Native code interoperability

Scala Native provides an interop layer that makes it easy to interact with foreign native code. This includes C and other languages that can expose APIs via C ABI (e.g. C++, D, Rust etc.)

All of the interop APIs discussed here are defined in scala.scalanative.unsafe package. For brevity, we’re going to refer to that namespace as just unsafe.

Extern objects

Extern objects are simple wrapper objects that demarcate scopes where methods are treated as their native C ABI-friendly counterparts. They are roughly analogous to header files with top-level function declarations in C.

For example, to call C’s malloc one might declare it as following:

import scala.scalanative.unsafe._

@extern
object libc {
  def malloc(size: CSize): Ptr[Byte] = extern
}

extern on the right hand side of the method definition signifies that the body of the method is defined elsewhere in a native library that is available on the library path (see Linking with native libraries). The signature of the external function must match the signature of the original C function (see Finding the right signature).

Finding the right signature

To find a correct signature for a given C function one must provide an equivalent Scala type for each of the arguments:

C Type Scala Type


void Unit bool unsafe.CBool char unsafe.CChar signed char unsafe.CSignedChar unsigned char unsafe.CUnsignedChar[^1] short unsafe.CShort unsigned short unsafe.CUnsignedShort[^2] int unsafe.CInt long int unsafe.CLongInt unsigned int unsafe.CUnsignedInt[^3] unsigned long int unsafe.CUnsignedLongInt[^4] long unsafe.CLong unsigned long unsafe.CUnsignedLong[^5] long long unsafe.CLongLong unsigned long long unsafe.CUnsignedLongLong[^6] size_t unsafe.CSize ssize_t unsafe.CSSize ptrdiff_t unsafe.CPtrDiff[^7] wchar_t unsafe.CWideChar char16_t unsafe.CChar16 char32_t unsafe.CChar32 float unsafe.CFloat double unsafe.CDouble void* unsafe.CVoidPtr[^8] int* unsafe.Ptr[unsafe.CInt][^9] char* unsafe.CString[^10][^11] int (*)(int) unsafe.CFuncPtr1[unsafe.CInt, unsafe.CInt][^12][^13] struct { int x, y; }* unsafe.Ptr[unsafe.CStruct2[unsafe.CInt, unsafe.CInt]][^14][^15] struct { int x, y; } Not supported

Linking with native libraries

C compilers typically require to pass an additional -l mylib flag to dynamically link with a library. In Scala Native, one can annotate libraries to link with using the @link annotation.

import scala.scalanative.unsafe._

@link("mylib")
@extern
object mylib {
  def f(): Unit = extern
}

Whenever any of the members of mylib object are reachable, the Scala Native linker will automatically link with the corresponding native library.

As in C, library names are specified without the lib prefix. For example, the library libuv corresponds to @link("uv") in Scala Native.

It is possible to rename functions using the @name annotation. Its use is recommended to enforce the Scala naming conventions in bindings:

import scala.scalanative.unsafe._

@link("uv")
@extern
object uv {
  @name("uv_uptime")
  def uptime(result: Ptr[CDouble]): Int = extern
}

If a library has multiple components, you could split the bindings into separate objects as it is permitted to use the same @link annotation more than once.

Variadic functions

Scala Native supports native interoperability with C’s variadic argument list type (i.e. va_list), and partially for ... varargs. For example vprintf and printf defined in C as:

int vprintf(const char * format, va_list arg);
int printf(const char * format, ... );  

can be declared in Scala as:

import scala.scalanative.unsafe._

@extern
object mystdio {
  def vprintf(format: CString, args: CVarArgList): CInt = extern
  def printf(format: CString, args: Any*): CInt = extern
}

The limitation of ... interop requires that it’s arguments needs to passed directly to variadic arguments function or arguments need to be inlined. This is required to obtain enough information on how arguments show be passed in regards to C ABI. Passing a sequence to extern method variadic arguments is not allowed and would result in compilation failure.

For va_list interop, one can wrap a function in a nicer API like:

import scala.scalanative.unsafe._

def myprintf(format: CString, args: CVarArg*): CInt =
  Zone { 
    mystdio.vprintf(format, toCVarArgList(args.toSeq))
  }

See Memory management for a guide of using unsafe.Zone And then call it just like a regular Scala function:

myprintf(c"2 + 3 = %d, 4 + 5 = %d", 2 + 3, 4 + 5)
printf(c"2 + 3 = %d, 4 + 5 = %d", 2 + 3, 4 + 5)

Exported methods

When linking Scala Native as library, you can mark functions that should visible in created library with @exported(name: String) annotation. In case if you omit or use null as the argument for name extern function name match the name of the method. Currently, only static object methods can be exported. To export accessors of field or variable in static object use @exportAccessors(getterName: String, setterName: String). If you omit the explicit names in the annotation constructor, Scala Native would create exported methods with set_ and get_ prefixes and name of field.

int ScalaNativeInit(void); function is special exported function that needs to be called before invoking any code defined in Scala Native. It returns 0 on successful initialization and non-zero value in the otherwise. For dynamic libraries a constructor would be generated to invoke ScalaNativeInit[ function automatically upon loading library or startup of the program. If for some reason you need to disable automatic initialization of Scala Native upon loading dynamic library and invoke it manually in user code set `SCALANATIVE_NO_DYLIB_CTOR]{.title-ref} environment variable. You can also disable generation of library constructors by defining -DSCALANATIVE_NO_DYLIB_CTOR in NativeConfig::compileOptions of your build.

import scala.scalanative.unsafe._

object myLib{
  @exportAccessors("mylib_current_count", "mylib_set_counter")
  var counter: Int = 0

  @exportAccessors("error_message")
  val ErrorMessage: CString = c"Something bad just happend!"

  @exported
  def addLongs(l: Long, r: Long): Long = l + r

  @exported("mylib_addInts")
  def addInts(l: Int, r: Int): Int = l + r
}
// libmylib.h
int ScalaNativeInit(void);
long addLongs(long, long);
int mylib_addInts(int, int);
int mylib_current_count();
void mylib_set_counter(int);

// test.c
#include "libmylib.h"
#include <assert.h>
#include <stdio.h>

int main(int argc, char** argv){
  // This function needs to be called before invoking any methods defined in Scala Native.
  // Might be called automatically unless SCALANATIVE_NO_DYLIB_CTOR env variable is set.
  assert(ScalaNativeInit() == 0);
  addLongs(0L, 4L);
  mylib_addInts(4, 0);
  printf("Current count %d\n", mylib_current_count());
  mylib_set_counter(42);
  // ...
}

Pointer types

Scala Native provides a built-in equivalent of C’s pointers via unsafe.Ptr[T] data type. Under the hood pointers are implemented using unmanaged machine pointers.

Operations on pointers are closely related to their C counterparts and are compiled into equivalent machine code:

Operation C syntax Scala Syntax


Load value *ptr !ptr Store value *ptr = value !ptr = value Pointer to index ptr + i, &ptr[i] ptr + i Elements between ptr1 - ptr2 ptr1 - ptr2 Load at index ptr[i] ptr(i) Store at index ptr[i] = value ptr(i) = value Pointer to field &ptr->name ptr.atN Load a field ptr->name ptr._N Store a field ptr->name = value ptr._N = value

Where N is the index of the field name in the struct. See Memory layout types for details.

Function pointers

It is possible to use external functions that take function pointers. For example given the following signature in C:

void test(void (* f)(char *));

One can declare it as follows in Scala Native:

def test(f: unsafe.CFuncPtr1[CString, Unit]): Unit = unsafe.extern

CFuncPtrN types are final classes containing pointer to underlying C function pointer. They automatically handle boxing call arguments and unboxing result. You can create them from C pointer using CFuncPtr helper methods:

def fnDef(str: CString): CInt = ???

val anyPtr: CVoidPtr = CFuncPtr.toPtr {
  CFuncPtr1.fromScalaFunction(fnDef)
}

type StringLengthFn = CFuncPtr1[CString, CInt]
val func: StringLengthFn = CFuncPtr.fromPtr[StringLengthFn](anyPtr)
func(c"hello")

It’s also possible to create CFuncPtrN from Scala FunctionN. You can do this by using implicit method conversion method from the corresponding companion object.

import scalanative.unsafe.CFuncPtr0
def myFunc(): Unit = println("hi there!")

val myFuncPtr: CFuncPtr0[Unit] = CFuncPtr0.fromScalaFunction(myFunc)
val myImplFn: CFuncPtr0[Unit] = myFunc _
val myLambdaFuncPtr: CFuncPtr0[Unit] = () => println("hello!")

On Scala 2.12 or newer, the Scala language automatically converts from closures to SAM types:

val myfuncptr: unsafe.CFuncPtr0[Unit] = () => println("hi there!")

Memory management

Unlike standard Scala objects that are managed automatically by the underlying runtime system, one has to be extra careful when working with unmanaged memory.

  1. Zone allocation. (since 0.3)

    Zones (also known as memory regions/contexts) are a technique for semi-automatic memory management. Using them one can bind allocations to a temporary scope in the program and the zone allocator will automatically clean them up for you as soon as execution goes out of it:

    import scala.scalanative.unsafe._
    
    // For Scala 3
    Zone {
      val buffer = alloc[Byte](n)
    }
    // For Scala 2, works, but is not idiomatic on Scala 3
    Zone.acquire { implicit z =>
      val buffer = alloc[Byte](n)
    }
    

    alloc requests memory sufficient to contain n values of a given type. If number of elements is not specified, it defaults to a single element. Memory is zeroed out by default.

    Zone allocation is the preferred way to allocate temporary unmanaged memory. It’s idiomatic to use implicit zone parameters to abstract over code that has to zone allocate.

    One typical example of this are C strings that are created from Scala strings using unsafe.toCString. The conversion takes implicit zone parameter and allocates the result in that zone.

    When using zone allocated memory one has to be careful not to capture this memory beyond the lifetime of the zone. Dereferencing zone-allocated memory after the end of the zone is undefined behavior.

  2. Stack allocation.

    Scala Native provides a built-in way to perform stack allocations of using unsafe.stackalloc function:

    val buffer = unsafe.stackalloc[Byte](256)
    

    This code will allocate 256 bytes that are going to be available until the enclosing method returns. Number of elements to be allocated is optional and defaults to 1 otherwise. Memory is zeroed out by default.

    When using stack allocated memory one has to be careful not to capture this memory beyond the lifetime of the method. Dereferencing stack allocated memory after the method’s execution has completed is undefined behavior.

  3. Manual heap allocation.

    Scala Native’s library contains a bindings for a subset of the standard libc functionality. This includes the trio of malloc, realloc and free functions that are defined in unsafe.stdlib extern object.

    Calling those will let you allocate memory using system’s standard dynamic memory allocator. Every single manual allocation must also be freed manually as soon as it’s not needed any longer.

    Apart from the standard system allocator one might also bind to plethora of 3-rd party allocators such as jemalloc to serve the same purpose.

Undefined behavior

Similarly to their C counter-parts, behavior of operations that access memory is subject to undefined behaviour for following conditions:

  1. Dereferencing null.

  2. Out-of-bounds memory access.

  3. Use-after-free.

  4. Use-after-return.

  5. Double-free, invalid free.

Memory layout types

Memory layout types are auxiliary types that let one specify memory layout of unmanaged memory. They are meant to be used purely in combination with native pointers and do not have a corresponding first-class values backing them.

  • unsafe.Ptr[unsafe.CStructN[T1, ..., TN]]

    Pointer to a C struct with up to 22 fields. Type parameters are the types of corresponding fields. One may access fields of the struct using _N helper methods on a pointer value:

    val ptr = unsafe.stackalloc[unsafe.CStruct2[Int, Int]]()
    ptr._1 = 10
    ptr._2 = 20
    println(s"first ${ptr._1}, second ${ptr._2}")
    

    Here _N is an accessor for the field number N.

  • unsafe.Ptr[unsafe.CArray[T, N]]

    Pointer to a C array with statically-known length N. Length is encoded as a type-level natural number. Natural numbers are types that are composed of base naturals Nat._0, ... Nat._9 and an additional Nat.DigitN constructors, where N refers to number of digits in the given number. So for example number 1024 is going to be encoded as following:

    import scalanative.unsafe._, Nat._
    
    type _1024 = Digit4[_1, _0, _2, _4]
    

    Once you have a natural for the length, it can be used as an array length:

    val arrptr = unsafe.stackalloc[CArray[Byte, _1024]]()
    

    You can find an address of n-th array element via arrptr.at(n).

Byte strings

Scala Native supports byte strings via c"..." string interpolator that gets compiled down to pointers to statically-allocated zero-terminated strings (similarly to C):

import scalanative.unsafe._
import scalanative.libc._

// CString is an alias for Ptr[CChar]
val msg: CString = c"Hello, world!"
stdio.printf(msg)

It does not allow any octal values or escape characters not supported by Scala compiler, like \a or \?, but also unicode escapes. It is possible to use C-style hex values up to value 0xFF, eg. c"Hello \x61\x62\x63"

Additionally, we also expose two helper functions unsafe.fromCString and unsafe.toCString to convert between C-style CString (sequence of Bytes, usually interpreted as UTF-8 or ASCII) and Java-style String (sequence of 2-byte Chars usually interpreted as UTF-16).

It’s worth to remember that unsafe.toCString and c\"\...\" interpreter cannot be used interchangeably as they handle literals differently. Helper methods unsafe.fromCString` andunsafe.toCStringare charset aware. They will always assume `String` is UTF-16, and take a `Charset` parameter to know what encoding to assume for the byte string (`CString`) - if not present it is UTF-8.  If passed a null as an argument, they will return a null of the appropriate type instead of throwing a NullPointerException.   Platform-specific types -----------------------  Scala Native defines the typeSizeand its unsigned counterpart,USize. A size corresponds toInton 32-bit architectures and toLongon 64-bit ones.  Size and alignment of types ---------------------------  In order to statically determine the size of a type, you can use thesizeoffunction which is Scala Native's counterpart of the eponymous C operator. It returns the size in bytes:  .. code-block:: scala      println(unsafe.sizeof[Byte])    // 1     println(unsafe.sizeof[CBool])   // 1     println(unsafe.sizeof[CShort])  // 2     println(unsafe.sizeof[CInt])    // 4     println(unsafe.sizeof[CLong])   // 8  It can also be used to obtain the size of a structure:  .. code-block:: scala      type TwoBytes = unsafe.CStruct2[Byte, Byte]     println(unsafe.sizeof[TwoBytes])  // 2  Additionally, you can also usealignmentofto find the alignment of a given type:  .. code-block:: scala      println(unsafe.alignmentof[Int])                         // 4     println(unsafe.alignmentof[unsafe.CStruct2[Byte, Long]]) // 8  Unsigned integer types ----------------------  Scala Native provides support for four unsigned integer types:  1.unsigned.UByte2.unsigned.UShort3.unsigned.UInt4.unsigned.ULong5.unsigned.USizeThey share the same primitive operations as signed integer types. Primitive operation between two integer values are supported only if they have the same signedness (they must both signed or both unsigned.)  Conversions between signed and unsigned integers must be done explicitly usingbyteValue.toUByte,shortValue.toUShort,intValue.toUInt,longValue.toULong,sizeValue.toUSizeand converselyunsignedByteValue.toByte,unsignedShortValue.toShort,unsignedIntValue.toInt,unsignedLongValue.toLong,unsignedSizeValue.toSize.  Universal equality is supported between signed and unsigned integers, for example-1.toUByte == 255or65535 == -1.toUShortwould yieldtrue,  However, similar to signed integers on JVM, class equality between different (boxed) integer types is not supported. Usage of `-1.toUByte.equals(255)` would returnfalse, as we're comparing different boxed types (scala.scalanative.unsigned.UBytewithjava.lang.Integer``)

Continue to native.

[^1]: See Unsigned integer types.

[^2]: See Unsigned integer types.

[^3]: See Unsigned integer types.

[^4]: See Unsigned integer types.

[^5]: See Unsigned integer types.

[^6]: See Unsigned integer types.

[^7]: See Pointer types.

[^8]: See Pointer types.

[^9]: See Pointer types.

[^10]: See Pointer types.

[^11]: See Byte strings.

[^12]: See Pointer types.

[^13]: See Function pointers.

[^14]: See Pointer types.

[^15]: See Memory layout types.