", SvNV(get_sv("a", 0)));
/** Treat $a as a string **/
eval_pv(
"$a = 'rekcaH lreP rehtonA tsuJ'; $a = reverse($a);", TRUE);
printf("a = %s\n", SvPV_nolen(get_sv("a", 0)));
perl_destruct(my_perl);
perl_free(my_perl);
PERL_SYS_TERM();
}
All of those strange functions with sv in their names help convert Perl
scalars to C types. They're described in perlguts and perlapi.
a = Just Another Perl Hacker
In the example above, we've created a global variable to temporarily
store the computed value of our eval'ed expression. It is also
possible and in most cases a better strategy to fetch the return value
from eevvaall_ppvv(()) instead. Example:
...
SV *val = eval_pv("reverse 'rekcaH lreP rehtonA tsuJ'", TRUE);
printf("%s\n", SvPV_nolen(val));
...
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 eevvaall_ssvv(()) function lets us evaluate strings of Perl code, so we can
define some functions that use it to "specialize" in matches and
substitutions: mmaattcchh(()), ssuubbssttiittuuttee(()), and mmaattcchheess(()).
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.
SSize_t 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;
PUSHMARK(SP);
eval_sv(sv, G_SCALAR);
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(0), *retval;
sv_setpvf(command, "my $string = '%s'; $string =~ %s",
SvPV_nolen(string), pattern);
retval = my_eval_sv(command, TRUE);
SvREFCNT_dec(command);
return SvIV(retval);
}
/** substitute(string, pattern)
**
** Used for =~ operations that
** modify their left-hand side (s/// and tr///)
**
** Returns the number of successful matches, and
** modifies the input string if there were any.
**/
I32 substitute(SV **string, char *pattern)
{
SV *command = newSV(0), *retval;
sv_setpvf(command, "$string = '%s'; ($string =~ %s)",
SvPV_nolen(*string), pattern);
retval = my_eval_sv(command, TRUE);
SvREFCNT_dec(command);
*string = get_sv("string", 0);
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
**/
SSize_t matches(SV *string, char *pattern, AV **match_list)
{
SV *command = newSV(0);
SSize_t num_matches;
*match_list = get_av("array", 0);
num_matches = av_top_index(*match_list) + 1;
return num_matches;
}
main (int argc, char **argv, char **env)
{
char *embedding[] = { "", "-e", "0", NULL };
AV *match_list;
I32 num_matches, i;
SV *text;
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(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_nolen(*av_fetch(match_list, i, FALSE)));
printf("\n");
/** Remove all vowels from text **/
num_matches = substitute(&text, "s/[aeiou]//gi");
if (num_matches) {
printf("substitute: s/[aeiou]//gi...%lu substitutions made.\n",
(unsigned long)num_matches);
printf("Now text is: %s\n\n", SvPV_nolen(text));
}
/** Attempt a substitution **/
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.
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.
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 ppooww(()) 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 PPeerrllPPoowweerr(()) that
contains all the perlguts necessary to push the two arguments into
eexxppoo(()) and to pop the return value out. Take a deep breath...
#include <EXTERN.h>
#include <perl.h>
static PerlInterpreter *my_perl;
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", NULL };
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();
exit(EXIT_SUCCESS);
}
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
choose to clean out the symbol table associated with the file after
it's no longer needed. Using "call_argv" in perlapi, 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.
Note that the process will continue to grow for each file that it uses.
In addition, there might be "AUTOLOAD"ed 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 "my" in
perlfunc 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 $@;
delete_package($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", NULL };
char *args[] = { "", DO_CLEAN, NULL };
char filename[BUFFER_SIZE];
int failing, exitstatus;
PERL_SYS_INIT3(&argc,&argv,&env);
if((my_perl = perl_alloc()) == NULL) {
fprintf(stderr, "no memory!");
exit(EXIT_FAILURE);
}
perl_construct(my_perl);
PL_origalen = 1; /* don't let $0 assignment update the
proctitle or embedding[0] */
failing = perl_parse(my_perl, NULL, 2, embedding, NULL);
PL_exit_flags |= PERL_EXIT_DESTRUCT_END;
if(!failing)
failing = perl_run(my_perl);
if(!failing) {
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 $@ */
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.
$0 assignments
When a perl script assigns a value to $0 then the perl runtime will try
to make this value show up as the program name reported by "ps" by
updating the memory pointed to by the argv passed to perl_parse() and
also calling API functions like setproctitle() where available. This
behaviour might not be appropriate when embedding perl and can be
disabled by assigning the value 1 to the variable "PL_origalen" before
perl_parse() is called.
The persistent.c example above is for instance likely to segfault when
$0 is assigned to if the "PL_origalen = 1;" assignment is removed.
This because perl will try to write to the read only memory of the
"embedding[]" strings.
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"
/* 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 ppeerrll_ddeessttrruucctt(()) 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.
See also "Thread-aware system interfaces" in perlxs.
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 */
#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, NULL };
char *two_args[] = { "two_perl", SAY_HELLO, NULL };
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_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();
exit(EXIT_SUCCESS);
}
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.
PERL_SET_CONTEXT(interp) should also be called whenever "interp" is
used by a thread that did not create it (using either perl_alloc(), or
the more esoteric perl_clone()).
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 uussee(())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
ppeerrll_ppaarrssee(()), 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
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
MMoodduullee::::bboooottssttrraapp(()) 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 MMoodduullee::::bboooottssttrraapp(()) 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
ppeerrll_ppaarrssee(()):
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`
Consult perlxs, perlguts, and perlapi for more details.
Using embedded Perl with POSIX locales
(See perllocale for information about these.) When a Perl interpreter
normally starts up, it tells the system it wants to use the system's
default locale. This is often, but not necessarily, the "C" or "POSIX"
locale. Absent a "use locale" within the perl code, this mostly has no
effect (but see "Not within the scope of "use locale"" in perllocale).
Also, there is not a problem if the locale you want to use in your
embedded perl is the same as the system default. However, this doesn't
work if you have set up and want to use a locale that isn't the system
default one. Starting in Perl v5.20, you can tell the embedded Perl
interpreter that the locale is already properly set up, and to skip
doing its own normal initialization. It skips if the environment
variable "PERL_SKIP_LOCALE_INIT" is set (even if set to 0 or ""). A
properly before starting the Perl interpreter. It will then properly
switch back to using the thread-safe functions.
Hiding Perl_
If you completely hide the short forms of the Perl public API, add
-DPERL_NO_SHORT_NAMES to the compilation flags. This means that for
example instead of writing
warn("%d bottles of beer on the wall", bottlecount);
you will have to write the explicit full form
Perl_warn(aTHX_ "%d bottles of beer on the wall", bottlecount);
(See "Background and PERL_IMPLICIT_CONTEXT" in perlguts for the
explanation of the "aTHX_". ) Hiding the short forms is very useful
for avoiding all sorts of nasty (C preprocessor or otherwise) conflicts
with other software packages (Perl defines about 2400 APIs with these
short names, take or leave few hundred, so there certainly is room for
conflict.)
MORAL
You can sometimes write faster code in C, but you can always write code
faster in Perl. Because you can use each from the other, combine them
as you wish.
AUTHOR
Jon Orwant <orwant@media.mit.edu> and Doug MacEachern
<dougm@covalent.net>, with small contributions from Tim Bunce, Tom
Christiansen, Guy Decoux, Hallvard Furuseth, Dov Grobgeld, and Ilya
Zakharevich.
Doug MacEachern has an article on embedding in Volume 1, Issue 4 of The
Perl Journal ( <http://www.tpj.com/> ). Doug is also the developer of
the most widely-used Perl embedding: the mod_perl system
(perl.apache.org), which embeds Perl in the Apache web server. Oracle,
Binary Evolution, ActiveState, and Ben Sugars's nsapi_perl have used
this model for Oracle, Netscape and Internet Information Server Perl
plugins.
COPYRIGHT
Copyright (C) 1995, 1996, 1997, 1998 Doug MacEachern and Jon Orwant.
All Rights Reserved.
This document may be distributed under the same terms as Perl itself.
perl v5.34.3 2023-11-28 PERLEMBED(1)