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._

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 [1]
int unsafe.CInt
long int unsafe.CLongInt
unsigned int unsafe.CUnsignedInt [1]
unsigned long int unsafe.CUnsignedLongInt [1]
long unsafe.CLong
unsigned long unsafe.CUnsignedLong [1]
long long unsafe.CLongLong
unsigned long long unsafe.CUnsignedLongLong [1]
size_t unsafe.CSize
ptrdiff_t unsafe.CPtrDiff [2]
wchar_t unsafe.CWideChar
char16_t unsafe.CChar16
char32_t unsafe.CChar32
float unsafe.CFloat
double unsafe.CDouble
void* unsafe.Ptr[Byte] [2]
int* unsafe.Ptr[unsafe.CInt] [2]
char* unsafe.CString [2] [3]
int (*)(int) unsafe.CFuncPtr1[unsafe.CInt, unsafe.CInt] [2] [4]
struct { int x, y; }* unsafe.Ptr[unsafe.CStruct2[unsafe.CInt, unsafe.CInt]] [2] [5]
struct { int x, y; } Not supported
[1](1, 2, 3, 4, 5, 6) See Unsigned integer types.
[2](1, 2, 3, 4, 5, 6) See Pointer types.
[3]See Byte strings.
[4]See Function pointers.
[5]See Memory layout types.

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._

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._

object uv {
  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), but not ... varargs. For example vprintf can be declared as:

import scala.scalanative.unsafe._

object mystdio {
  def vprintf(format: CString, args: CVarArgList): CInt = extern

One can wrap a function in a nicer API like:

import scala.scalanative.unsafe._

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

And then call it just like a regular Scala function:

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

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: Ptr[Byte] = CFuncPtr.toPtr {

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

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._
    Zone { 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 not 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

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!"

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` and ``unsafe.toCString are 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 type Word and its unsigned counterpart, UWord. A word corresponds to Int on 32-bit architectures and to Long on 64-bit ones.

Size and alignment of types

In order to statically determine the size of a type, you can use the sizeof function which is Scala Native’s counterpart of the eponymous C operator. It returns the size in bytes:

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:

type TwoBytes = unsafe.CStruct2[Byte, Byte]
println(unsafe.sizeof[TwoBytes])  // 2

Aditionally you can also use alignmentof to find alignment of a given type:

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.UByte
  2. unsigned.UShort
  3. unsigned.UInt
  4. unsigned.ULong

They 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 using byteValue.toUByte, shortValue.toUShort, intValue.toUInt, longValue.toULong and conversely unsignedByteValue.toByte, unsignedShortValue.toShort, unsignedIntValue.toInt, unsignedLongValue.toLong.

Continue to Libraries.