Native calling interface
Call into dynamic libraries that follow the C calling convention
The simplest imaginable use of
NativeCall would look something like this:
use NativeCall;sub some_argless_function() is native('something')some_argless_function();
The first line imports various traits and types. The next line looks like a relatively ordinary Raku sub declaration—with a twist. We use the
native trait in order to specify that the sub is actually defined in a native library. The platform-specific extension (e.g.,
.dll), as well as any customary prefixes (e.g.,
lib) will be added for you.
The first time you call "some_argless_function", the "libsomething" will be loaded and the "some_argless_function" will be located in it. A call will then be made. Subsequent calls will be faster, since the symbol handle is retained.
Of course, most functions take arguments or return values—but everything else that you can do is just adding to this simple pattern of declaring a Raku sub, naming it after the symbol you want to call and marking it with the
You will also need to declare and use native types, which cannot be imported from the shared library in the same way. Please check the native types page for more information.
Except in the case you are using your own compiled libraries, or any other kind of bundled library, shared libraries are versioned, i.e., they will be in a file
libfoo.so.x.y.z, and this shared library will be symlinked to
libxxx.so.x. By default, Raku will pick up that file if it's the only existing one. This is why it safer, and advisable, to always include a version, this way:
sub some_argless_function() is native('foo',v1.2.3)
Please check the section on the ABI/API version for more information.
A non-core Raku module, App::GPTrixie, has a Raku program,
gptrixie, to establish a starting code framework when initiating a new NativeCall project. It describes itself as A tool to generate NativeCall code from C headers and its Github repository is here. For now it is only for the C language but C++ support is planned.
Sometimes you want the name of your Raku subroutine to be different from the name used in the library you're loading. Maybe the name is long or has different casing or is otherwise cumbersome within the context of the module you are trying to create.
NativeCall provides a
symbol trait for you to specify the name of the native routine in your library that may be different from your Raku subroutine name.
unit ;use NativeCall;our sub init() is native('foo') is symbol('FOO_INIT')
libfoo there is a routine called
FOO_INIT but, since we're creating a module called
Foo and we'd rather call the routine as
Foo::init (instead of
Foo::FOO_INIT), we use the
symbol trait to specify the name of the symbol in
libfoo and call the subroutine whatever we want (
init in this case).
Normal Raku signatures and the
returns trait are used in order to convey the type of arguments a native function expects and what it returns. Here is an example.
use NativeCall;sub add(int32, int32) returns int32 is native("calculator")
Here, we have declared that the function takes two 32-bit integers and returns a 32-bit integer. You can find the other types that you may pass in the native types page. Note that the lack of a
returns trait is used to indicate
void return type. Do not use the
void type anywhere except in the
For strings, there is an additional
encoded trait to give some extra hints on how to do the marshaling.
use NativeCall;sub message_box(Str is encoded('utf8')) is native('gui')
To specify how to marshal string return types, just apply this trait to the routine itself.
use NativeCall;sub input_box() returns Str is encoded('utf8') is native('gui')
Note that a
NULL string pointer can be passed by using the
Str type object; a
NULL return will also be represented by the type object.
If the C function requires the lifetime of data, say some buffer of type
CArray[uint8], passed by pointer to exceed the function call, you have to ensure that the backing
CArray[uint8] Raku object is not destroyed before the C function expects it.
This is particularly important for strings, as the (usually preferable)
is encoded-way of passing strings creates only a temporary object that is automatically destroyed after the call. In this case, the argument must be manually encoded:
use NativeCall;# void set_foo(const char *)# Records a char pointer for later usagesub set_foo(CArray[uint8]) is native('foo')# void use_foo(void)# Uses the pointer stored by set_foo()sub use_foo() is native('foo')my = "FOO";# Manually marshal the $string into $array. Take care that $array# is not destroyed (e.g. by going out of scope) while the library# still holds on to it, after set_foo().## $string.encode takes care of UTF-8 encoding. If $array is used# as a string by the native function, don't forget to append the# NUL byte that terminates a C string: ------------vmy = CArray[uint8].new(.encode.list, 0);set_foo();# ...use_foo();# It's fine if $array goes out of scope starting from here.
When working with native functions, sometimes you need to specify what kind of native data structure is going to be used.
is repr is the term employed for that.
use NativeCall;is repr('CStruct')sub clock_gettime(uint32 , timespec --> uint32) is native ;my timespec .=new;my = clock_gettime( 0, );say "$result, $this-time"; # OUTPUT: «0, timespec<65385480>␤»
The original function we are calling, clock_gettime, uses a pointer to the
timespec struct as second argument. We declare it as a class here, but specify its representation as
is repr('CStruct'), to indicate it corresponds to a C data structure. When we create an object of that class, we are creating exactly the kind of pointer
clock_gettime expects. This way, data can be transferred seamlessly to and from the native interface.
When the signature of your native function needs a pointer to some native type (
uint32, etc.) all you need to do is declare the argument
use NativeCall;# C prototype is void my_version(int *major, int *minor)sub my_version(int32 is rw, int32 is rw) is native('foo')my_version(my int32 , my int32 ); # Pass a pointer to
Sometimes you need to get a pointer (for example, a library handle) back from a C library. You don't care about what it points to - you just need to keep hold of it. The
Pointer type provides for this.
use NativeCall;sub Foo_init() returns Pointer is native("foo")sub Foo_free(Pointer) is native("foo")
This works out OK, but you may fancy working with a type named something better than
Pointer. It turns out that any class with the representation
CPointer can serve this role. This means you can expose libraries that work on handles by writing a class like this:
use NativeCall;is repr('CPointer')
Note that the
CPointer representation can do nothing more than hold a C pointer. This means that your class cannot have extra attributes. However, for simple libraries this may be a neat way to expose an object oriented interface to it.
Of course, you can always have an empty class:
And just use the class as you would use
Pointer, but with potential for better type safety and more readable code.
Once again, type objects are used to represent
C libraries can expose pointers to C functions as return values of functions and as members of structures such as structs and unions.
Example of invoking a function pointer
$fptr returned by a function
f, using a signature defining the desired function parameters and return value:
sub f() returns Pointer is native('mylib')my = f();my = nativecast(:(Str, size_t --> int32), );say newfunc("test", 4);
NativeCall has some support for arrays. It is constrained to work with machine-size integers, doubles and strings, sized numeric types, arrays of pointers, arrays of structs, and arrays of arrays.
Raku arrays, which support amongst other things laziness, are laid out in memory in a radically different way to C arrays. Therefore, the NativeCall library offers a much more primitive CArray type, which you must use if working with C arrays.
Here is an example of passing a C array.
sub RenderBarChart(Str, int32, CArray[Str], CArray[num64]) is native("chart")my := CArray[Str].new; = 'Me'; = 'You'; = 'Hagrid';my := CArray[num64].new; = 59.5e0; = 61.2e0; = 180.7e0;RenderBarChart('Weights (kg)', 3, , );
Note that binding was used to
@titles, not assignment! If you assign, you are putting the values into a Raku array, and it will not work out. If this all freaks you out, forget you ever knew anything about the
@ sigil and just use
$ all the way when using NativeCall.
use NativeCall;my = CArray[Str].new; = 'Me'; = 'You'; = 'Hagrid';
Getting return values for arrays works out just the same.
Some library APIs may take an array as a buffer that will be populated by the C function and, for instance, return the actual number of items populated:
use NativeCall;sub get_n_ints(CArray[int32], int32) returns int32 is native('ints')
In these cases it is important that the CArray has at least the number of elements that are going to be populated before passing it to the native subroutine, otherwise the C function may stomp all over Raku's memory leading to possibly unpredictable behavior:
my = 10;my = CArray[int32].allocate(); # instantiates an array with 10 elementsmy = get_n_ints(, );
allocate was introduced in Rakudo 2018.05. Before that, you had to use this mechanism to extend an array to a number of elements:
my = CArray[int32].new;my = 10;[ - 1] = 0; # extend the array to 10 items
The memory management of arrays is important to understand. When you create an array yourself, then you can add elements to it as you wish and it will be expanded for you as required. However, this may result in it being moved in memory (assignments to existing elements will never cause this, however). This means you'd best know what you're doing if you twiddle with an array after passing it to a C library.
By contrast, when a C library returns an array to you, then the memory can not be managed by NativeCall, and it doesn't know where the array ends. Presumably, something in the library API tells you this (for example, you know that when you see a null element, you should read no further). Note that NativeCall can offer you no protection whatsoever here - do the wrong thing, and you will get a segfault or cause memory corruption. This isn't a shortcoming of NativeCall, it's the way the big bad native world works. Scared? Here, have a hug. Good luck!
Besides the usual methods available on every Raku instance,
CArray provides the following methods that can be used to interact with it from the Raku point of view:
elemsprovides the number of elements within the array;
AT-POSprovides a specific element at the given position (starting from zero). This method is not intended to be used directly, but invoked using the subscript notation
listprovides the List of elements within the array building it from the native array iterator.
As an example, consider the following simple piece of code:
use NativeCall;my = CArray[int32].new( 1, 2, 3, 4, 5 );say 'Number of elements: ' ~ .elems;# walk the arrayfor .list -># get every element by its index-based positionfor 0...elems - 1 ->
that produces the following output
Number of elements: 5Current element is: 1Current element is: 2Current element is: 3Current element is: 4Current element is: 5Element at position 0 is 1Element at position 1 is 2Element at position 2 is 3Element at position 3 is 4Element at position 4 is 5
Thanks to representation polymorphism, it's possible to declare a normal looking Raku class that, under the hood, stores its attributes in the same way a C compiler would lay them out in a similar struct definition. All it takes is a quick use of the
The attributes can only be of the types that NativeCall knows how to marshal into struct fields. Currently, structs can contain machine-sized integers, doubles, strings, and other NativeCall objects (
CArrays, and those using the
CStruct reprs). Other than that, you can do the usual set of things you would with a class; you could even have some of the attributes come from roles or have them inherited from another class. Of course, methods are completely fine too. Go wild!
CStruct objects are passed to native functions by reference and native functions must also return
CStruct objects by reference. The memory management rules for these references are very much like the rules for arrays, though simpler since a struct is never resized. When you create a struct, the memory is managed for you and when the variable(s) pointing to the instance of a
CStruct go away, the memory will be freed when the GC gets to it. When a
CStruct-based type is used as the return type of a native function, the memory is not managed for you by the GC.
NativeCall currently doesn't put object members in containers, so assigning new values to them (with
=) doesn't work. Instead, you have to bind (with
:=) new values to the private members:
As you may have predicted by now, a
NULL pointer is represented by the type object of the struct type.
Likewise, it is possible to declare a Raku class that stores its attributes the same way a C compiler would lay them out in a similar
union definition; using the
use NativeCall;is repr('CUnion')say nativesizeof(MyUnion.new); # OUTPUT: «8␤»# ie. max(sizeof(MyUnion.flags32), sizeof(MyUnion.flags64))
CUnions can be in turn referenced by—or embedded into—a surrounding
CUnion. To say the former we use
has as usual, and to do the latter we use the
HAS declarator instead:
is repr('CStruct')say nativesizeof(MyStruct.new); # OUTPUT: «16␤»# ie. sizeof(struct Point *) + sizeof(int32_t)is repr('CStruct')say nativesizeof(MyStruct2.new); # OUTPUT: «24␤»# ie. sizeof(struct Point) + sizeof(int32_t)
When allocating a struct for use as a struct, make sure that you allocate your own memory in your C functions. If you're passing a struct into a C function which needs a
char* allocated ahead of time, be sure to assign a container for a variable of type
Str prior to passing your struct into the function.
In this code we first set up our members,
$.an_int32. After that we declare our
init_struct() function for the
init() method to wrap around; this function is then called from
BUILD to effectively assign the values before returning the created object.
Please note that BUILD is binding attributes of native types, such as
$.an_int32. You can always bind native types this way.
typedef struct a_string_and_an_int32_t_a_string_and_an_int32_t;
Here's the structure. Notice how we've got a
char * there.
void init_struct(a_string_and_an_int32_t *target, char *str, int32_t int32)
In this function we initialize the C structure by assigning an integer by value, and passing the string by reference. The function allocates memory that it points
char *a_string to within the structure as it copies the string. (Note you will also have to manage deallocation of the memory as well to avoid memory leaks.)
# A long time ago in a galaxy far, far away...my = AStringAndAnInt.new(a_string => "str", an_int => 123);say "foo is and ";# OUTPUT: «foo is str and 123␤»
You can type your
Pointer by passing the type as a parameter. It works with the native type but also with
CStruct defined types. NativeCall will not implicitly allocate the memory for it even when calling
new on them. It's mostly useful in the case of a C routine returning a pointer, or if it's a pointer embedded in a
use NativeCall;sub strdup(Str --> Pointer[Str]) is nativemy Pointer[Str] = strdup("Success!");say .deref;
You have to call
Pointers to access the embedded type. In the example above, declaring the type of the pointer avoids typecasting error when dereferenced. Please note that the original
strdup returns a pointer to
char; we are using
my Pointer[int32] ; #For a pointer on int32;my Pointer[MyCstruct] = some_c_routine();my MyCstruct = .deref;say .field1;
It's quite common for a native function to return a pointer to an array of elements. Typed pointers can be dereferenced as an array to obtain individual elements.
my = 5;# returns a pointer to an array of length $nmy Pointer[Point] = some_other_c_routine();# display the 5 elements in the arrayfor 1 .. ->
Pointers can also be updated to reference successive elements in the array:
my Pointer[Point] = ;# show differences between successive pointsfor 1 ..^
Void pointers can also be used by declaring them
Pointer[void]. Please consult the native types documentation for more information on the subject.
Let's say there is some C code that caches strings passed, like so:
#include <stdlib.h>static char *__VERSION;char *get_version()char *set_version(char *version)
If you were to write bindings for
set_version, they would initially look like this, but will not work as intended:
sub get_version(--> Str) is native('./version')sub set_version(Str --> Str) is native('./version')say set_version('1.0.0'); # 1.0.0say get_version; # Differs on each runsay set_version('1.0.1'); # Double free; segfaults
This code segfaults on the second
set_version call because it tries to free the string passed on the first call after the garbage collector had already done so. If the garbage collector shouldn't free a string passed to a native function, use
explicitly-manage with it:
say set_version(explicitly-manage('1.0.0')); # 1.0.0say get_version; # 1.0.0say set_version(explicitly-manage('1.0.1')); # 1.0.1say get_version; # 1.0.1
Bear in mind all memory management for explicitly managed strings must be handled by the C library itself or through the NativeCall API to prevent memory leaks.
my = Blob.new(0x22, 0x33);my = nativecast(Pointer, );
$src can then be used as an argument for any native function that takes a
Pointer. The opposite, putting values pointed to by a
Pointer into a
Buf or using it to initialize a
Blob is not directly supported. You might want to use
NativeHelpers::Blob to do this kind of operations.
my = blob-from-pointer( , :2elems, :type(Blob[int8]));say ;
NativeCall also supports native functions that take functions as arguments. One example of this is using function pointers as callbacks in an event-driven system. When binding these functions via NativeCall, one needs only provide the equivalent signature as a constraint on the code parameter. In the case of NativeCall, however, as of Rakudo 2019.07, a space between the function argument and the signature, and the colon of a normal
Signature literal is omitted, as in:
use NativeCall;# void SetCallback(int (*callback)(const char *))my sub SetCallback( (Str --> int32)) is native('mylib')
Note: the native code is responsible for memory management of values passed to Raku callbacks this way. In other words, NativeCall will not free() strings passed to callbacks.
It is important that any code passed as a native callback handles its exceptions and, if applicable, returns an appropriate error value to the native code that invoked it. It is not allowed to throw an exception out of a native callback, and doing so will lead to process termination.
native trait accepts the library name, the full path, or a subroutine returning either of the two. When using the library name, the name is assumed to be prepended with
lib and appended with
.so (or just appended with
.dll on Windows), and will be searched for in the paths in the
PATH on Windows) environment variable.
use NativeCall;constant LIBMYSQL = 'mysqlclient';constant LIBFOO = '/usr/lib/libfoo.so.1';sub LIBBAR# and latersub mysql_affected_rows returns int32 is native(LIBMYSQL) ;sub bar is native(LIBFOO)sub baz is native(LIBBAR)
You can also put an incomplete path like './foo' and NativeCall will automatically put the right extension according to the platform specification. If you wish to suppress this expansion, simply pass the string as the body of a block.
sub bar is native()
BE CAREFUL: the
native trait and
constant are evaluated at compile time. Don't write a constant that depends on a dynamic variable like:
# WRONG:constant LIBMYSQL = <P6LIB_MYSQLCLIENT> || 'mysqlclient';
This will keep the value given at compile time. A module will be precompiled and
LIBMYSQL will keep the value it acquires when the module gets precompiled.
If you write
native('foo') NativeCall will search
libfoo.so under Unix like system (
libfoo.dynlib on OS X,
foo.dll on win32). In most modern system it will require you or the user of your module to install the development package because it's recommended to always provide an API/ABI version to a shared library, so
libfoo.so ends often being a symbolic link provided only by a development package.
To avoid that, the
native trait allows you to specify the API/ABI version. It can be a full version or just a part of it. (Try to stick to Major version, some BSD code does not care for Minor.)
use NativeCall;sub foo1 is native('foo', v1) # Will try to load libfoo.so.1sub foo2 is native('foo', v1.2.3) # Will try to load libfoo.so.1.2.3my List = ('foo', 'v1');sub foo3 is native()
native trait also accepts a
Callable as argument, allowing you to provide your own way to handle the way it will find the library file to load.
use NativeCall;sub foo is native(sub )
It will only be called at the first invocation of the sub.
If you want to call a C function that's already loaded, either from the standard library or from your own program, you can omit the value, so
For example on a UNIX-like operating system, you could use the following code to print the home directory of the current user:
use NativeCall;my is repr('CStruct')sub getuid() returns uint32 is native ;sub getpwuid(uint32 ) returns PwStruct is native ;say getpwuid(getuid()).pw_dir;
Though of course
$*HOME is a much easier way :-)!
Variables exported by a library – also named "global" or "extern" variables – can be accessed using
cglobal. For example:
my := cglobal('libc.so.6', 'errno', int32)
This code binds to
$var a new Proxy object that redirects all its accesses to the integer variable named "errno" as exported by the
NativeCall offers support to use classes and methods from C++ as shown in https://github.com/rakudo/rakudo/blob/master/t/04-nativecall/13-cpp-mangling.t (and its associated C++ file). Note that at the moment it's not as tested and developed as C support.
NativeCall library exports several subroutines to help you work with data from native libraries.
sub nativecast(, ) is export(:DEFAULT)
This will cast the
$source to an object of
$target-type. The source pointer will typically have been obtained from a call to a native subroutine that returns a pointer or as a member of a
struct, this may be specified as
void * in the
C library definition for instance, but you may also cast from a pointer to a less specific type to a more specific one.
As a special case, if a Signature is supplied as
$target-type then a
subroutine will be returned which will call the native function pointed to by
$source in the same way as a subroutine declared with the
native trait. This is described in Function Pointers.
sub cglobal(, , ) is export is rw
This returns a Proxy object that provides access to the
$symbol that is exposed by the specified library. The library can be specified in the same ways that they can be to the
sub nativesizeof() is export(:DEFAULT)
This returns the size in bytes of the supplied object, it can be thought of as being equivalent to
sizeof in C. The object can be a builtin native type such as
CArray or a class with the
sub explicitly-manage() is export(:DEFAULT)
This returns a
CStr object for the given
Str. If the string returned is passed to a NativeCall subroutine, it will not be freed by the runtime's garbage collector.
Some specific examples, and instructions to use examples above in particular platforms.
The PostgreSQL examples in DBIish make use of the NativeCall library and
is native to use the native
_putenv function call in Windows.
NOTE: Please bear in mind that, under the hood, Debian has substituted MySQL with MariaDB since the Stretch version, so if you want to install MySQL, use MySQL APT repository instead of the default repository.
To use the MySQL example in DBIish, you'll need to install MySQL server locally; on Debian-esque systems it can be installed with something like:
wget https://dev.mysql.com/get/mysql-apt-config_0.8.10-1_all.debsudo dpkg -i mysql-apt-config_0.8.10-1_all.deb # Don't forget to select 5.6.xsudo apt-get updatesudo apt-get install mysql-community-server -ysudo apt-get install libmysqlclient18 -y
Prepare your system along these lines before trying out the examples:
$ mysql -u root -pSET PASSWORD = PASSWORD('sa');DROP DATABASE test;CREATE DATABASE test;
Here is an example of a Windows API call:
use NativeCall;sub MessageBoxA(int32, Str, Str, int32)returns int32is native('user32')MessageBoxA(0, "We have NativeCall", "ohai", 64);
This is an example for calling a standard function and using the returned information in a Raku program.
getaddrinfo is a POSIX standard function for obtaining network information about a network node, e.g.,
google.com. It is an interesting function to look at because it illustrates a number of the elements of NativeCall.
The Linux manual provides the following information about the C callable function:
int getaddrinfo(const char *node, const char *service,const struct addrinfo *hints,struct addrinfo **res);
The function returns a response code 0 for success and 1 for error. The data are extracted from a linked list of
addrinfo elements, with the first element pointed to by
From the table of NativeCall types we know that an
int32. We also know that a
char * is one of the forms C for a C
Str, which maps simply to Str. But
addrinfo is a structure, which means we will need to write our own type class. However, the function declaration is straightforward:
sub getaddrinfo( Str , Str , Addrinfo , Pointer is rw )returns int32is native
$res is to be written by the function, so it must be labeled with the
is rw trait. Since the library is standard POSIX, the library name can be the type definition or null.
We now have to handle structure
Addrinfo. The Linux Manual provides this information:
int, char* parts are straightforward. Some research indicates that
socklen_t can be architecture dependent, but is an unsigned integer of at least 32 bits. So
socklen_t can be mapped to the
The complication is
sockaddr which differs depending on whether
ai_socktype is undefined, INET, or INET6 (a standard v4 IP address or a v6 address).
So we create a Raku
class to map to the C
struct addrinfo; while we're at it, we also create another class for
SockAddr which is needed for it.
is repr('CStruct')is repr('CStruct')
is rw on the last three attributes reflects that these were defined in C to be pointers.
The important thing here for mapping to a C
Struct is the structure of the state part of the class, that is the attributes. However, a class can have methods and
NativeCall does not 'touch' them for mapping to C. This means that we can add extra methods to the class to unpack the attributes in a more readable manner, e.g.,
By defining an appropriate
flags will return a string of keys rather than a bit packed integer.
The most useful information in the
sockaddr structure is the address of node, which depends on the family of the Socket. So we can add method
address to the Raku class that interprets the address depending on the family.
In order to get a human readable IP address, there is the C function
inet_ntop which returns a
char * given a buffer with the
Putting all these together, leads to the following program:
#!/usr/bin/env rakuuse v6;use NativeCall;constant \INET_ADDRSTRLEN = 16;constant \INET6_ADDRSTRLEN = 46;(AF_UNSPEC => 0;AF_INET => 2;AF_INET6 => 10;);(SOCK_STREAM => 1;SOCK_DGRAM => 2;SOCK_RAW => 3;SOCK_RDM => 4;SOCK_SEQPACKET => 5;SOCK_DCCP => 6;SOCK_PACKET => 10;);(AI_PASSIVE => 0x0001;AI_CANONNAME => 0x0002;AI_NUMERICHOST => 0x0004;AI_V4MAPPED => 0x0008;AI_ALL => 0x0010;AI_ADDRCONFIG => 0x0020;AI_IDN => 0x0040;AI_CANONIDN => 0x0080;AI_IDN_ALLOW_UNASSIGNED => 0x0100;AI_IDN_USE_STD3_ASCII_RULES => 0x0200;AI_NUMERICSERV => 0x0400;);sub inet_ntop(int32, Pointer, Blob, int32 --> Str)is nativeis repr('CStruct')is repr('CStruct')is repr('CStruct')is repr('CStruct')sub getaddrinfo(Str , Str , Addrinfo ,Pointer is rw --> int32)is native ;sub freeaddrinfo(Pointer)is nativesub MAIN()
This produces the following output:
return val: 0Name: google.comAF_INET SOCK_STREAM188.8.131.52AF_INET SOCK_DGRAM184.108.40.206AF_INET SOCK_RAW220.127.116.11AF_INET6 SOCK_STREAM2607:f8b0:4006:800::200eAF_INET6 SOCK_DGRAM2607:f8b0:4006:800::200eAF_INET6 SOCK_RAW2607:f8b0:4006:800::200e