Skip to content

Write Perl bindings to non-Perl libraries with FFI. No XS required.

Notifications You must be signed in to change notification settings

PerlFFI/FFI-Platypus

Repository files navigation

FFI::Platypus static linux windows macos msys2-mingw

Write Perl bindings to non-Perl libraries with FFI. No XS required.

SYNOPSIS

use FFI::Platypus 2.00;

# for all new code you should use api => 2
my $ffi = FFI::Platypus->new(
  api => 2,
  lib => undef, # search libc
);

# call dynamically
$ffi->function( puts => ['string'] => 'int' )->call("hello world");

# attach as a xsub and call (much faster)
$ffi->attach( puts => ['string'] => 'int' );
puts("hello world");

DESCRIPTION

Platypus is a library for creating interfaces to machine code libraries written in languages like C, C++, Go, Fortran, Rust, Pascal. Essentially anything that gets compiled into machine code. This implementation uses libffi to accomplish this task. libffi is battle tested by a number of other scripting and virtual machine languages, such as Python and Ruby to serve a similar role. There are a number of reasons why you might want to write an extension with Platypus instead of XS:

  • FFI / Platypus does not require messing with the guts of Perl

    XS is less of an API and more of the guts of perl splayed out to do whatever you want. That may at times be very powerful, but it can also be a frustrating exercise in hair pulling.

  • FFI / Platypus is portable

    Lots of languages have FFI interfaces, and it is subjectively easier to port an extension written in FFI in Perl or another language to FFI in another language or Perl. One goal of the Platypus Project is to reduce common interface specifications to a common format like JSON that could be shared between different languages.

  • FFI / Platypus could be a bridge to Raku

    One of those "other" languages could be Raku and Raku already has an FFI interface I am told.

  • FFI / Platypus can be reimplemented

    In a bright future with multiple implementations of Perl 5, each interpreter will have its own implementation of Platypus, allowing extensions to be written once and used on multiple platforms, in much the same way that Ruby-FFI extensions can be use in Ruby, JRuby and Rubinius.

  • FFI / Platypus is pure perl (sorta)

    One Platypus script or module works on any platform where the libraries it uses are available. That means you can deploy your Platypus script in a shared filesystem where they may be run on different platforms. It also means that Platypus modules do not need to be installed in the platform specific Perl library path.

  • FFI / Platypus is not C or C++ centric

    XS is implemented primarily as a bunch of C macros, which requires at least some understanding of C, the C pre-processor, and some C++ caveats (since on some platforms Perl is compiled and linked with a C++ compiler). Platypus on the other hand could be used to call other compiled languages, like Fortran, Go, Rust, Pascal, C++, or even assembly, allowing you to focus on your strengths.

  • FFI / Platypus does not require a parser

    Inline isolates the extension developer from XS to some extent, but it also requires a parser. The various Inline language bindings are a great technical achievement, but I think writing a parser for every language that you want to interface with is a bit of an anti-pattern.

This document consists of an API reference, a set of examples, some support and development (for contributors) information. If you are new to Platypus or FFI, you may want to skip down to the EXAMPLES to get a taste of what you can do with Platypus.

Platypus has extensive documentation of types at FFI::Platypus::Type and its custom types API at FFI::Platypus::API.

You are strongly encouraged to use API level 2 for all new code. There are a number of improvements and design fixes that you get for free. You should even consider updating existing modules to use API level 2 where feasible. How do I do that you might ask? Simply pass in the API level to the platypus constructor.

my $ffi = FFI::Platypus->new( api => 2 );

The Platypus documentation has already been updated to assume API level 1.

CONSTRUCTORS

new

my $ffi = FFI::Platypus->new( api => 2, %options);

Create a new instance of FFI::Platypus.

Any types defined with this instance will be valid for this instance only, so you do not need to worry about stepping on the toes of other CPAN FFI / Platypus Authors.

Any functions found will be out of the list of libraries specified with the lib attribute.

options

  • api

    [version 0.91]

    Sets the API level. The recommended value for all new code is 2. The Platypus documentation assumes API level 2 except for a few places that specifically document older versions. You should only use a lower value for a legacy code base that cannot be migrated to a newer API level. Legal values are:

    • 0

      Original API level. See FFI::Platypus::TypeParser::Version0 for details on the differences.

    • 1

      Enable version 1 API type parser which allows pass-by-value records and type decoration on basic types.

    • 2

      Enable version 2 API. The Platypus documentation assumes this api level is set.

      API version 2 is identical to version 1, except:

      • Pointer functions that return NULL will return undef instead of empty list

        This fixes a long standing design bug in Platypus.

      • Array references may be passed to pointer argument types

        This replicates the behavior of array argument types with no size. So the types sint8* and sint8[] behave identically when an array reference is passed in. They differ in that, as before, you can pass a scalar reference into type sint8*.

      • The fixed string type can be specified without pointer modifier

        That is you can use string(10) instead of string(10)* as you were previously able to in API 0.

  • lib

    Either a pathname (string) or a list of pathnames (array ref of strings) to pre-populate the lib attribute. Use [undef] to search the current process for symbols.

    0.48

    undef (without the array reference) can be used to search the current process for symbols.

  • ignore_not_found

    [version 0.15]

    Set the ignore_not_found attribute.

  • lang

    [version 0.18]

    Set the lang attribute.

ATTRIBUTES

lib

$ffi->lib($path1, $path2, ...);
my @paths = $ffi->lib;

The list of libraries to search for symbols in.

The most portable and reliable way to find dynamic libraries is by using FFI::CheckLib, like this:

use FFI::CheckLib 0.06;
$ffi->lib(find_lib_or_die lib => 'archive');
  # finds libarchive.so on Linux
  #       libarchive.bundle on OS X
  #       libarchive.dll (or archive.dll) on Windows
  #       cygarchive-13.dll on Cygwin
  #       ...
  # and will die if it isn't found

FFI::CheckLib has a number of options, such as checking for specific symbols, etc. You should consult the documentation for that module.

As a special case, if you add undef as a "library" to be searched, Platypus will also search the current process for symbols. This is mostly useful for finding functions in the standard C library, without having to know the name of the standard c library for your platform (as it turns out it is different just about everywhere!).

You may also use the "find_lib" method as a shortcut:

$ffi->find_lib( lib => 'archive' );

ignore_not_found

[version 0.15]

$ffi->ignore_not_found(1);
my $ignore_not_found = $ffi->ignore_not_found;

Normally the attach and function methods will throw an exception if it cannot find the name of the function you provide it. This will change the behavior such that function will return undef when the function is not found and attach will ignore functions that are not found. This is useful when you are writing bindings to a library and have many optional functions and you do not wish to wrap every call to function or attach in an eval.

lang

[version 0.18]

$ffi->lang($language);

Specifies the foreign language that you will be interfacing with. The default is C. The foreign language specified with this attribute changes the default native types (for example, if you specify Rust, you will get i32 as an alias for sint32 instead of int as you do with C).

If the foreign language plugin supports it, this will also enable Platypus to find symbols using the demangled names (for example, if you specify CPP for C++ you can use method names like Foo::get_bar() with "attach" or "function".

api

[version 1.11]

my $level = $ffi->api;

Returns the API level of the Platypus instance.

METHODS

type

$ffi->type($typename);
$ffi->type($typename => $alias);

Define a type. The first argument is the native or C name of the type. The second argument (optional) is an alias name that you can use to refer to this new type. See FFI::Platypus::Type for legal type definitions.

Examples:

$ffi->type('sint32');            # only checks to see that sint32 is a valid type
$ffi->type('sint32' => 'myint'); # creates an alias myint for sint32
$ffi->type('bogus');             # dies with appropriate diagnostic

custom_type

$ffi->custom_type($alias => {
  native_type         => $native_type,
  native_to_perl      => $coderef,
  perl_to_native      => $coderef,
  perl_to_native_post => $coderef,
});

Define a custom type. See FFI::Platypus::Type#Custom-Types for details.

load_custom_type

$ffi->load_custom_type($name => $alias, @type_args);

Load the custom type defined in the module $name, and make an alias $alias. If the custom type requires any arguments, they may be passed in as @type_args. See FFI::Platypus::Type#Custom-Types for details.

If $name contains :: then it will be assumed to be a fully qualified package name. If not, then FFI::Platypus::Type:: will be prepended to it.

types

my @types = $ffi->types;
my @types = FFI::Platypus->types;

Returns the list of types that FFI knows about. This will include the native libffi types (example: sint32, opaque and double) and the normal C types (example: unsigned int, uint32_t), any types that you have defined using the type method, and custom types.

The list of types that Platypus knows about varies somewhat from platform to platform, FFI::Platypus::Type includes a list of the core types that you can always count on having access to.

It can also be called as a class method, in which case, no user defined or custom types will be included in the list.

type_meta

my $meta = $ffi->type_meta($type_name);
my $meta = FFI::Platypus->type_meta($type_name);

Returns a hash reference with the meta information for the given type.

It can also be called as a class method, in which case, you won't be able to get meta data on user defined types.

The format of the meta data is implementation dependent and subject to change. It may be useful for display or debugging.

Examples:

my $meta = $ffi->type_meta('int');        # standard int type
my $meta = $ffi->type_meta('int[64]');    # array of 64 ints
$ffi->type('int[128]' => 'myintarray');
my $meta = $ffi->type_meta('myintarray'); # array of 128 ints

mangler

$ffi->mangler(\&mangler);

Specify a customer mangler to be used for symbol lookup. This is usually useful when you are writing bindings for a library where all of the functions have the same prefix. Example:

$ffi->mangler(sub {
  my($symbol) = @_;
  return "foo_$symbol";
});

$ffi->function( get_bar => [] => 'int' );  # attaches foo_get_bar

my $f = $ffi->function( set_baz => ['int'] => 'void' );
$f->call(22); # calls foo_set_baz

function

my $function = $ffi->function($name => \@argument_types => $return_type);
my $function = $ffi->function($address => \@argument_types => $return_type);
my $function = $ffi->function($name => \@argument_types => $return_type, \&wrapper);
my $function = $ffi->function($address => \@argument_types => $return_type, \&wrapper);

Returns an object that is similar to a code reference in that it can be called like one.

Caveat: many situations require a real code reference, so at the price of a performance penalty you can get one like this:

my $function = $ffi->function(...);
my $coderef = sub { $function->(@_) };

It may be better, and faster to create a real Perl function using the attach method.

In addition to looking up a function by name you can provide the address of the symbol yourself:

my $address = $ffi->find_symbol('my_function');
my $function = $ffi->function($address => ...);

Under the covers, function uses find_symbol when you provide it with a name, but it is useful to keep this in mind as there are alternative ways of obtaining a functions address. Example: a C function could return the address of another C function that you might want to call.

[version 0.76]

If the last argument is a code reference, then it will be used as a wrapper around the function when called. The first argument to the wrapper will be the inner function, or if it is later attached an xsub. This can be used if you need to verify/modify input/output data.

Examples:

my $function = $ffi->function('my_function_name', ['int', 'string'] => 'string');
my $return_string = $function->(1, "hi there");

[version 0.91]

my $function = $ffi->function( $name => \@fixed_argument_types => \@var_argument_types => $return_type);
my $function = $ffi->function( $name => \@fixed_argument_types => \@var_argument_types => $return_type, \&wrapper);
my $function = $ffi->function( $name => \@fixed_argument_types => \@var_argument_types);
my $function = $ffi->function( $name => \@fixed_argument_types => \@var_argument_types => \&wrapper);

Version 0.91 and later allows you to creat functions for c variadic functions (such as printf, scanf, etc) which can take a variable number of arguments. The first set of arguments are the fixed set, the second set are the variable arguments to bind with. The variable argument types must be specified in order to create a function object, so if you need to call variadic function with different set of arguments then you will need to create a new function object each time:

# int printf(const char *fmt, ...);
$ffi->function( printf => ['string'] => ['int'] => 'int' )
    ->call("print integer %d\n", 42);
$ffi->function( printf => ['string'] => ['string'] => 'int' )
    ->call("print string %s\n", 'platypus');

Some older versions of libffi and possibly some platforms may not support variadic functions. If you try to create a one, then an exception will be thrown.

[version 1.26]

If the return type is omitted then void will be the assumed return type.

attach

$ffi->attach($name => \@argument_types => $return_type);
$ffi->attach([$c_name => $perl_name] => \@argument_types => $return_type);
$ffi->attach([$address => $perl_name] => \@argument_types => $return_type);
$ffi->attach($name => \@argument_types => $return_type, \&wrapper);
$ffi->attach([$c_name => $perl_name] => \@argument_types => $return_type, \&wrapper);
$ffi->attach([$address => $perl_name] => \@argument_types => $return_type, \&wrapper);

Find and attach a C function as a real live Perl xsub. The advantage of attaching a function over using the function method is that it is much much much faster since no object resolution needs to be done. The disadvantage is that it locks the function and the FFI::Platypus instance into memory permanently, since there is no way to deallocate an xsub.

If just one $name is given, then the function will be attached in Perl with the same name as it has in C. The second form allows you to give the Perl function a different name. You can also provide an address (the third form), just like with the function method.

Examples:

$ffi->attach('my_function_name', ['int', 'string'] => 'string');
$ffi->attach(['my_c_function_name' => 'my_perl_function_name'], ['int', 'string'] => 'string');
my $string1 = my_function_name($int);
my $string2 = my_perl_function_name($int);

[version 0.20]

If the last argument is a code reference, then it will be used as a wrapper around the attached xsub. The first argument to the wrapper will be the inner xsub. This can be used if you need to verify/modify input/output data.

Examples:

$ffi->attach('my_function', ['int', 'string'] => 'string', sub {
  my($my_function_xsub, $integer, $string) = @_;
  $integer++;
  $string .= " and another thing";
  my $return_string = $my_function_xsub->($integer, $string);
  $return_string =~ s/Belgium//; # HHGG remove profanity
  $return_string;
});

[version 0.91]

$ffi->attach($name => \@fixed_argument_types => \@var_argument_types, $return_type);
$ffi->attach($name => \@fixed_argument_types => \@var_argument_types, $return_type, \&wrapper);

As of version 0.91 you can attach a variadic functions, if it is supported by the platform / libffi that you are using. For details see the function documentation. If not supported by the implementation then an exception will be thrown.

closure

my $closure = $ffi->closure($coderef);
my $closure = FFI::Platypus->closure($coderef);

Prepares a code reference so that it can be used as a FFI closure (a Perl subroutine that can be called from C code). For details on closures, see FFI::Platypus::Type#Closures and FFI::Platypus::Closure.

cast

my $converted_value = $ffi->cast($original_type, $converted_type, $original_value);

The cast function converts an existing $original_value of type $original_type into one of type $converted_type. Not all types are supported, so care must be taken. For example, to get the address of a string, you can do this:

my $address = $ffi->cast('string' => 'opaque', $string_value);

Something that won't work is trying to cast an array to anything:

my $address = $ffi->cast('int[10]' => 'opaque', \@list);  # WRONG

attach_cast

$ffi->attach_cast("cast_name", $original_type, $converted_type);
$ffi->attach_cast("cast_name", $original_type, $converted_type, \&wrapper);
my $converted_value = cast_name($original_value);

This function attaches a cast as a permanent xsub. This will make it faster and may be useful if you are calling a particular cast a lot.

[version 1.26]

A wrapper may be added as the last argument to attach_cast and works just like the wrapper for attach and function methods.

sizeof

my $size = $ffi->sizeof($type);
my $size = FFI::Platypus->sizeof($type);

Returns the total size of the given type in bytes. For example to get the size of an integer:

my $intsize = $ffi->sizeof('int');   # usually 4
my $longsize = $ffi->sizeof('long'); # usually 4 or 8 depending on platform

You can also get the size of arrays

my $intarraysize = $ffi->sizeof('int[64]');  # usually 4*64
my $intarraysize = $ffi->sizeof('long[64]'); # usually 4*64 or 8*64
                                             # depending on platform

Keep in mind that "pointer" types will always be the pointer / word size for the platform that you are using. This includes strings, opaque and pointers to other types.

This function is not very fast, so you might want to save this value as a constant, particularly if you need the size in a loop with many iterations.

alignof

[version 0.21]

my $align = $ffi->alignof($type);

Returns the alignment of the given type in bytes.

kindof

[version 1.24]

my $kind = $ffi->kindof($type);

Returns the kind of a type. This is a string with a value of one of

  • void
  • scalar
  • string
  • closure
  • record
  • record-value
  • pointer
  • array
  • object

countof

[version 1.24]

my $count = $ffi->countof($type);

For array types returns the number of elements in the array (returns 0 for variable length array). For the void type returns 0. Returns 1 for all other types.

def

[version 1.24]

$ffi->def($package, $type, $value);
my $value = $ff->def($package, $type);

This method allows you to store data for types. If the $package is not provided, then the caller's package will be used. $type must be a legal Platypus type for the FFI::Platypus instance.

unitof

[version 1.24]

my $unittype = $ffi->unitof($type);

For array and pointer types, returns the basic type without the array or pointer part. In other words, for sin16[] or sint16* it will return sint16.

find_lib

[version 0.20]

$ffi->find_lib( lib => $libname );

This is just a shortcut for calling FFI::CheckLib#find_lib and updating the "lib" attribute appropriately. Care should be taken though, as this method simply passes its arguments to FFI::CheckLib#find_lib, so if your module or script is depending on a specific feature in FFI::CheckLib then make sure that you update your prerequisites appropriately.

find_symbol

my $address = $ffi->find_symbol($name);

Return the address of the given symbol (usually function).

bundle

[version 0.96 api = 1+]

$ffi->bundle($package, \@args);
$ffi->bundle(\@args);
$ffi->bundle($package);
$ffi->bundle;

This is an interface for bundling compiled code with your distribution intended to eventually replace the package method documented above. See FFI::Platypus::Bundle for details on how this works.

package

[version 0.15 api = 0]

$ffi->package($package, $file); # usually __PACKAGE__ and __FILE__ can be used
$ffi->package;                  # autodetect

Note: This method is officially discouraged in favor of bundle described above.

If you use FFI::Build (or the older deprecated Module::Build::FFI to bundle C code with your distribution, you can use this method to tell the FFI::Platypus instance to look for symbols that came with the dynamic library that was built when your distribution was installed.

abis

my $href = $ffi->abis;
my $href = FFI::Platypus->abis;

Get the legal ABIs supported by your platform and underlying implementation. What is supported can vary a lot by CPU and by platform, or even between 32 and 64 bit on the same CPU and platform. They keys are the "ABI" names, also known as "calling conventions". The values are integers used internally by the implementation to represent those ABIs.

abi

$ffi->abi($name);

Set the ABI or calling convention for use in subsequent calls to "function" or "attach". May be either a string name or integer value from the "abis" method above.

EXAMPLES

Here are some examples. These examples are provided in full with the Platypus distribution in the "examples" directory. There are also some more examples in FFI::Platypus::Type that are related to types.

Passing and Returning Integers

C Source

int add(int a, int b) {
  return a+b;
}

Perl Source

use FFI::Platypus 2.00;
use FFI::CheckLib qw( find_lib_or_die );
use File::Basename qw( dirname );

my $ffi = FFI::Platypus->new( api => 2, lib => './add.so' );
$ffi->attach( add => ['int', 'int'] => 'int' );

print add(1,2), "\n";  # prints 3

Execute

$ cc -shared -o add.so add.c
$ perl add.pl
3

Discussion

Basic types like integers and floating points are the easiest to pass across the FFI boundary. Because they are values that are passed on the stack (or through registers) you don't need to worry about memory allocations or ownership.

Here we are building our own C dynamic library using the native C compiler on a Unix like platform. The exact incantation that you will use to do this would unfortunately depend on your platform and C compiler.

By default, Platypus uses the Platypus C language plugin, which gives you easy access to many of the basic types used by C APIs. (for example int, unsigned long, double, size_t and others).

If you are working with another language like Fortran, Go, Rust or Zig, you will find similar examples where you can use the Platypus language plugin for that language and use the native types.

String Arguments (with puts)

C API

cppreference - puts

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new( api => 2, lib => undef );
$ffi->attach( puts => ['string'] => 'int' );

puts("hello world");

Execute

$ perl puts.pl
hello world

Discussion

Passing strings into a C function as an argument is also pretty easy using Platypus. Just use the string type, which is equivalent to the C <char *> or const char * types.

In this example we are using the C Standard Library's puts function, so we don't need to build our own C code. We do still need to tell Platypus where to look for the puts symbol though, which is why we set lib to undef. This is a special value which tells Platypus to search the Perl runtime executable itself (including any dynamic libraries) for symbols. That helpfully includes the C Standard Library.

Returning Strings

C Source

#include <string.h>
#include <stdlib.h>

const char *
string_reverse(const char *input)
{
  static char *output = NULL;
  int i, len;

  if(output != NULL)
    free(output);

  if(input == NULL)
    return NULL;

  len = strlen(input);
  output = malloc(len+1);

  for(i=0; input[i]; i++)
    output[len-i-1] = input[i];
  output[len] = '\0';

  return output;
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './string_reverse.so',
);

$ffi->attach( string_reverse => ['string'] => 'string' );

print string_reverse("\nHello world");

string_reverse(undef);

Execute

$ cc -shared -o string_reverse.so string_reverse.c
$ perl string_reverse.pl
dlrow olleH

Discussion

The C code here takes an input ASCII string and reverses it, returning the result. Note that it retains ownership of the string, the caller is expected to use it before the next call to reverse_string, or copy it.

The Perl code simply declares the return value as string and is very simple. This does bring up an inconsistency though, strings passed in to a function as arguments are passed by reference, whereas the return value is copied! This is usually what you want because C APIs usually follow this pattern where you are expected to make your own copy of the string.

At the end of the program we call reverse_string with undef, which gets translated to C as NULL. This allows it to free the output buffer so that the memory will not leak.

Returning and Freeing Strings with Embedded NULLs

C Source

#include <string.h>
#include <stdlib.h>

char *
string_crypt(const char *input, int len, const char *key)
{
  char *output;
  int i, n;

  if(input == NULL)
    return NULL;

  output = malloc(len+1);
  output[len] = '\0';

  for(i=0, n=0; i<len; i++, n++) {
    if(key[n] == '\0')
      n = 0;
    output[i] = input[i] ^ key[n];
  }

  return output;
}

void
string_crypt_free(char *output)
{
  if(output != NULL)
    free(output);
}

Perl Source

use FFI::Platypus 2.00;
use FFI::Platypus::Buffer qw( buffer_to_scalar );
use YAML ();

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './xor_cipher.so',
);

$ffi->attach( string_crypt_free => ['opaque'] );

$ffi->attach( string_crypt => ['string','int','string'] => 'opaque' => sub{
  my($xsub, $input, $key) = @_;
  my $ptr = $xsub->($input, length($input), $key);
  my $output = buffer_to_scalar $ptr, length($input);
  string_crypt_free($ptr);
  return $output;
});

my $orig = "hello world";
my $key  = "foobar";

print YAML::Dump($orig);
my $encrypted = string_crypt($orig, $key);
print YAML::Dump($encrypted);
my $decrypted = string_crypt($encrypted, $key);
print YAML::Dump($decrypted);

Execute

$ cc -shared -o xor_cipher.so xor_cipher.c
$ perl xor_cipher.pl
--- hello world
--- "\x0e\n\x03\x0e\x0eR\x11\0\x1d\x0e\x05"
--- hello world

Discussion

The C code here also returns a string, but it has some different expectations, so we can't just use the string type like we did in the previous example and copy the string.

This C code implements a simple XOR cipher. Given an input string and a key it returns an encrypted or decrypted output string where the characters are XORd with the key. There are some challenges here though. First the input and output strings can have embedded NULLs in them. For the string passed in, we can provide the length of the input string. For the output, the string type expects a NULL terminated string, so we can't use that. So instead we get a pointer to the output using the opaque type. Because we know that the output string is the same length as the input string we can convert the pointer to a regular Perl string using the buffer_to_scalar function. (For more details about working with buffers and strings see FFI::Platypus::Buffer).

Next, the C code here does not keep the pointer to the output string, as in the previous example. We are expected to call string_encrypt_free when we are done. Since we are getting the pointer back from the C code instead of copying the string that is easy to do.

Finally, we are using a wrapper to hide a lot of this complexity from our caller. The last argument to the attach call is a code reference which will wrap around the C function, which is passed in as the first argument of the wrapper. This is a good practice when writing modules, to hide the complexity of C.

Pointers

C Source

void
swap(int *a, int *b)
{
  int tmp = *b;
  *b = *a;
  *a = tmp;
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './swap.so',
);

$ffi->attach( swap => ['int*','int*'] );

my $a = 1;
my $b = 2;

print "[a,b] = [$a,$b]\n";

swap( \$a, \$b );

print "[a,b] = [$a,$b]\n";

Execute

$ cc -shared -o swap.so swap.c
$ perl swap.pl
[a,b] = [1,2]
[a,b] = [2,1]

Discussion

Pointers are often use in C APIs to return simple values like this. Platypus provides access to pointers to primitive types by appending * to the primitive type. Here for example we are using int* to create a function that takes two pointers to integers and swaps their values.

When calling the function from Perl we pass in a reference to a scalar. Strictly speaking Perl allows modifying the argument values to subroutines, so we could have allowed just passing in a scalar, but in the design of Platypus we decided that forcing the use of a reference here emphasizes that you are passing a reference to the variable, not just the value.

Not pictured in this example, but you can also pass in undef for a pointer value and that will be translated into NULL on the C side. You can also return a pointer to a primitive type from a function, again this will be returned to Perl as a reference to a scalar. Platypus also supports string pointers (string*). (Though the C equivalent to a string* is a double pointer to char char**).

Opaque Pointers (objects)

C Source

#include <string.h>
#include <stdlib.h>

typedef struct person_t {
  char *name;
  unsigned int age;
} person_t;

person_t *
person_new(const char *name, unsigned int age) {
  person_t *self = malloc(sizeof(person_t));
  self->name = strdup(name);
  self->age  = age;
}

const char *
person_name(person_t *self) {
  return self->name;
}

unsigned int
person_age(person_t *self) {
  return self->age;
}

void
person_free(person_t *self) {
  free(self->name);
  free(self);
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './person.so',
);

$ffi->type( 'opaque' => 'person_t' );

$ffi->attach( person_new =>  ['string','unsigned int'] => 'person_t'       );
$ffi->attach( person_name => ['person_t']              => 'string'       );
$ffi->attach( person_age =>  ['person_t']              => 'unsigned int' );
$ffi->attach( person_free => ['person_t']                                  );

my $person = person_new( 'Roger Frooble Bits', 35 );

print "name = ", person_name($person), "\n";
print "age  = ", person_age($person),  "\n";

person_free($person);

Execute

$ cc -shared -o person.so person.c
$ perl person.pl
name = Roger Frooble Bits
age  = 35

Discussion

An opaque pointer is a pointer (memory address) that is pointing to something but you do not know the structure of that something. In C this is usually a void*, but it could also be a pointer to a struct without a defined body.

This is often used to as an abstraction around objects in C. Here in the C code we have a person_t struct with functions to create (a constructor), free (a destructor) and query it (methods).

The Perl code can then use the constructor, methods and destructors without having to understand the internals. The person_t internals can also be changed without having to modify the calling code.

We use the Platypus type method to create an alias of opaque called person_t. While this is not necessary, it does make the Perl code easier to understand.

In later examples we will see how to hide the use of opaque types further using the object type, but for some code direct use of opaque is appropriate.

Opaque Pointers (buffers and strings)

C API

Perl Source

use FFI::Platypus 2.00;
use FFI::Platypus::Memory qw( malloc free memcpy strdup );

my $ffi = FFI::Platypus->new( api => 2 );
my $buffer = malloc 14;
my $ptr_string = strdup("hello there!!\n");

memcpy $buffer, $ptr_string, 15;

print $ffi->cast('opaque' => 'string', $buffer);

free $ptr_string;
free $buffer;

Execute

$ perl malloc.pl
hello there!!

Discussion

Another useful application of the opaque type is for dealing with buffers, and C strings that you do not immediately need to convert into Perl strings. This example is completely contrived, but we are using malloc to create a buffer of 14 bytes. We create a C string using strdup, and then copy it into the buffer using memcpy. When we are done with the opaque pointers we can free them using free since they. (This is generally only okay when freeing memory that was allocated by malloc, which is the case for strdup).

These memory tools, along with others are provided by the FFI::Platypus::Memory module, which is worth reviewing when you need to manipulate memory from Perl when writing your FFI code.

Just to verify that the memcpy did the right thing we convert the buffer into a Perl string and print it out using the Platypus cast method.

Arrays

C Source

void
array_reverse(int a[], int len) {
  int tmp, i;

  for(i=0; i < len/2; i++) {
    tmp = a[i];
    a[i] = a[len-i-1];
    a[len-i-1] = tmp;
  }
}

void
array_reverse10(int a[10]) {
  array_reverse(a, 10);
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './array_reverse.so',
);

$ffi->attach( array_reverse   => ['int[]','int'] );
$ffi->attach( array_reverse10 => ['int[10]'] );

my @a = (1..10);
array_reverse10( \@a );
print "$_ " for @a;
print "\n";

@a = (1..20);
array_reverse( \@a, 20 );
print "$_ " for @a;
print "\n";

Execute

$ cc -shared -o array_reverse.so array_reverse.c
$ perl array_reverse.pl
10 9 8 7 6 5 4 3 2 1
20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Discussion

Arrays in C are passed as pointers, so the C code here reverses the array in place, rather than returning it. Arrays can also be fixed or variable length. If the array is variable length the length of the array must be provided in some way. In this case we explicitly pass in a length. Another way might be to end the array with 0, if you don't otherwise expect any 0 to appear in your data. For this reason, Platypus adds a zero (or NULL in the case of pointers) element at the end of the array when passing it into a variable length array type, although we do not use it here.

With Platypus you can declare an array type as being either fixed or variable length. Because Perl stores arrays in completely differently than C, a temporary array is created by Platypus, passed into the C function as a pointer. When the function returns the array is re-read by Platypus and the Perl array is updated with the new values. The temporary array is then freed.

You can use any primitive type for arrays, even string. You can also return an array from a function. As in our discussion about strings, when you return an array the value is copied, which is usually what you want.

Pointers as Arrays

C Source

#include <stdlib.h>

int
array_sum(const int *a) {
  int i, sum;
  if(a == NULL)
    return -1;
  for(i=0, sum=0; a[i] != 0; i++)
    sum += a[i];
  return sum;
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './array_sum.so',
);

$ffi->attach( array_sum => ['int*'] => 'int' );

print array_sum(undef), "\n";     # -1
print array_sum([0]), "\n";       # 0
print array_sum([1,2,3,0]), "\n"; # 6

Execute

$ cc -shared -o array_sum.so array_sum.c
$ perl array_sum.pl
-1
0
6

Discussion

Starting with the Platypus version 2 API, you can also pass an array reference in to a pointer argument.

In C pointer and array arguments are often used somewhat interchangeably. In this example we have an array_sum function that takes a zero terminated array of integers and computes the sum. If the pointer to the array is zero (0) then we return -1 to indicate an error.

This is the main advantage from Perl for using pointer argument rather than an array one: the array argument will not let you pass in undef / NULL.

Sending Strings to GUI on Unix with libnotify

C API

Libnotify Reference Manual

Perl Source

use FFI::CheckLib;
use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => find_lib_or_die(lib => 'notify'),
);

$ffi->attach( notify_init              => ['string']                                  );
$ffi->attach( notify_uninit            => []                                          );
$ffi->attach( notify_notification_new  => ['string', 'string', 'string']  => 'opaque' );
$ffi->attach( notify_notification_show => ['opaque', 'opaque']                        );

my $message = join "\n",
  "Hello from Platypus!",
  "Welcome to the fun",
  "world of FFI";

notify_init('Platypus Hello');
my $n = notify_notification_new('Platypus Hello World', $message, 'dialog-information');
notify_notification_show($n, undef);
notify_uninit();

Execute

$ perl notify.pl

And this is what it will look like:

Test

Discussion

The GNOME project provides an API to send notifications to its desktop environment. Nothing here is particularly new: all of the types and techniques are ones that we have seen before, except we are using a third party library, instead of using our own C code or the standard C library functions.

When using a third party library you have to know the name or location of it, which is not typically portable, so here we use FFI::CheckLib's find_lib_or_die function. If the library is not found the script will die with a useful diagnostic. FFI::CheckLib has a number of useful features and will integrate nicely with Alien::Build based Aliens.

The Win32 API with MessageBoxW

Win32 API

MessageBoxW function (winuser.h)

Perl Source

use utf8;
use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api  => 2,
  lib  => [undef],
);

# see FFI::Platypus::Lang::Win32
$ffi->lang('Win32');

# Send a Unicode string to the Windows API MessageBoxW function.
use constant MB_OK                   => 0x00000000;
use constant MB_DEFAULT_DESKTOP_ONLY => 0x00020000;
$ffi->attach( [MessageBoxW => 'MessageBox'] => [ 'HWND', 'LPCWSTR', 'LPCWSTR', 'UINT'] => 'int' );
MessageBox(undef, "I ❤️ Platypus", "Confession", MB_OK|MB_DEFAULT_DESKTOP_ONLY);

Execute

$ perl win32_messagebox.pl

And this is what it will look like:

Test

Discussion

The API used by Microsoft Windows presents some unique challenges. On 32 bit systems a different ABI is used than what is used by the standard C library. It also provides a rats nest of type aliases. Finally if you want to talk Unicode to any of the Windows API you will need to use UTF-16LE instead of UTF-8 which is native to Perl. (The Win32 API refers to these as LPWSTR and LPCWSTR types). As much as possible the Win32 "language" plugin attempts to handle these challenges transparently. For more details see FFI::Platypus::Lang::Win32.

Discussion

The libnotify library is a desktop GUI notification system for the GNOME Desktop environment. This script sends a notification event that should show up as a balloon, for me it did so in the upper right hand corner of my screen.

Structured Data Records (by pointer or by reference)

C API

cppreference - localtime

Perl Source

use FFI::Platypus 2.00;
use FFI::C;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => [undef],
);
FFI::C->ffi($ffi);

package Unix::TimeStruct {

  FFI::C->struct(tm => [
    tm_sec    => 'int',
    tm_min    => 'int',
    tm_hour   => 'int',
    tm_mday   => 'int',
    tm_mon    => 'int',
    tm_year   => 'int',
    tm_wday   => 'int',
    tm_yday   => 'int',
    tm_isdst  => 'int',
    tm_gmtoff => 'long',
    _tm_zone  => 'opaque',
  ]);

  # For now 'string' is unsupported by FFI::C, but we
  # can cast the time zone from an opaque pointer to
  # string.
  sub tm_zone {
    my $self = shift;
    $ffi->cast('opaque', 'string', $self->_tm_zone);
  }

  # attach the C localtime function
  $ffi->attach( localtime => ['time_t*'] => 'tm', sub {
    my($inner, $class, $time) = @_;
    $time = time unless defined $time;
    $inner->(\$time);
  });
}

# now we can actually use our Unix::TimeStruct class
my $time = Unix::TimeStruct->localtime;
printf "time is %d:%d:%d %s\n",
  $time->tm_hour,
  $time->tm_min,
  $time->tm_sec,
  $time->tm_zone;

Execute

$ perl time_struct.pl
time is 3:48:19 MDT

Discussion

C and other machine code languages frequently provide interfaces that include structured data records (defined using the struct keyword in C). Some libraries will provide an API which you are expected to read or write before and/or after passing them along to the library.

For C pointers to strict, union, nested struct and nested union structures, the easiest interface to use is via FFI::C. If you are working with a struct that must be passed by value (not pointers), then you will want to use FFI::Platypus::Record class instead. We will discuss an example of that next.

The C localtime function takes a pointer to a C struct. We simply define the members of the struct using the FFI::C struct method. Because we used the ffi method to tell FFI::C to use our local instance of FFI::Platypus it registers the tm type for us, and we can just start using it as a return type!

Structured Data Records (on stack or by value)

C Source

#include <stdint.h>
#include <string.h>

typedef struct color_t {
   char    name[8];
   uint8_t red;
   uint8_t green;
   uint8_t blue;
} color_t;

color_t
color_increase_red(color_t color, uint8_t amount)
{
  strcpy(color.name, "reddish");
  color.red += amount;
  return color;
}

Perl Source

use FFI::Platypus 2.00;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => './color.so'
);

package Color {

  use FFI::Platypus::Record;
  use overload
    '""' => sub { shift->as_string },
    bool => sub { 1 }, fallback => 1;

  record_layout_1($ffi,
    'string(8)' => 'name', qw(
    uint8     red
    uint8     green
    uint8     blue
  ));

  sub as_string {
    my($self) = @_;
    sprintf "%s: [red:%02x green:%02x blue:%02x]",
      $self->name, $self->red, $self->green, $self->blue;
  }

}

$ffi->type('record(Color)' => 'color_t');
$ffi->attach( color_increase_red => ['color_t','uint8'] => 'color_t' );

my $gray = Color->new(
  name  => 'gray',
  red   => 0xDC,
  green => 0xDC,
  blue  => 0xDC,
);

my $slightly_red = color_increase_red($gray, 20);

print "$gray\n";
print "$slightly_red\n";

Execute

$ cc -shared -o color.so color.c
$ perl color.pl
gray: [red:dc green:dc blue:dc]
reddish: [red:f0 green:dc blue:dc]

Discussion

In the C source of this example, we pass a C struct by value by copying it onto the stack. On the Perl side we create a Color class using FFI::Platypus::Record, which allows us to pass the structure the way the C source wants us to.

Generally you should only reach for FFI::Platypus::Record if you need to pass small records on the stack like this. For more complicated (including nested) data you want to use FFI::C using pointers.

Avoiding Copy Using Memory Windows (with libzmq3)

C API

ØMQ/3.2.6 API Reference

Perl Source

use constant ZMQ_IO_THREADS  => 1;
use constant ZMQ_MAX_SOCKETS => 2;
use constant ZMQ_REQ => 3;
use constant ZMQ_REP => 4;
use FFI::CheckLib qw( find_lib_or_die );
use FFI::Platypus 2.00;
use FFI::Platypus::Memory qw( malloc );
use FFI::Platypus::Buffer qw( scalar_to_buffer window );

my $endpoint = "ipc://zmq-ffi-$$";
my $ffi = FFI::Platypus->new(
  api => 2,
  lib => find_lib_or_die lib => 'zmq',
);

$ffi->attach(zmq_version => ['int*', 'int*', 'int*'] => 'void');

my($major,$minor,$patch);
zmq_version(\$major, \$minor, \$patch);
print "libzmq version $major.$minor.$patch\n";
die "this script only works with libzmq 3 or better" unless $major >= 3;

$ffi->type('opaque'       => 'zmq_context');
$ffi->type('opaque'       => 'zmq_socket');
$ffi->type('opaque'       => 'zmq_msg_t');
$ffi->attach(zmq_ctx_new  => [] => 'zmq_context');
$ffi->attach(zmq_ctx_set  => ['zmq_context', 'int', 'int'] => 'int');
$ffi->attach(zmq_socket   => ['zmq_context', 'int'] => 'zmq_socket');
$ffi->attach(zmq_connect  => ['opaque', 'string'] => 'int');
$ffi->attach(zmq_bind     => ['zmq_socket', 'string'] => 'int');
$ffi->attach(zmq_send     => ['zmq_socket', 'opaque', 'size_t', 'int'] => 'int');
$ffi->attach(zmq_msg_init => ['zmq_msg_t'] => 'int');
$ffi->attach(zmq_msg_recv => ['zmq_msg_t', 'zmq_socket', 'int'] => 'int');
$ffi->attach(zmq_msg_data => ['zmq_msg_t'] => 'opaque');
$ffi->attach(zmq_errno    => [] => 'int');
$ffi->attach(zmq_strerror => ['int'] => 'string');

my $context = zmq_ctx_new();
zmq_ctx_set($context, ZMQ_IO_THREADS, 1);

my $socket1 = zmq_socket($context, ZMQ_REQ);
zmq_connect($socket1, $endpoint);

my $socket2 = zmq_socket($context, ZMQ_REP);
zmq_bind($socket2, $endpoint);

{ # send
  our $sent_message = "hello there";
  my($pointer, $size) = scalar_to_buffer $sent_message;
  my $r = zmq_send($socket1, $pointer, $size, 0);
  die zmq_strerror(zmq_errno()) if $r == -1;
}

{ # recv
  my $msg_ptr  = malloc 100;
  zmq_msg_init($msg_ptr);
  my $size     = zmq_msg_recv($msg_ptr, $socket2, 0);
  die zmq_strerror(zmq_errno()) if $size == -1;
  my $data_ptr = zmq_msg_data($msg_ptr);
  window(my $recv_message, $data_ptr, $size);
  print "recv_message = $recv_message\n";
}

Execute

$ perl zmq3.pl
libzmq version 4.3.4
recv_message = hello there

Discussion

ØMQ is a high-performance asynchronous messaging library. There are a few things to note here.

Firstly, sometimes there may be multiple versions of a library in the wild and you may need to verify that the library on a system meets your needs (alternatively you could support multiple versions and configure your bindings dynamically). Here we use zmq_version to ask libzmq which version it is.

zmq_version returns the version number via three integer pointer arguments, so we use the pointer to integer type: int *. In order to pass pointer types, we pass a reference. In this case it is a reference to an undefined value, because zmq_version will write into the pointers the output values, but you can also pass in references to integers, floating point values and opaque pointer types. When the function returns the $major variable (and the others) has been updated and we can use it to verify that it supports the API that we require.

Finally we attach the necessary functions, send and receive a message. When we receive we use the FFI::Platypus::Buffer function window instead of buffer_to_scalar. They have a similar effect in that the provide a scalar from a region of memory, but window doesn't have to copy any data, so it is cheaper to call. The only downside is that a windowed scalar like this is read-only.

libarchive

C Documentation

https://www.libarchive.org/

Perl Source

use FFI::Platypus 2.00;
use FFI::CheckLib qw( find_lib_or_die );

# This example uses FreeBSD's libarchive to list the contents of any
# archive format that it suppors.  We've also filled out a part of
# the ArchiveWrite class that could be used for writing archive formats
# supported by libarchive

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => find_lib_or_die(lib => 'archive'),
);
$ffi->type('object(Archive)'      => 'archive_t');
$ffi->type('object(ArchiveRead)'  => 'archive_read_t');
$ffi->type('object(ArchiveWrite)' => 'archive_write_t');
$ffi->type('object(ArchiveEntry)' => 'archive_entry_t');

package Archive {
  # base class is "abstract" having no constructor or destructor

  $ffi->mangler(sub {
    my($name) = @_;
    "archive_$name";
  });
  $ffi->attach( error_string => ['archive_t'] => 'string' );
}

package ArchiveRead {
  our @ISA = qw( Archive );

  $ffi->mangler(sub {
    my($name) = @_;
    "archive_read_$name";
  });

  $ffi->attach( new                   => ['string']                        => 'archive_read_t' );
  $ffi->attach( [ free => 'DESTROY' ] => ['archive_t']                                         );
  $ffi->attach( support_filter_all    => ['archive_t']                     => 'int'            );
  $ffi->attach( support_format_all    => ['archive_t']                     => 'int'            );
  $ffi->attach( open_filename         => ['archive_t','string','size_t']   => 'int'            );
  $ffi->attach( next_header2          => ['archive_t', 'archive_entry_t' ] => 'int'            );
  $ffi->attach( data_skip             => ['archive_t']                     => 'int'            );
  # ... define additional read methods
}

package ArchiveWrite {

  our @ISA = qw( Archive );

  $ffi->mangler(sub {
    my($name) = @_;
    "archive_write_$name";
  });

  $ffi->attach( new                   => ['string'] => 'archive_write_t' );
  $ffi->attach( [ free => 'DESTROY' ] => ['archive_write_t'] );
  # ... define additional write methods
}

package ArchiveEntry {

  $ffi->mangler(sub {
    my($name) = @_;
    "archive_entry_$name";
  });

  $ffi->attach( new => ['string']     => 'archive_entry_t' );
  $ffi->attach( [ free => 'DESTROY' ] => ['archive_entry_t'] );
  $ffi->attach( pathname              => ['archive_entry_t'] => 'string' );
  # ... define additional entry methods
}

use constant ARCHIVE_OK => 0;

# this is a Perl version of the C code here:
# https://github.com/libarchive/libarchive/wiki/Examples#List_contents_of_Archive_stored_in_File

my $archive_filename = shift @ARGV;
unless(defined $archive_filename)
{
  print "usage: $0 archive.tar\n";
  exit;
}

my $archive = ArchiveRead->new;
$archive->support_filter_all;
$archive->support_format_all;

my $r = $archive->open_filename($archive_filename, 1024);
die "error opening $archive_filename: ", $archive->error_string
  unless $r == ARCHIVE_OK;

my $entry = ArchiveEntry->new;

while($archive->next_header2($entry) == ARCHIVE_OK)
{
  print $entry->pathname, "\n";
  $archive->data_skip;
}

Execute

$ perl archive_object.pl archive.tar
archive.pl
archive_object.pl

Discussion

libarchive is the implementation of tar for FreeBSD provided as a library and available on a number of platforms.

One interesting thing about libarchive is that it provides a kind of object oriented interface via opaque pointers. This example creates an abstract class Archive, and concrete classes ArchiveWrite, ArchiveRead and ArchiveEntry. The concrete classes can even be inherited from and extended just like any Perl classes because of the way the custom types are implemented. We use Platypus's object type for this implementation, which is a wrapper around an opaque (can also be an integer) type that is blessed into a particular class.

Another advanced feature of this example is that we define a mangler to modify the symbol resolution for each class. This means we can do this when we define a method for Archive:

$ffi->attach( support_filter_all => ['archive_t'] => 'int' );

Rather than this:

$ffi->attach(
  [ archive_read_support_filter_all => 'support_read_filter_all' ] =>
  ['archive_t'] => 'int' );
);

As nice as libarchive is, note that we have to shoehorn then archive_free function name into the Perl convention of using DESTROY as the destructor. We can easily do that for just this one function with:

$ffi->attach( [ free => 'DESTROY' ] => ['archive_t'] );

The libarchive is a large library with hundreds of methods. For comprehensive FFI bindings for libarchive see Archive::Libarchive.

unix open

C API

Input-output system calls in C

Perl Source

use FFI::Platypus 2.00;

{
  package FD;

  use constant O_RDONLY => 0;
  use constant O_WRONLY => 1;
  use constant O_RDWR   => 2;

  use constant IN  => bless \do { my $in=0  }, __PACKAGE__;
  use constant OUT => bless \do { my $out=1 }, __PACKAGE__;
  use constant ERR => bless \do { my $err=2 }, __PACKAGE__;

  my $ffi = FFI::Platypus->new( api => 2, lib => [undef]);

  $ffi->type('object(FD,int)' => 'fd');

  $ffi->attach( [ 'open' => 'new' ] => [ 'string', 'int', 'mode_t' ] => 'fd' => sub {
    my($xsub, $class, $fn, @rest) = @_;
    my $fd = $xsub->($fn, @rest);
    die "error opening $fn $!" if $$fd == -1;
    $fd;
  });

  $ffi->attach( write => ['fd', 'string', 'size_t' ] => 'ssize_t' );
  $ffi->attach( read  => ['fd', 'string', 'size_t' ] => 'ssize_t' );
  $ffi->attach( close => ['fd'] => 'int' );
}

my $fd = FD->new("file_handle.txt", FD::O_RDONLY);

my $buffer = "\0" x 10;

while(my $br = $fd->read($buffer, 10))
{
  FD::OUT->write($buffer, $br);
}

$fd->close;

Execute

$ perl file_handle.pl
Hello World

Discussion

The Unix file system calls use an integer handle for each open file. We can use the same object type that we used for libarchive above, except we let platypus know that the underlying type is int instead of opaque (the latter being the default for the object type). Mainly just for demonstration since Perl has much better IO libraries, but now we have an OO interface to the Unix IO functions.

Varadic Functions (with libcurl)

C API

Perl Source

use FFI::Platypus 2.00;
use FFI::CheckLib qw( find_lib_or_die );
use constant CURLOPT_URL => 10002;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => find_lib_or_die(lib => 'curl'),
);

my $curl_handle = $ffi->function( 'curl_easy_init' => [] => 'opaque' )
                      ->call;

$ffi->function( 'curl_easy_setopt' => ['opaque', 'enum' ] => ['string'] )
    ->call($curl_handle, CURLOPT_URL, "https://pl.atypus.org" );

$ffi->function( 'curl_easy_perform' => ['opaque' ] => 'enum' )
    ->call($curl_handle);

$ffi->function( 'curl_easy_cleanup' => ['opaque' ] )
    ->call($curl_handle);

Execute

$ perl curl.pl
<!doctype html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <title>pl.atypus.org - Home for the Perl Platypus Project</title>
...

Discussion

The libcurl library makes extensive use of "varadic" functions.

The C programming language and ABI have the concept of "varadic" functions that can take a variable number and variable type of arguments. Assuming you have a libffi that supports it (and most modern systems should), then you can create bindings to a varadic function by providing two sets of array references, one for the fixed arguments (for reasons, C varadic functions must have at least one) and one for variable arguments. In this example we call curl_easy_setopt as a varadic function.

For functions that have a large or infinite number of possible signatures it may be impracticable or impossible to attach them all. You can instead do as we did in this example, create a function object using the function method and call it immediately. This is not as performant either when you create or call as using the attach method, but in some cases the performance penalty may be worth it or unavoidable.

Callbacks (with libcurl)

C API

Perl Source

use FFI::Platypus 2.00;
use FFI::CheckLib qw( find_lib_or_die );
use FFI::Platypus::Buffer qw( window );
use constant CURLOPT_URL           => 10002;
use constant CURLOPT_WRITEFUNCTION => 20011;

my $ffi = FFI::Platypus->new(
  api => 2,
  lib => find_lib_or_die(lib => 'curl'),
);

my $curl_handle = $ffi->function( 'curl_easy_init' => [] => 'opaque' )
                      ->call;

$ffi->function( 'curl_easy_setopt' => [ 'opaque', 'enum' ] => ['string'] )
    ->call($curl_handle, CURLOPT_URL, "https://pl.atypus.org" );

my $html;

my $closure = $ffi->closure(sub {
  my($ptr, $len, $num, $user) = @_;
  window(my $buf, $ptr, $len*$num);
  $html .= $buf;
  return $len*$num;
});

$ffi->function( 'curl_easy_setopt' => [ 'opaque', 'enum' ] => ['(opaque,size_t,size_t,opaque)->size_t'] => 'enum' )
    ->call($curl_handle, CURLOPT_WRITEFUNCTION, $closure);

$ffi->function( 'curl_easy_perform' => [ 'opaque' ] => 'enum' )
    ->call($curl_handle);

$ffi->function( 'curl_easy_cleanup' => [ 'opaque' ] )
    ->call($curl_handle);

if($html =~ /<title>(.*?)<\/title>/) {
  print "$1\n";
}

Execute

$ perl curl_callback.pl
pl.atypus.org - Home for the Perl Platypus Project

Discussion

This example is similar to the previous one, except instead of letting libcurl write the content body to STDOUT, we give it a callback to send the data to instead. The closure method can be used to create a callback function pointer that can be called from C. The type for the callback is in the form (arg_type,arg_type,etc)->return_type where the argument types are in parentheticals with an arrow between the argument types and the return type.

Inside the closure or callback we use the window function from FFI::Platypus::Buffer again to avoid an extra copy. We still have to copy the buffer to append it to $hmtl but it is at least one less copy.

bundle your own code

C Source

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);
}

Perl Source

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;
}

Execute

With prove:

$ prove -lvm
t/foo.t ..
# Seeded srand with seed '20221105' from local date.
ok 1 - Foo=SCALAR->isa('Foo')
ok 2
ok 3
1..3
ok
All tests successful.
Files=1, Tests=3,  0 wallclock secs ( 0.00 usr  0.00 sys +  0.10 cusr  0.00 csys =  0.10 CPU)
Result: PASS

With ExtUtils::MakeMaker:

$ 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
"/home/ollisg/opt/perl/5.37.5/bin/perl5.37.5" -MFFI::Build::MM=cmd -e fbx_build
CC ffi/foo.c
LD blib/lib/auto/share/dist/Foo/lib/libFoo.so
$ make test
"/home/ollisg/opt/perl/5.37.5/bin/perl5.37.5" -MFFI::Build::MM=cmd -e fbx_build
"/home/ollisg/opt/perl/5.37.5/bin/perl5.37.5" -MFFI::Build::MM=cmd -e fbx_test
PERL_DL_NONLAZY=1 "/home/ollisg/opt/perl/5.37.5/bin/perl5.37.5" "-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,  1 wallclock secs ( 0.00 usr  0.00 sys +  0.03 cusr  0.00 csys =  0.03 CPU)
Result: PASS

Discussion

You can bundle your own C code with your Perl extension. There are a number of reasons you might want to do this Sometimes you need to optimize a tight loop for speed. Or you might need a little bit of glue code for your bindings to a library that isn't inherently FFI friendly. Either way what you want is the FFI::Build system on the install step and the FFI::Platypus::Bundle interface on the runtime step. If you are using Dist::Zilla for your distribution, you will also want to check out the Dist::Zilla::Plugin::FFI::Build plugin to make this as painless as possible.

One of the nice things about the bundle interface is that it is smart enough to work with either App::Prove or ExtUtils::MakeMaker. This means, unlike XS, you do not need to explicitly compile your C code in development mode, that will be done for you when you call $ffi->bundle

FAQ

How do I get constants defined as macros in C header files

This turns out to be a challenge for any language calling into C, which frequently uses #define macros to define constants like so:

#define FOO_STATIC  1
#define FOO_DYNAMIC 2
#define FOO_OTHER   3

As macros are expanded and their definitions are thrown away by the C pre-processor there isn't any way to get the name/value mappings from the compiled dynamic library.

You can manually create equivalent constants in your Perl source:

use constant FOO_STATIC  => 1;
use constant FOO_DYNAMIC => 2;
use constant FOO_OTHER   => 3;

If there are a lot of these types of constants you might want to consider using a tool (Convert::Binary::C can do this) that can extract the constants for you.

See also the "Integer constants" example in FFI::Platypus::Type.

You can also use the new Platypus bundle interface to define Perl constants from C space. This is more reliable, but does require a compiler at install time. It is recommended mainly for writing bindings against libraries that have constants that can vary widely from platform to platform. See FFI::Platypus::Constant for details.

What about enums?

The C enum types are integers. The underlying type is up to the platform, so Platypus provides enum and senum types for unsigned and singed enums respectively. At least some compilers treat signed and unsigned enums as different types. The enum values are essentially the same as macro constants described above from an FFI perspective. Thus the process of defining enum values is identical to the process of defining macro constants in Perl.

For more details on enumerated types see "Enum types" in FFI::Platypus::Type.

There is also a type plugin (FFI::Platypus::Type::Enum) that can be helpful in writing interfaces that use enums.

Memory leaks

There are a couple places where memory is allocated, but never deallocated that may look like memory leaks by tools designed to find memory leaks like valgrind. This memory is intended to be used for the lifetime of the perl process so there normally this isn't a problem unless you are embedding a Perl interpreter which doesn't closely match the lifetime of your overall application.

Specifically:

  • type cache

    some types are cached and not freed. These are needed as long as there are FFI functions that could be called.

  • attached functions

    Attaching a function as an xsub will definitely allocate memory that won't be freed because the xsub could be called at any time, including in END blocks.

The Platypus team plans on adding a hook to free some of this "leaked" memory for use cases where Perl and Platypus are embedded in a larger application where the lifetime of the Perl process is significantly smaller than the overall lifetime of the whole process.

I get seg faults on some platforms but not others with a library using pthreads.

On some platforms, Perl isn't linked with libpthreads if Perl threads are not enabled. On some platforms this doesn't seem to matter, libpthreads can be loaded at runtime without much ill-effect. (Linux from my experience doesn't seem to mind one way or the other). Some platforms are not happy about this, and about the only thing that you can do about it is to build Perl such that it links with libpthreads even if it isn't a threaded Perl.

This is not really an FFI issue, but a Perl issue, as you will have the same problem writing XS code for the such libraries.

Doesn't work on Perl 5.10.0.

The first point release of Perl 5.10 was buggy, and is not supported by Platypus. Please upgrade to a newer Perl.

CAVEATS

Platypus and Native Interfaces like libffi rely on the availability of dynamic libraries. Things not supported include:

  • Systems that lack dynamic library support

    Like MS-DOS

  • Systems that are not supported by libffi

    Like OpenVMS

  • Languages that do not support using dynamic libraries from other languages

    This used to be the case with Google's Go, but is no longer the case. This is a problem for C / XS code as well.

  • Languages that do not compile to machine code

    Like .NET based languages and Java.

The documentation has a bias toward using FFI / Platypus with C. This is my fault, as my background mainly in C/C++ programmer (when I am not writing Perl). In many places I use "C" as a short form for "any language that can generate machine code and is callable from C". I welcome pull requests to the Platypus core to address this issue. In an attempt to ease usage of Platypus by non C programmers, I have written a number of foreign language plugins for various popular languages (see the SEE ALSO below). These plugins come with examples specific to those languages, and documentation on common issues related to using those languages with FFI. In most cases these are available for easy adoption for those with the know-how or the willingness to learn. If your language doesn't have a plugin YET, that is just because you haven't written it yet.

SUPPORT

The intent of the FFI-Platypus team is to support the same versions of Perl that are supported by the Perl toolchain. As of this writing that means 5.16 and better.

IRC: #native on irc.perl.org

(click for instant chat room login)

If something does not work the way you think it should, or if you have a feature request, please open an issue on this project's GitHub Issue tracker:

https://github.com/perlFFI/FFI-Platypus/issues

CONTRIBUTING

If you have implemented a new feature or fixed a bug then you may make a pull request on this project's GitHub repository:

https://github.com/PerlFFI/FFI-Platypus/pulls

This project is developed using Dist::Zilla. The project's git repository also comes with the Makefile.PL file necessary for building, testing (and even installing if necessary) without Dist::Zilla. Please keep in mind though that these files are generated so if changes need to be made to those files they should be done through the project's dist.ini file. If you do use Dist::Zilla and already have the necessary plugins installed, then I encourage you to run dzil test before making any pull requests. This is not a requirement, however, I am happy to integrate especially smaller patches that need tweaking to fit the project standards. I may push back and ask you to write a test case or alter the formatting of a patch depending on the amount of time I have and the amount of code that your patch touches.

This project's GitHub issue tracker listed above is not Write-Only. If you want to contribute then feel free to browse through the existing issues and see if there is something you feel you might be good at and take a whack at the problem. I frequently open issues myself that I hope will be accomplished by someone in the future but do not have time to immediately implement myself.

Another good area to help out in is documentation. I try to make sure that there is good document coverage, that is there should be documentation describing all the public features and warnings about common pitfalls, but an outsider's or alternate view point on such things would be welcome; if you see something confusing or lacks sufficient detail I encourage documentation only pull requests to improve things.

The Platypus distribution comes with a test library named libtest that is normally automatically built by ./Build test. If you prefer to use prove or run tests directly, you can use the ./Build libtest command to build it. Example:

% perl Makefile.PL
% make
% make ffi-test
% prove -bv t
# or an individual test
% perl -Mblib t/ffi_platypus_memory.t

The build process also respects these environment variables:

  • FFI_PLATYPUS_DEBUG_FAKE32

    When building Platypus on 32 bit Perls, it will use the Math::Int64 C API and make Math::Int64 a prerequisite. Setting this environment variable will force Platypus to build with both of those options on a 64 bit Perl as well.

    % env FFI_PLATYPUS_DEBUG_FAKE32=1 perl Makefile.PL
    DEBUG_FAKE32:
      + making Math::Int64 a prereq
      + Using Math::Int64's C API to manipulate 64 bit values
    Generating a Unix-style Makefile
    Writing Makefile for FFI::Platypus
    Writing MYMETA.yml and MYMETA.json
    %
    
  • FFI_PLATYPUS_NO_ALLOCA

    Platypus uses the non-standard and somewhat controversial C function alloca by default on platforms that support it. I believe that Platypus uses it responsibly to allocate small amounts of memory for argument type parameters, and does not use it to allocate large structures like arrays or buffers. If you prefer not to use alloca despite these precautions, then you can turn its use off by setting this environment variable when you run Makefile.PL:

    helix% env FFI_PLATYPUS_NO_ALLOCA=1 perl Makefile.PL
    NO_ALLOCA:
      + alloca() will not be used, even if your platform supports it.
    Generating a Unix-style Makefile
    Writing Makefile for FFI::Platypus
    Writing MYMETA.yml and MYMETA.json
  • V

    When building platypus may hide some of the excessive output when probing and building, unless you set V to a true value.

    % env V=1 perl Makefile.PL
    % make V=1
    ...
    

Coding Guidelines

  • Do not hesitate to make code contribution. Making useful contributions is more important than following byzantine bureaucratic coding regulations. We can always tweak things later.
  • Please make an effort to follow existing coding style when making pull requests.
  • The intent of the FFI-Platypus team is to support the same versions of Perl that are supported by the Perl toolchain. As of this writing that means 5.16 and better. As such, please do not include any code that requires a newer version of Perl.

Performance Testing

As Mark Twain was fond of saying there are four types of lies: lies, damn lies, statistics and benchmarks. That being said, it can sometimes be helpful to compare the runtime performance of Platypus if you are making significant changes to the Platypus Core. For that I use `FFI-Performance`, which can be found in my GitHub repository here:

System integrators

This distribution uses Alien::FFI in fallback mode, meaning if the system doesn't provide pkg-config and libffi it will attempt to download libffi and build it from source. If you are including Platypus in a larger system (for example a Linux distribution) you only need to make sure to declare pkg-config or pkgconf and the development package for libffi as prereqs for this module.

SEE ALSO

Extending Platypus

  • FFI::Platypus::Type

    Type definitions for Platypus.

  • FFI::C

    Interface for defining structured data records for use with Platypus. It supports C struct, union, nested structures and arrays of all of those. It only supports passing these types by reference or pointer, so if you need to pass structured data by value see FFI::Platypus::Record below.

  • FFI::Platypus::Record

    Interface for defining structured data records for use with Platypus. Included in the Platypus core. Supports pass by value which is uncommon in C, but frequently used in languages like Rust and Go. Consider using FFI::C instead if you don't need to pass by value.

  • FFI::Platypus::API

    The custom types API for Platypus.

  • FFI::Platypus::Memory

    Memory functions for FFI.

Languages

Other Tools Related Tools Useful for FFI

  • FFI::CheckLib

    Find dynamic libraries in a portable way.

  • Convert::Binary::C

    A great interface for decoding C data structures, including structs, enums, #defines and more.

  • pack and unpack

    Native to Perl functions that can be used to decode C struct types.

  • C::Scan

    This module can extract constants and other useful objects from C header files that may be relevant to an FFI application. One downside is that its use may require development packages to be installed.

Other Foreign Function Interfaces

  • Dyn

    A wrapper around dyncall, which is itself an alternative to libffi.

  • NativeCall

    Promising interface to Platypus inspired by Raku.

  • Win32::API

    Microsoft Windows specific FFI style interface.

  • FFI

    Older, simpler, less featureful FFI. It used to be implemented using FSF's ffcall. Because ffcall has been unsupported for some time, I reimplemented this module using FFI::Platypus.

  • C::DynaLib

    Another FFI for Perl that doesn't appear to have worked for a long time.

  • C::Blocks

    Embed a tiny C compiler into your Perl scripts.

  • P5NCI

    Yet another FFI like interface that does not appear to be supported or under development anymore.

Other

  • Alien::FFI

    Provides libffi for Platypus during its configuration and build stages.

ACKNOWLEDGMENTS

In addition to the contributors mentioned below, I would like to acknowledge Brock Wilcox (AWWAIID) and Meredith Howard (MHOWARD) whose work on FFI::Sweet not only helped me get started with FFI but significantly influenced the design of Platypus.

Dan Book, who goes by Grinnz on IRC for answering user questions about FFI and Platypus.

In addition I'd like to thank Alessandro Ghedini (ALEXBIO) whose work on another Perl FFI library helped drive some of the development ideas for FFI::Platypus.

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.

About

Write Perl bindings to non-Perl libraries with FFI. No XS required.

Topics

Resources

Stars

Watchers

Forks

Packages

No packages published