NAME
FFI::Platypus::Bundle - Bundle foreign code with your Perl module
VERSION
version 2.10
SYNOPSIS
ffi/foo.c
:
#include <ffi_platypus_bundle.h>
#include <string.h>
typedef struct {
char *name;
int value;
} foo_t;
foo_t*
foo__new(const char *class_name, const char *name, int value) {
(void)class_name;
foo_t *self = malloc( sizeof( foo_t ) );
self->name = strdup(name);
self->value = value;
return self;
}
const char *
foo__name(foo_t *self) {
return self->name;
}
int
foo__value(foo_t *self) {
return self->value;
}
void
foo__DESTROY(foo_t *self) {
free(self->name);
free(self);
}
lib/Foo.pm
:
package Foo;
use strict;
use warnings;
use FFI::Platypus 2.00;
my $ffi = FFI::Platypus->new( api => 2 );
$ffi->type('object(Foo)' => 'foo_t');
$ffi->mangler(sub {
my $name = shift;
$name =~ s/^/foo__/;
$name;
});
$ffi->bundle;
$ffi->attach( new => [ 'string', 'string', 'int' ] => 'foo_t' );
$ffi->attach( name => [ 'foo_t' ] => 'string' );
$ffi->attach( value => [ 'foo_t' ] => 'int' );
$ffi->attach( DESTROY => [ 'foo_t' ] => 'void' );
1;
t/foo.t
use Test2::V0;
use Foo;
my $foo = Foo->new("platypus", 10);
isa_ok $foo, 'Foo';
is $foo->name, "platypus";
is $foo->value, 10;
done_testing;
Makefile.PL
:
use ExtUtils::MakeMaker;
use FFI::Build::MM;
my $fbmm = FFI::Build::MM->new;
WriteMakefile(
$fbmm->mm_args(
NAME => 'Foo',
DISTNAME => 'Foo',
VERSION => '1.00',
# ...
)
);
sub MY::postamble
{
$fbmm->mm_postamble;
}
or dist.ini
:
name = Foo
version = 0.01
...
[FFI::Build]
version = 1.04
DESCRIPTION
This document serves as a tutorial for using the new bundling interface provided by FFI::Platypus as of api version 1. It requires FFI::Platypus of at least 1.00.
Sometimes when writing FFI bindings you need to include a little C code (or your favorite compiled language) to finish things off. Alternatively, you might just want to write some C code (or your favorite compiled language) to include with your Perl module to make a tight loop faster. The bundling interface has you covered.
Basic example
To illustrate we will go through the files in the synopsis and explain how and why they work. To start with we have some C code which emulates object oriented code using foo__
as a prefix. We use a C struct that we call foo_t
to store our object data. On the C level the struct acts as a class, when combined with its functions that act as methods. The constructor just allocates the memory it needs for the foo_t
instance, fills in the appropriate fields and returns the pointer:
foo_t*
foo__new(const char *class_name, const char *name, int value)
{
(void) class_name;
foo_t *self = malloc( sizeof( foo_t ) );
self->name = strdup(name);
self->value = value;
return self;
}
We include a class name as the first argument, because Perl will include that when calling the constructor, but we do not use it here. An exercise for the reader would be to add hierarchical inheritance.
There are also some methods which return member values. This class has only read only members, but you could have read/write or other methods depending on your needs.
const char *
foo__name(foo_t *self)
{
return self->name;
}
We also include a destructor so that the memory owned by the object can be freed when it is no longer needed.
void
foo__DESTROY(foo_t *self)
{
free(self->name);
free(self);
}
This might start to look a little like a Perl module, and when we look at the Perl code that binds to this code, you will see why. First lets prepare the FFI::Platypus instance and specify the correct api version:
my $ffi = FFI::Platypus->new( api => 2 );
The bundle interface is only supported with api version 1, so if you try to use version 0 it will not work. Next we define an object type for foo_t
which will associate it with the Perl class Foo
.
$ffi->type('object(Foo)' => 'foo_t');
As object type is a blessed reference to an opaque (default) or integer type which can be used as a Perl object. Platypus does the translating of Perl object to and from the foo_t pointers that the C code understands. For more details on Platypus types see FFI::Platypus::Type.
Next we set the mangler on the Platypus instance so that we can refer to function names without the foo__
prefix. You could just not use the prefix in your C code and skip this step, or you could refer to the function names in their full in your Perl code, however, this saves extra typing and allows you to bundle more than one class with your Perl code without having to worry about name conflicts.
$ffi->mangler(sub {
my $name = shift;
$name =~ s/^/foo__/;
$name;
});
Finally we let Platypus know that we will be bundling code.
$ffi->bundle;
By default, this searches for the appropriate place for your dynamic libraries using the current package. In some cases you may need to override this, for example if your dist is named Foo-Bar
but your specific class is named Foo::Bar::Baz
, you'd want something like this:
package Foo::Bar::Baz;
use FFI::Platypus 2.00;
my $ffi = FFI::Platypus->new( api => 2 );
$ffi->bundle('Foo::Bar');
...
Now, finally we can attach the methods for our class:
$ffi->attach( new => [ 'string', 'int' ] => 'foo_t' );
$ffi->attach( name => [ 'foo_t' ] => 'string' );
$ffi->attach( value => [ 'foo_t' ] => 'int' );
$ffi->attach( DESTROY => [ 'foo_t' ] => 'void' );
Note that we do not have to include the foo__
prefix because of the way we set up the mangler. If we hadn't done that then we could instead attach with the full names:
$ffi->attach( [ 'foo__new' => 'new' ] => [ 'string', 'int' ] => 'foo_t' );
$ffi->attach( [ 'foo__name' => 'name' ] => [ 'foo_t' ] => 'string' );
...
You're done! You can now use this class. Lets write a test to make sure it works,
use Test2::V0;
use Foo;
my $foo = Foo->new("platypus", 10);
isa_ok $foo, 'Foo';
is $foo->name, "platypus";
is $foo->value, 10;
done_testing;
and use prove
to check that it works:
% prove -lvm
t/foo.t ..
ok 1 - An object of class 'Foo' isa 'Foo'
ok 2
ok 3
1..3
ok
All tests successful.
Files=1, Tests=3, 0 wallclock secs ( 0.02 usr 0.00 sys + 0.14 cusr 0.03 csys = 0.19 CPU)
Result: PASS
Platypus automatically compiles and links the dynamic library for you:
% ls ffi/_build
foo.c.o libFoo.so
The C code will be rebuilt next time if the source code is newer than the object or dynamic libraries files. If the source files are not changed, then it won't be rebuilt to save time. If you are using the code without MakeMaker, or another build system you are responsible for cleaning up these files. This is intended as a convenience to allow you to test your code without having to invoke MakeMaker, or dzil
or whatever build system you are using.
When you distribute your module though, you will want the dynamic library built just once at build-time and installed correctly so that it can be found at run-time. You don't need to make any changes to your C or Perl code, but you do need to tell MakeMaker to build and install the appropriate files using FFI::Build::MM:
use ExtUtils::MakeMaker;
use FFI::Build::MM;
my $fbmm = FFI::Build::MM->new;
WriteMakefile(
$fbmm->mm_args(
NAME => 'Foo',
DISTNAME => 'Foo',
VERSION => '1.00',
# ...
)
);
sub MY::postamble
{
$fbmm->mm_postamble;
}
And we can invoke all the normal MakeMaker style stuff and our C code will be compiled, linked and installed at the appropriate steps.
% perl Makefile.PL
Generating a Unix-style Makefile
Writing Makefile for Foo
Writing MYMETA.yml and MYMETA.json
% make
cp lib/Foo.pm blib/lib/Foo.pm
"/Users/ollisg/perl5/perlbrew/perls/perl-5.30.0/bin/perl" -MFFI::Build::MM=cmd -e fbx_build
CC ffi/foo.c
LD blib/lib/auto/share/dist/Foo/lib/libFoo.dylib
% make test
"/Users/ollisg/perl5/perlbrew/perls/perl-5.30.0/bin/perl" -MFFI::Build::MM=cmd -e fbx_build
"/Users/ollisg/perl5/perlbrew/perls/perl-5.30.0/bin/perl" -MFFI::Build::MM=cmd -e fbx_test
PERL_DL_NONLAZY=1 "/Users/ollisg/perl5/perlbrew/perls/perl-5.30.0/bin/perl" "-MExtUtils::Command::MM" "-MTest::Harness" "-e" "undef *Test::Harness::Switches; test_harness(0, 'blib/lib', 'blib/arch')" t/*.t
t/foo.t .. ok
All tests successful.
Files=1, Tests=3, 0 wallclock secs ( 0.01 usr 0.00 sys + 0.06 cusr 0.01 csys = 0.08 CPU)
Result: PASS
If the Makefile.PL
file above looks overly complicated, you can use the Dist::Zilla::Plugin::FFI::Build plugin to simplify your life if you are using Dist::Zilla:
[FFI::Build]
version = 1.04
Specifying version 1.04 will ensure that any .o
or .so
files are pruned from your build tree and not distributed by mistake.
Initialization example
The bundle interface also gives you entry points which will be called automatically when your code is loaded and unloaded if they are found.
ffi_pl_bundle_init
-
void ffi_pl_bundle_init(const char *package, int argc, void *argv[]);
Called when the dynamic library is loaded.
package
is the Perl package that calledbundle
from Perl space.argc
andargv
represents an array of opaque pointers that can be passed as an array to bundle as the last argument. (the countargc
is a little redundant becauseargv
is also NULL terminated). ffi_pl_bundle_constant
-
void ffi_pl_bundle_constant(const char *package, ffi_platypus_constant_t *c);
Called immediately after
ffi_pl_bundle_init
, and is intended to allow you to set Perl constants from C space. For details on how this works and what methods you can call on theffi_platypus_constant_t
instance, see FFI::Platypus::Constant. ffi_pl_bundle_fini
-
void ffi_pl_bundle_fini(const char *package);
Called when the dynamic library is unloaded.
package
is the Perl package that calledbundle
from Perl space when the library was loaded. CAVEAT: if you attach any functions then this will never be called, because attaching functions locks the Platypus instance into memory along with the libraries which it is using.
Here is an example that passes the version and a callback back into Perl space that emulates the Perl 5.10 say
feature.
ffi/init.c
:
#include <ffi_platypus_bundle.h>
char buffer[512];
const char *version;
void (*say)(const char *);
void
ffi_pl_bundle_init(const char *package, int argc, void *argv[])
{
version = argv[0];
say = argv[1];
say("in init!");
snprintf(buffer, 512, "package = %s, version = %s", package, version);
say(buffer);
snprintf(buffer, 512, "args = %d", argc);
say(buffer);
}
void
ffi_pl_bundle_fini(const char *package)
{
say("in fini!");
}
lib/Init.pm
:
package Init;
use strict;
use warnings;
use FFI::Platypus 2.00;
our $VERSION = '1.00';
{
my $ffi = FFI::Platypus->new( api => 2 );
my $say = $ffi->closure(sub {
my $string = shift;
print "$string\n";
});
$ffi->bundle([
$ffi->cast( 'string' => 'opaque', $VERSION ),
$ffi->cast( '(string)->void' => 'opaque', $say ),
]);
undef $ffi;
undef $say;
}
1;
The deinitialization order for the $say
callback and the $ffi
instance is essential here, so we do it manually with undef
:
undef $ffi;
undef $say;
First we deallocate $ffi
which calls ffi_pl_bundle_fini
, which calls $say
, so we want to make sure the latter is still allocated. Once ffi_pl_bundle_fini
is done, we can safely deallocate $say
.
If ffi_pl_bundle_fini
didn't call back into Perl space like this then we don't have to be as careful about deallocating things in Perl space.
Compiler or linker flags example
There are times when you will want to specify your own compiler and linker flags for the C code that you are bundling. The TL;DR
is that you can put a .fbx
file in your ffi
directory. This is a Perl script that returns a hash reference that is passed into the FFI::Build constructor. This allows you to set a number of options, including compiler and linker flags. A more detailed example follows:
You may want or need to set compiler and linker flags for your bundled C code. For example, say we have a header file, but instead of putting it in the ffi
directory we want to put it in a separate directory called include
.
include/answer.h
:
#ifndef ANSWER_H
#define ANSWER_H
int answer(void);
#endif
ffi/answer.c
:
int
answer(void)
{
/* the answer to life the universe and everything */
return 42;
}
lib/Answer.pm
:
package Answer;
use strict;
use warnings;
use FFI::Platypus 2.00;
use Exporter qw( import );
our @EXPORT = qw( answer );
my $ffi = FFI::Platypus->new( api => 2 );
$ffi->bundle;
$ffi->attach( answer => [] => 'int' );
1;
If you try to use this module just as-is you will get an error, about not being able to find the header file. Probably something like this:
ffi/answer.c:1:10: fatal error: 'answer.h' file not found
So we put a answer.fbx
file in the ffi
directory. (In case you are wondering FBX stands for "Ffi Build and file eXtensions should whenever possible be three characters long"). The name of the file can be anything so long as it ends in .fbx
, we just choose answer
here because that is the name of the project.
ffi/answer.fbx
:
our $DIR;
return {
cflags => "-I/include",
source => "$DIR/*.c",
}
The $DIR
variable is provided by the builder code. It is the root of the distribution, and is helpful if you need a fully qualified path. In this case you could have also used ffi/*.c
.
The script returns a hash reference which is passed into the FFI::Build constructor, so you can use any of the options supported by that class. Now we should be able to use our bundled module:
% perl -Ilib -MAnswer=answer -E 'say answer'
42
Using bundled code with Alien.
A useful technique is to use Platypus with Alien technology. The Alien namespace is reserved for providing external non-Perl dependencies for CPAN modules. The nominal Alien module when installed looks for the library locally, and if it can't be found it fetches it from the internet, builds it, and installs it in a private directory so that it can be used by other CPAN modules. For Aliens that provide shared libraries, and that have simple interfaces that do not require additional C code you can easily just pass the shared libraries to Platypus directly. For modules that require some bundled C code and an Alien you have to link the Alien library with your bundled code. If the Alien uses the Alien::Base interface then all you have to do is give the name of the Alien to FFI::Build.
For example, the bzip2
library provides an interface that requires the caller to allocate a C struct
and then pass it to its various functions. The struct
is actually pretty simple and you could use FFI::C or FFI::Platypus::Record, but here is an example of how you would connect bundled C code with an Alien.
ffi/compress.c
:
#include <bzlib.h>
#include <stdlib.h>
int
bzip2__new(bz_stream **stream, int blockSize100k, int verbosity, int workFactor )
{
*stream = malloc(sizeof(bz_stream));
(*stream)->bzalloc = NULL;
(*stream)->bzfree = NULL;
(*stream)->opaque = NULL;
return BZ2_bzCompressInit(*stream, blockSize100k, verbosity, workFactor );
}
lib/Bzip2.pm
:
package Bzip2;
use strict;
use warnings;
use FFI::Platypus 2.00;
use FFI::Platypus::Memory qw( free );
my $ffi = FFI::Platypus->new( api => 2 );
$ffi->bundle;
$ffi->mangler(sub {
my $name = shift;
$name =~ s/^/bzip2__/ unless $name =~ /^BZ2_/;
$name;
});
=head2 new
my $bzip2 = Bzip2->new($block_size_100k, $verbosity, $work_flow);
=cut
$ffi->attach( new => ['opaque*', 'int', 'int', 'int'] => 'int' => sub {
my $xsub = shift;
my $class = shift;
my $ptr;
my $ret = $xsub->(\$ptr, @_);
return bless \$ptr, $class;
});
$ffi->attach( [ BZ2_bzCompressEnd => 'DESTROY' ] => ['opaque'] => 'int' => sub {
my $xsub = shift;
my $self = shift;
my $ret = $xsub->($$self);
free $$self;
});
1;
The .fbx
file that goes with this to make it work with Alien::Libbz2 is now pretty trivial:
ffi/bz2.fbx
:
{
alien => ['Alien::Libbz2'],
source => ['ffi/*.c'],
};
AUTHOR
Author: Graham Ollis <plicease@cpan.org>
Contributors:
Bakkiaraj Murugesan (bakkiaraj)
Dylan Cali (calid)
pipcet
Zaki Mughal (zmughal)
Fitz Elliott (felliott)
Vickenty Fesunov (vyf)
Gregor Herrmann (gregoa)
Shlomi Fish (shlomif)
Damyan Ivanov
Ilya Pavlov (Ilya33)
Petr Písař (ppisar)
Mohammad S Anwar (MANWAR)
Håkon Hægland (hakonhagland, HAKONH)
Meredith (merrilymeredith, MHOWARD)
Diab Jerius (DJERIUS)
Eric Brine (IKEGAMI)
szTheory
José Joaquín Atria (JJATRIA)
Pete Houston (openstrike, HOUSTON)
Lukas Mai (MAUKE)
COPYRIGHT AND LICENSE
This software is copyright (c) 2015-2022 by Graham Ollis.
This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.