Calling C and Fortran Code

Though most code can be written in Julia, there are many high-quality, mature libraries for numerical computing already written in C and Fortran. To allow easy use of this existing code, Julia makes it simple and efficient to call C and Fortran functions. Julia has a "no boilerplate" philosophy: functions can be called directly from Julia without any "glue" code, code generation, or compilation – even from the interactive prompt. This is accomplished just by making an appropriate call with ccall syntax, which looks like an ordinary function call.

The code to be called must be available as a shared library. Most C and Fortran libraries ship compiled as shared libraries already, but if you are compiling the code yourself using GCC (or Clang), you will need to use the -shared and -fPIC options. The machine instructions generated by Julia's JIT are the same as a native C call would be, so the resulting overhead is the same as calling a library function from C code. ^[1]

Shared libraries and functions are referenced by a tuple of the form (:function, "library") or ("function", "library") where function is the C-exported function name, and library refers to the shared library name. Shared libraries available in the (platform-specific) load path will be resolved by name. The full path to the library may also be specified.

A function name may be used alone in place of the tuple (just :function or "function"). In this case the name is resolved within the current process. This form can be used to call C library functions, functions in the Julia runtime, or functions in an application linked to Julia.

By default, Fortran compilers generate mangled names (for example, converting function names to lowercase or uppercase, often appending an underscore), and so to call a Fortran function via ccall you must pass the mangled identifier corresponding to the rule followed by your Fortran compiler. Also, when calling a Fortran function, all inputs must be passed as pointers to allocated values on the heap or stack. This applies not only to arrays and other mutable objects which are normally heap-allocated, but also to scalar values such as integers and floats which are normally stack-allocated and commonly passed in registers when using C or Julia calling conventions.

Finally, you can use ccall to actually generate a call to the library function. The arguments to ccall are:

1. A (:function, "library") pair (most common),

   OR

   a :function name symbol or "function" name string (for symbols in the current process or libc),

   OR

   a function pointer (for example, from dlsym).

2. The function's return type

3. A tuple of input types, corresponding to the function signature

4. The actual argument values to be passed to the function, if any; each is a separate parameter.

As a complete but simple example, the following calls the clock function from the standard C library on most Unix-derived systems:

julia> t = ccall(:clock, Int32, ())
2292761

julia> t
2292761

julia> typeof(t)
Int32

clock takes no arguments and returns an Int32. One common mistake is forgetting that a 1-tuple of argument types must be written with a trailing comma. For example, to call the getenv function to get a pointer to the value of an environment variable, one makes a call like this:

julia> path = ccall(:getenv, Cstring, (Cstring,), "SHELL")
Cstring(@0x00007fff5fbffc45)

julia> unsafe_string(path)
"/bin/bash"

Note that the argument type tuple must be written as (Cstring,), not (Cstring). This is because (Cstring) is just the expression Cstring surrounded by parentheses, rather than a 1-tuple containing Cstring:

julia> (Cstring)
Cstring

julia> (Cstring,)
(Cstring,)

In practice, especially when providing reusable functionality, one generally wraps ccall uses in Julia functions that set up arguments and then check for errors in whatever manner the C or Fortran function specifies. And if an error occurs it is thrown as a normal Julia exception. This is especially important since C and Fortran APIs are notoriously inconsistent about how they indicate error conditions. For example, the getenv C library function is wrapped in the following Julia function, which is a simplified version of the actual definition from env.jl:

function getenv(var::AbstractString)
    val = ccall(:getenv, Cstring, (Cstring,), var)
    if val == C_NULL
        error("getenv: undefined variable: ", var)
    end
    return unsafe_string(val)
end

The C getenv function indicates an error by returning NULL, but other standard C functions indicate errors in various different ways, including by returning -1, 0, 1 and other special values. This wrapper throws an exception clearly indicating the problem if the caller tries to get a non-existent environment variable:

julia> getenv("SHELL")
"/bin/bash"

julia> getenv("FOOBAR")
getenv: undefined variable: FOOBAR

Here is a slightly more complex example that discovers the local machine's hostname. In this example, the networking library code is assumed to be in a shared library named "libc". In practice, this function is usually part of the C standard library, and so the "libc" portion should be omitted, but we wish to show here the usage of this syntax.

function gethostname()
    hostname = Vector{UInt8}(undef, 256) # MAXHOSTNAMELEN
    err = ccall((:gethostname, "libc"), Int32,
                (Ptr{UInt8}, Csize_t),
                hostname, sizeof(hostname))
    Base.systemerror("gethostname", err != 0)
    hostname[end] = 0 # ensure null-termination
    return GC.@preserve hostname unsafe_string(pointer(hostname))
end

This example first allocates an array of bytes. It then calls the C library function gethostname to populate the array with the hostname. Finally, it takes a pointer to the hostname buffer, and converts the pointer to a Julia string, assuming that it is a NUL-terminated C string.

It is common for C libraries to use this pattern of requiring the caller to allocate memory to be passed to the callee and populated. Allocation of memory from Julia like this is generally accomplished by creating an uninitialized array and passing a pointer to its data to the C function. This is why we don't use the Cstring type here: as the array is uninitialized, it could contain NUL bytes. Converting to a Cstring as part of the ccall checks for contained NUL bytes and could therefore throw a conversion error.

Dereferencing pointer(hostname) with unsafe_string is an unsafe operation as it requires access to the memory allocated for hostname that may have been in the meanwhile garbage collected. The macro GC.@preserve prevents this from happening and therefore accessing an invalid memory location.

Creating C-Compatible Julia Function Pointers

It is possible to pass Julia functions to native C functions that accept function pointer arguments. For example, to match C prototypes of the form:

typedef returntype (*functiontype)(argumenttype, ...)

The macro @cfunction generates the C-compatible function pointer for a call to a Julia function. The arguments to @cfunction are:

1. A Julia function
2. The function's return type
3. A tuple of input types, corresponding to the function signature

A classic example is the standard C library qsort function, declared as:

void qsort(void *base, size_t nmemb, size_t size,
           int (*compare)(const void*, const void*));

The base argument is a pointer to an array of length nmemb, with elements of size bytes each. compare is a callback function which takes pointers to two elements a and b and returns an integer less/greater than zero if a should appear before/after b (or zero if any order is permitted).

Now, suppose that we have a 1-d array A of values in Julia that we want to sort using the qsort function (rather than Julia's built-in sort function). Before we consider calling qsort and passing arguments, we need to write a comparison function:

julia> function mycompare(a, b)::Cint
           return (a < b) ? -1 : ((a > b) ? +1 : 0)
       end
mycompare (generic function with 1 method)

qsort expects a comparison function that return a C int, so we annotate the return type to be Cint.

In order to pass this function to C, we obtain its address using the macro @cfunction:

julia> mycompare_c = @cfunction(mycompare, Cint, (Ref{Cdouble}, Ref{Cdouble}));

@cfunction requires three arguments: the Julia function (mycompare), the return type (Cint), and a literal tuple of the input argument types, in this case to sort an array of Cdouble (Float64) elements.

The final call to qsort looks like this:

julia> A = [1.3, -2.7, 4.4, 3.1]
4-element Vector{Float64}:
  1.3
 -2.7
  4.4
  3.1

julia> ccall(:qsort, Cvoid, (Ptr{Cdouble}, Csize_t, Csize_t, Ptr{Cvoid}),
             A, length(A), sizeof(eltype(A)), mycompare_c)

julia> A
4-element Vector{Float64}:
 -2.7
  1.3
  3.1
  4.4

As the example shows, the original Julia array A has now been sorted: [-2.7, 1.3, 3.1, 4.4]. Note that Julia takes care of converting the array to a Ptr{Cdouble}), computing the size of the element type in bytes, and so on.

For fun, try inserting a println("mycompare($a, $b)") line into mycompare, which will allow you to see the comparisons that qsort is performing (and to verify that it is really calling the Julia function that you passed to it).

Mapping C Types to Julia

It is critical to exactly match the declared C type with its declaration in Julia. Inconsistencies can cause code that works correctly on one system to fail or produce indeterminate results on a different system.

Note that no C header files are used anywhere in the process of calling C functions: you are responsible for making sure that your Julia types and call signatures accurately reflect those in the C header file.^[2]

Automatic Type Conversion

Julia automatically inserts calls to the Base.cconvert function to convert each argument to the specified type. For example, the following call:

ccall((:foo, "libfoo"), Cvoid, (Int32, Float64), x, y)

will behave as if it were written like this:

ccall((:foo, "libfoo"), Cvoid, (Int32, Float64),
      Base.unsafe_convert(Int32, Base.cconvert(Int32, x)),
      Base.unsafe_convert(Float64, Base.cconvert(Float64, y)))

Base.cconvert normally just calls convert, but can be defined to return an arbitrary new object more appropriate for passing to C. This should be used to perform all allocations of memory that will be accessed by the C code. For example, this is used to convert an Array of objects (e.g. strings) to an array of pointers.

Base.unsafe_convert handles conversion to Ptr types. It is considered unsafe because converting an object to a native pointer can hide the object from the garbage collector, causing it to be freed prematurely.

Type Correspondences

First, let's review some relevant Julia type terminology:

Bits Types

There are several special types to be aware of, as no other type can be defined to behave the same:

• Float32

  Exactly corresponds to the float type in C (or REAL*4 in Fortran).

• Float64

  Exactly corresponds to the double type in C (or REAL*8 in Fortran).

• ComplexF32

  Exactly corresponds to the complex float type in C (or COMPLEX*8 in Fortran).

• ComplexF64

  Exactly corresponds to the complex double type in C (or COMPLEX*16 in Fortran).

• Signed

  Exactly corresponds to the signed type annotation in C (or any INTEGER type in Fortran). Any Julia type that is not a subtype of Signed is assumed to be unsigned.

• Ref{T}

  Behaves like a Ptr{T} that can manage its memory via the Julia GC.

• Array{T,N}

  When an array is passed to C as a Ptr{T} argument, it is not reinterpret-cast: Julia requires that the element type of the array matches T, and the address of the first element is passed.

  Therefore, if an Array contains data in the wrong format, it will have to be explicitly converted using a call such as trunc(Int32, a).

  To pass an array A as a pointer of a different type without converting the data beforehand (for example, to pass a Float64 array to a function that operates on uninterpreted bytes), you can declare the argument as Ptr{Cvoid}.

  If an array of eltype Ptr{T} is passed as a Ptr{Ptr{T}} argument, Base.cconvert will attempt to first make a null-terminated copy of the array with each element replaced by its Base.cconvert version. This allows, for example, passing an argv pointer array of type Vector{String} to an argument of type Ptr{Ptr{Cchar}}.

On all systems we currently support, basic C/C++ value types may be translated to Julia types as follows. Every C type also has a corresponding Julia type with the same name, prefixed by C. This can help when writing portable code (and remembering that an int in C is not the same as an Int in Julia).

System Independent Types

The Cstring type is essentially a synonym for Ptr{UInt8}, except the conversion to Cstring throws an error if the Julia string contains any embedded NUL characters (which would cause the string to be silently truncated if the C routine treats NUL as the terminator). If you are passing a char* to a C routine that does not assume NUL termination (e.g. because you pass an explicit string length), or if you know for certain that your Julia string does not contain NUL and want to skip the check, you can use Ptr{UInt8} as the argument type. Cstring can also be used as the ccall return type, but in that case it obviously does not introduce any extra checks and is only meant to improve readability of the call.

System Dependent Types

int main(int argc, char **argv);

argv = [ "a.out", "arg1", "arg2" ]
ccall(:main, Int32, (Int32, Ptr{Ptr{UInt8}}), length(argv), argv)

subroutine test(str1, str2)
character(len=*) :: str1,str2

str1 = "foo"
str2 = "bar"
ccall(:test, Cvoid, (Ptr{UInt8}, Ptr{UInt8}, Csize_t, Csize_t),
                    str1, str2, sizeof(str1), sizeof(str2))

Struct Type Correspondences

Composite types such as struct in C or TYPE in Fortran90 (or STRUCTURE / RECORD in some variants of F77), can be mirrored in Julia by creating a struct definition with the same field layout.

When used recursively, isbits types are stored inline. All other types are stored as a pointer to the data. When mirroring a struct used by-value inside another struct in C, it is imperative that you do not attempt to manually copy the fields over, as this will not preserve the correct field alignment. Instead, declare an isbits struct type and use that instead. Unnamed structs are not possible in the translation to Julia.

Packed structs and union declarations are not supported by Julia.

You can get an approximation of a union if you know, a priori, the field that will have the greatest size (potentially including padding). When translating your fields to Julia, declare the Julia field to be only of that type.

Arrays of parameters can be expressed with NTuple. For example, the struct in C notation written as

struct B {
    int A[3];
};

b_a_2 = B.A[2];

can be written in Julia as

struct B
    A::NTuple{3, Cint}
end

b_a_2 = B.A[3]  # note the difference in indexing (1-based in Julia, 0-based in C)

Arrays of unknown size (C99-compliant variable length structs specified by [] or [0]) are not directly supported. Often the best way to deal with these is to deal with the byte offsets directly. For example, if a C library declared a proper string type and returned a pointer to it:

struct String {
    int strlen;
    char data[];
};

In Julia, we can access the parts independently to make a copy of that string:

str = from_c::Ptr{Cvoid}
len = unsafe_load(Ptr{Cint}(str))
unsafe_string(str + Core.sizeof(Cint), len)

Type Parameters

The type arguments to ccall and @cfunction are evaluated statically, when the method containing the usage is defined. They therefore must take the form of a literal tuple, not a variable, and cannot reference local variables.

This may sound like a strange restriction, but remember that since C is not a dynamic language like Julia, its functions can only accept argument types with a statically-known, fixed signature.

However, while the type layout must be known statically to compute the intended C ABI, the static parameters of the function are considered to be part of this static environment. The static parameters of the function may be used as type parameters in the call signature, as long as they don't affect the layout of the type. For example, f(x::T) where {T} = ccall(:valid, Ptr{T}, (Ptr{T},), x) is valid, since Ptr is always a word-size primitive type. But, g(x::T) where {T} = ccall(:notvalid, T, (T,), x) is not valid, since the type layout of T is not known statically.

SIMD Values

Note: This feature is currently implemented on 64-bit x86 and AArch64 platforms only.

If a C/C++ routine has an argument or return value that is a native SIMD type, the corresponding Julia type is a homogeneous tuple of VecElement that naturally maps to the SIMD type. Specifically:

• The tuple must be the same size as the SIMD type. For example, a tuple representing an __m128 on x86 must have a size of 16 bytes.
• The element type of the tuple must be an instance of VecElement{T} where T is a primitive type that is 1, 2, 4 or 8 bytes.

For instance, consider this C routine that uses AVX intrinsics:

#include <immintrin.h>

__m256 dist( __m256 a, __m256 b ) {
    return _mm256_sqrt_ps(_mm256_add_ps(_mm256_mul_ps(a, a),
                                        _mm256_mul_ps(b, b)));
}

The following Julia code calls dist using ccall:

const m256 = NTuple{8, VecElement{Float32}}

a = m256(ntuple(i -> VecElement(sin(Float32(i))), 8))
b = m256(ntuple(i -> VecElement(cos(Float32(i))), 8))

function call_dist(a::m256, b::m256)
    ccall((:dist, "libdist"), m256, (m256, m256), a, b)
end

println(call_dist(a,b))

The host machine must have the requisite SIMD registers. For example, the code above will not work on hosts without AVX support.

Memory Ownership

malloc/free

Memory allocation and deallocation of such objects must be handled by calls to the appropriate cleanup routines in the libraries being used, just like in any C program. Do not try to free an object received from a C library with Libc.free in Julia, as this may result in the free function being called via the wrong library and cause the process to abort. The reverse (passing an object allocated in Julia to be freed by an external library) is equally invalid.

When to use T, Ptr{T} and Ref{T}

In Julia code wrapping calls to external C routines, ordinary (non-pointer) data should be declared to be of type T inside the ccall, as they are passed by value. For C code accepting pointers, Ref{T} should generally be used for the types of input arguments, allowing the use of pointers to memory managed by either Julia or C through the implicit call to Base.cconvert. In contrast, pointers returned by the C function called should be declared to be of output type Ptr{T}, reflecting that the memory pointed to is managed by C only. Pointers contained in C structs should be represented as fields of type Ptr{T} within the corresponding Julia struct types designed to mimic the internal structure of corresponding C structs.

In Julia code wrapping calls to external Fortran routines, all input arguments should be declared as of type Ref{T}, as Fortran passes all variables by pointers to memory locations. The return type should either be Cvoid for Fortran subroutines, or a T for Fortran functions returning the type T.

Mapping C Functions to Julia

ccall / @cfunction argument translation guide

For translating a C argument list to Julia:

• T, where T is one of the primitive types: char, int, long, short, float, double, complex, enum or any of their typedef equivalents

  • T, where T is an equivalent Julia Bits Type (per the table above)
  • if T is an enum, the argument type should be equivalent to Cint or Cuint
  • argument value will be copied (passed by value)

• struct T (including typedef to a struct)

  • T, where T is a Julia leaf type
  • argument value will be copied (passed by value)

• void*

  • depends on how this parameter is used, first translate this to the intended pointer type, then determine the Julia equivalent using the remaining rules in this list
  • this argument may be declared as Ptr{Cvoid}, if it really is just an unknown pointer

• jl_value_t*

  • Any
  • argument value must be a valid Julia object

• jl_value_t* const*

  • Ref{Any}
  • argument list must be a valid Julia object (or C_NULL)
  • cannot be used for an output parameter, unless the user is able to separately arrange for the object to be GC-preserved

• T*

  • Ref{T}, where T is the Julia type corresponding to T
  • argument value will be copied if it is an inlinealloc type (which includes isbits otherwise, the value must be a valid Julia object

• T (*)(...) (e.g. a pointer to a function)

  • Ptr{Cvoid} (you may need to use @cfunction explicitly to create this pointer)

• ... (e.g. a vararg)

  • [for ccall]: T..., where T is the single Julia type of all remaining arguments
  • [for @ccall]: ; va_arg1::T, va_arg2::S, etc, where T and S are the Julia type (i.e. separate the regular arguments from varargs with a ;)
  • currently unsupported by @cfunction

• va_arg

  • not supported by ccall or @cfunction

ccall / @cfunction return type translation guide

For translating a C return type to Julia:

• void

  • Cvoid (this will return the singleton instance nothing::Cvoid)

• T, where T is one of the primitive types: char, int, long, short, float, double, complex, enum or any of their typedef equivalents

  • T, where T is an equivalent Julia Bits Type (per the table above)
  • if T is an enum, the argument type should be equivalent to Cint or Cuint
  • argument value will be copied (returned by-value)

• struct T (including typedef to a struct)

  • T, where T is a Julia Leaf Type
  • argument value will be copied (returned by-value)

• void*

  • depends on how this parameter is used, first translate this to the intended pointer type, then determine the Julia equivalent using the remaining rules in this list
  • this argument may be declared as Ptr{Cvoid}, if it really is just an unknown pointer

• jl_value_t*

  • Any
  • argument value must be a valid Julia object

• jl_value_t**

  • Ptr{Any} (Ref{Any} is invalid as a return type)

• T*

  • If the memory is already owned by Julia, or is an isbits type, and is known to be non-null:

    • Ref{T}, where T is the Julia type corresponding to T
    • a return type of Ref{Any} is invalid, it should either be Any (corresponding to jl_value_t*) or Ptr{Any} (corresponding to jl_value_t**)
    • C MUST NOT modify the memory returned via Ref{T} if T is an isbits type

  • If the memory is owned by C:

    • Ptr{T}, where T is the Julia type corresponding to T

• T (*)(...) (e.g. a pointer to a function)

  • Ptr{Cvoid} to call this directly from Julia you will need to pass this as the first argument to ccall. See Indirect Calls.

Passing Pointers for Modifying Inputs

Because C doesn't support multiple return values, often C functions will take pointers to data that the function will modify. To accomplish this within a ccall, you need to first encapsulate the value inside a Ref{T} of the appropriate type. When you pass this Ref object as an argument, Julia will automatically pass a C pointer to the encapsulated data:

width = Ref{Cint}(0)
range = Ref{Cfloat}(0)
ccall(:foo, Cvoid, (Ref{Cint}, Ref{Cfloat}), width, range)

Upon return, the contents of width and range can be retrieved (if they were changed by foo) by width[] and range[]; that is, they act like zero-dimensional arrays.

C Wrapper Examples

Let's start with a simple example of a C wrapper that returns a Ptr type:

mutable struct gsl_permutation
end

# The corresponding C signature is
#     gsl_permutation * gsl_permutation_alloc (size_t n);
function permutation_alloc(n::Integer)
    output_ptr = ccall(
        (:gsl_permutation_alloc, :libgsl), # name of C function and library
        Ptr{gsl_permutation},              # output type
        (Csize_t,),                        # tuple of input types
        n                                  # name of Julia variable to pass in
    )
    if output_ptr == C_NULL # Could not allocate memory
        throw(OutOfMemoryError())
    end
    return output_ptr
end

The GNU Scientific Library (here assumed to be accessible through :libgsl) defines an opaque pointer, gsl_permutation *, as the return type of the C function gsl_permutation_alloc. As user code never has to look inside the gsl_permutation struct, the corresponding Julia wrapper simply needs a new type declaration, gsl_permutation, that has no internal fields and whose sole purpose is to be placed in the type parameter of a Ptr type. The return type of the ccall is declared as Ptr{gsl_permutation}, since the memory allocated and pointed to by output_ptr is controlled by C.

The input n is passed by value, and so the function's input signature is simply declared as (Csize_t,) without any Ref or Ptr necessary. (If the wrapper was calling a Fortran function instead, the corresponding function input signature would instead be (Ref{Csize_t},), since Fortran variables are passed by pointers.) Furthermore, n can be any type that is convertible to a Csize_t integer; the ccall implicitly calls Base.cconvert(Csize_t, n).

Here is a second example wrapping the corresponding destructor:

# The corresponding C signature is
#     void gsl_permutation_free (gsl_permutation * p);
function permutation_free(p::Ref{gsl_permutation})
    ccall(
        (:gsl_permutation_free, :libgsl), # name of C function and library
        Cvoid,                             # output type
        (Ref{gsl_permutation},),          # tuple of input types
        p                                 # name of Julia variable to pass in
    )
end

Here, the input p is declared to be of type Ref{gsl_permutation}, meaning that the memory that p points to may be managed by Julia or by C. A pointer to memory allocated by C should be of type Ptr{gsl_permutation}, but it is convertible using Base.cconvert and therefore

Now if you look closely enough at this example, you may notice that it is incorrect, given our explanation above of preferred declaration types. Do you see it? The function we are calling is going to free the memory. This type of operation cannot be given a Julia object (it will crash or cause memory corruption). Therefore, it may be preferable to declare the p type as Ptr{gsl_permutation }, to make it harder for the user to mistakenly pass another sort of object there than one obtained via gsl_permutation_alloc.

If the C wrapper never expects the user to pass pointers to memory managed by Julia, then using p::Ptr{gsl_permutation} for the method signature of the wrapper and similarly in the ccall is also acceptable.

Here is a third example passing Julia arrays:

# The corresponding C signature is
#    int gsl_sf_bessel_Jn_array (int nmin, int nmax, double x,
#                                double result_array[])
function sf_bessel_Jn_array(nmin::Integer, nmax::Integer, x::Real)
    if nmax < nmin
        throw(DomainError())
    end
    result_array = Vector{Cdouble}(undef, nmax - nmin + 1)
    errorcode = ccall(
        (:gsl_sf_bessel_Jn_array, :libgsl), # name of C function and library
        Cint,                               # output type
        (Cint, Cint, Cdouble, Ref{Cdouble}),# tuple of input types
        nmin, nmax, x, result_array         # names of Julia variables to pass in
    )
    if errorcode != 0
        error("GSL error code $errorcode")
    end
    return result_array
end

The C function wrapped returns an integer error code; the results of the actual evaluation of the Bessel J function populate the Julia array result_array. This variable is declared as a Ref{Cdouble}, since its memory is allocated and managed by Julia. The implicit call to Base.cconvert(Ref{Cdouble}, result_array) unpacks the Julia pointer to a Julia array data structure into a form understandable by C.

Fortran Wrapper Example

The following example utilizes ccall to call a function in a common Fortran library (libBLAS) to computes a dot product. Notice that the argument mapping is a bit different here than above, as we need to map from Julia to Fortran. On every argument type, we specify Ref or Ptr. This mangling convention may be specific to your fortran compiler and operating system, and is likely undocumented. However, wrapping each in a Ref (or Ptr, where equivalent) is a frequent requirement of Fortran compiler implementations:

function compute_dot(DX::Vector{Float64}, DY::Vector{Float64})
    @assert length(DX) == length(DY)
    n = length(DX)
    incx = incy = 1
    product = ccall((:ddot_, "libLAPACK"),
                    Float64,
                    (Ref{Int32}, Ptr{Float64}, Ref{Int32}, Ptr{Float64}, Ref{Int32}),
                    n, DX, incx, DY, incy)
    return product
end

Garbage Collection Safety

When passing data to a ccall, it is best to avoid using the pointer function. Instead define a convert method and pass the variables directly to the ccall. ccall automatically arranges that all of its arguments will be preserved from garbage collection until the call returns. If a C API will store a reference to memory allocated by Julia, after the ccall returns, you must ensure that the object remains visible to the garbage collector. The suggested way to do this is to make a global variable of type Array{Ref,1} to hold these values, until the C library notifies you that it is finished with them.

Whenever you have created a pointer to Julia data, you must ensure the original data exists until you have finished using the pointer. Many methods in Julia such as unsafe_load and String make copies of data instead of taking ownership of the buffer, so that it is safe to free (or alter) the original data without affecting Julia. A notable exception is unsafe_wrap which, for performance reasons, shares (or can be told to take ownership of) the underlying buffer.

The garbage collector does not guarantee any order of finalization. That is, if a contained a reference to b and both a and b are due for garbage collection, there is no guarantee that b would be finalized after a. If proper finalization of a depends on b being valid, it must be handled in other ways.

Non-constant Function Specifications

In some cases, the exact name or path of the needed library is not known in advance and must be computed at run time. To handle such cases, the library component of a (name, library) specification can be a function call, e.g. (:dgemm_, find_blas()). The call expression will be executed when the ccall itself is executed. However, it is assumed that the library location does not change once it is determined, so the result of the call can be cached and reused. Therefore, the number of times the expression executes is unspecified, and returning different values for multiple calls results in unspecified behavior.

If even more flexibility is needed, it is possible to use computed values as function names by staging through eval as follows:

@eval ccall(($(string("a", "b")), "lib"), ...

This expression constructs a name using string, then substitutes this name into a new ccall expression, which is then evaluated. Keep in mind that eval only operates at the top level, so within this expression local variables will not be available (unless their values are substituted with $). For this reason, eval is typically only used to form top-level definitions, for example when wrapping libraries that contain many similar functions. A similar example can be constructed for @cfunction.

However, doing this will also be very slow and leak memory, so you should usually avoid this and instead keep reading. The next section discusses how to use indirect calls to efficiently achieve a similar effect.

Indirect Calls

The first argument to ccall can also be an expression evaluated at run time. In this case, the expression must evaluate to a Ptr, which will be used as the address of the native function to call. This behavior occurs when the first ccall argument contains references to non-constants, such as local variables, function arguments, or non-constant globals.

For example, you might look up the function via dlsym, then cache it in a shared reference for that session. For example:

macro dlsym(func, lib)
    z = Ref{Ptr{Cvoid}}(C_NULL)
    quote
        let zlocal = $z[]
            if zlocal == C_NULL
                zlocal = dlsym($(esc(lib))::Ptr{Cvoid}, $(esc(func)))::Ptr{Cvoid}
                $z[] = zlocal
            end
            zlocal
        end
    end
end

mylibvar = Libdl.dlopen("mylib")
ccall(@dlsym("myfunc", mylibvar), Cvoid, ())

Closure cfunctions

The first argument to @cfunction can be marked with a $, in which case the return value will instead be a struct CFunction which closes over the argument. You must ensure that this return object is kept alive until all uses of it are done. The contents and code at the cfunction pointer will be erased via a finalizer when this reference is dropped and atexit. This is not usually needed, since this functionality is not present in C, but can be useful for dealing with ill-designed APIs which don't provide a separate closure environment parameter.

function qsort(a::Vector{T}, cmp) where T
    isbits(T) || throw(ArgumentError("this method can only qsort isbits arrays"))
    callback = @cfunction $cmp Cint (Ref{T}, Ref{T})
    # Here, `callback` isa Base.CFunction, which will be converted to Ptr{Cvoid}
    # (and protected against finalization) by the ccall
    ccall(:qsort, Cvoid, (Ptr{T}, Csize_t, Csize_t, Ptr{Cvoid}),
        a, length(a), Base.elsize(a), callback)
    # We could instead use:
    #    GC.@preserve callback begin
    #        use(Base.unsafe_convert(Ptr{Cvoid}, callback))
    #    end
    # if we needed to use it outside of a `ccall`
    return a
end

Closing a Library

It is sometimes useful to close (unload) a library so that it can be reloaded. For instance, when developing C code for use with Julia, one may need to compile, call the C code from Julia, then close the library, make an edit, recompile, and load in the new changes. One can either restart Julia or use the Libdl functions to manage the library explicitly, such as:

lib = Libdl.dlopen("./my_lib.so") # Open the library explicitly.
sym = Libdl.dlsym(lib, :my_fcn)   # Get a symbol for the function to call.
ccall(sym, ...) # Use the pointer `sym` instead of the (symbol, library) tuple (remaining arguments are the same).
Libdl.dlclose(lib) # Close the library explicitly.

Note that when using ccall with the tuple input (e.g., ccall((:my_fcn, "./my_lib.so"), ...)), the library is opened implicitly and it may not be explicitly closed.

Calling Convention

The second argument to ccall can optionally be a calling convention specifier (immediately preceding return type). Without any specifier, the platform-default C calling convention is used. Other supported conventions are: stdcall, cdecl, fastcall, and thiscall (no-op on 64-bit Windows). For example (from base/libc.jl) we see the same gethostnameccall as above, but with the correct signature for Windows:

hn = Vector{UInt8}(undef, 256)
err = ccall(:gethostname, stdcall, Int32, (Ptr{UInt8}, UInt32), hn, length(hn))

For more information, please see the LLVM Language Reference.

There is one additional special calling convention llvmcall, which allows inserting calls to LLVM intrinsics directly. This can be especially useful when targeting unusual platforms such as GPGPUs. For example, for CUDA, we need to be able to read the thread index:

ccall("llvm.nvvm.read.ptx.sreg.tid.x", llvmcall, Int32, ())

As with any ccall, it is essential to get the argument signature exactly correct. Also, note that there is no compatibility layer that ensures the intrinsic makes sense and works on the current target, unlike the equivalent Julia functions exposed by Core.Intrinsics.

Accessing Global Variables

Global variables exported by native libraries can be accessed by name using the cglobal function. The arguments to cglobal are a symbol specification identical to that used by ccall, and a type describing the value stored in the variable:

julia> cglobal((:errno, :libc), Int32)
Ptr{Int32} @0x00007f418d0816b8

The result is a pointer giving the address of the value. The value can be manipulated through this pointer using unsafe_load and unsafe_store!.

Accessing Data through a Pointer

The following methods are described as "unsafe" because a bad pointer or type declaration can cause Julia to terminate abruptly.

Given a Ptr{T}, the contents of type T can generally be copied from the referenced memory into a Julia object using unsafe_load(ptr, [index]). The index argument is optional (default is 1), and follows the Julia-convention of 1-based indexing. This function is intentionally similar to the behavior of getindex and setindex! (e.g. [] access syntax).

The return value will be a new object initialized to contain a copy of the contents of the referenced memory. The referenced memory can safely be freed or released.

If T is Any, then the memory is assumed to contain a reference to a Julia object (a jl_value_t*), the result will be a reference to this object, and the object will not be copied. You must be careful in this case to ensure that the object was always visible to the garbage collector (pointers do not count, but the new reference does) to ensure the memory is not prematurely freed. Note that if the object was not originally allocated by Julia, the new object will never be finalized by Julia's garbage collector. If the Ptr itself is actually a jl_value_t*, it can be converted back to a Julia object reference by unsafe_pointer_to_objref(ptr). (Julia values v can be converted to jl_value_t* pointers, as Ptr{Cvoid}, by calling pointer_from_objref(v).)

The reverse operation (writing data to a Ptr{T}), can be performed using unsafe_store!(ptr, value, [index]). Currently, this is only supported for primitive types or other pointer-free (isbits) immutable struct types.

Any operation that throws an error is probably currently unimplemented and should be posted as a bug so that it can be resolved.

If the pointer of interest is a plain-data array (primitive type or immutable struct), the function unsafe_wrap(Array, ptr,dims, own = false) may be more useful. The final parameter should be true if Julia should "take ownership" of the underlying buffer and call free(ptr) when the returned Array object is finalized. If the own parameter is omitted or false, the caller must ensure the buffer remains in existence until all access is complete.

Arithmetic on the Ptr type in Julia (e.g. using +) does not behave the same as C's pointer arithmetic. Adding an integer to a Ptr in Julia always moves the pointer by some number of bytes, not elements. This way, the address values obtained from pointer arithmetic do not depend on the element types of pointers.

Thread-safety

Some C libraries execute their callbacks from a different thread, and since Julia isn't thread-safe you'll need to take some extra precautions. In particular, you'll need to set up a two-layered system: the C callback should only schedule (via Julia's event loop) the execution of your "real" callback. To do this, create an AsyncCondition object and wait on it:

cond = Base.AsyncCondition()
wait(cond)

The callback you pass to C should only execute a ccall to :uv_async_send, passing cond.handle as the argument, taking care to avoid any allocations or other interactions with the Julia runtime.

Note that events may be coalesced, so multiple calls to uv_async_send may result in a single wakeup notification to the condition.

More About Callbacks

For more details on how to pass callbacks to C libraries, see this blog post.

C++

For direct C++ interfacing, see the Cxx package. For tools to create C++ bindings, see the CxxWrap package.