Calling C Functions
Starting from the 1.3 beta version, Luna comes equipped with the ability to call foreign C functions from object files. This document describes how to use this system to create C based libraries.
Object files
Object files contain compiled code that can be dynamically loaded and used by other programs. The most common types are .so
files on Linux, .dylib
files on macOS and .dll
on Windows. Luna is able to work with object files coming from two sources: 1. System-wide available dynamic libraries. If a library is installed on your machine, Luna will be able to use it. Unfortunately, this requires all the users of your library to have the dynamic library installed in their systems. For most libraries this is not a good solution, it will, however, work fine for bindings to the most popular libraries and for quick prototyping. 2. Object files inside the project tree. You can drop any shared object file into the native_libs/PLATFORM
directory inside your project (PLATFORM
is macos
, linux
or windows
) and use it from Luna. This is a great solution for making sure users are able to use your library without any external dependencies. It's also the only way to include your own C extension or wrapper.
Using a shared library
[info] Changes ahead!
This describes the bare-bones way of interacting with foreign objects. There are some syntactic changes coming soon to cut down on boilerplate. The basic concepts, however, will remain unchanged.
OK, so you have your shared library in place, how to actually use it from Luna? For this example, we'll assume that we're working with the following directory structure:
The object files in this example contain just a single function, with the following C signature:
How do we use it?
Resolving the symbol
The workhorse of our FFI (Foreign Function Interface) is the lookupSymbol
function from Std.Foreign
module. It takes a shared object file name and a name of the symbol to lookup, and returns a FunPtr
object. The library resolution takes into account the most common naming conventions, so that it is possible to use the same call with a file named awesome.dylib
as well as libawesome.4.so
. So, in our example, you can just write:
Calling the FunPtr
Now that we've obtained the function pointer, it's time to call the function. This is what the call
method of FunPtr
does. It takes two arguments – a representation for the output type and a list of objects of type CArg
, representing the function arguments. Note that not every type can be represented as a C argument. The types that can (numeric types, pointers and a few more) are converted using their toCArg
method. So, back to our example. The awesomeFunction
expects and int
and returns an int
. Note that Luna's Int
s are different from C's, so the latter are represented by the class CInt
. This is what our function call looks like:
Translating between C and Luna data types
We have successfully called a C function. It is not very usable for most Luna code though – it requires us to use the C types throughout the program. We'll fix that with a few conversion methods. First of all, the function argument needs to be a CInt
. Most Luna programs, however, use standard Int
s and these cannot easily be represented in C. This can be fixed with the help of CInt.fromInt
function, which converts an ordinary Int
object into a CInt
. It also returns a CInt
, which again is not very handy. This can be fixed with the help of toInt
method of CInt
. The final version of our wrapper, which operates on Int
s and from the outside is nearly indistinguishable from pure Luna code looks like this:
Basic C Types
Luna defines counterparts of the standard C types in the Std.Foreign.C.Value
module. This section is a short overview of all of them.
Integer types
There is a counterpart for integral types commonly used throughout C codebases. All these wrappers have a common API – they define the fromInt
and toInt
methods, support comparison and basic arithmetic operators. The following table shows the C types and their Luna counterparts.
Floating point numbers
Luna defines the CDouble
and CFloat
classes as counterparts of C's double
and float
, respectively. Both classes define fromReal
and toReal
methods for conversion between them and Luna's Real
. They also support basic arithmetic and comparison operators.
Pointers
The basic pointer type is just Pointer
. It takes a single argument denoting the type of its content. So, for example, Pointer CInt
corresponds to int*
in C, while Pointer (Pointer None)
is void**
.
Creating and freeing pointers
To create a pointer able to hold a single value of type X
use the malloc
method on pointer class:
For this to work, the X
object must define a byteSize
method, returning the size in bytes of the structure. You can also use mallocElems
to create arrays like so:
Any pointer can be freed using its free
method – if you're done with your ptr
just call ptr.free
.
Reading and writing
Reading a pointer can be accomplished by its read
method. It works on a Ptr X
and returns a value of type X
. Keep in mind that in order for this to work, the X
type must define a readPtr
method that takes a bare pointer and plucks the fields one by one.
Similarly, to write a value to ptr
of type Pointer X
, call ptr.write x
. Again, this requires the type X
to define writePtr
method.
For the basic C types, the required methods are already defined in the standard library.
Pointer arithmetic
Any pointer can be moved by a specified number of bytes using the ptr.moveBytes i
method – it returns a new pointer, resulting from adding i
bytes to ptr
. There is also a moveElems
method, that will move the pointer by the specified number of elements (i.e. by number of elements * element.byteSize
bytes).
Managed Pointers
With standard pointers we need to think about freeing unused memory, or it will clutter up our RAM. We can fix that issue with managed pointers – pointers that can be automatically garbage collected when no longer needed. Since managed pointers are available the regular pointers should not be used any more. To create managed pointer for single value of X
type call, like for pointer, malloc
method just on the managed pointer class:
Allocating multiple elements with mallocElems
works just like for regular poinetrs. To create array with managed pointer use:
It is also possible to create managed pointer from existing pointer ptr
. For this finalizer function fin
is required. Finalizer will be run when the pointer will be garbage collected:
Methods like read
, write
, moveElems
works the same way for managed pointers like for regular pointers.
Real life example
Now that we've covered all the basics, let's dive into a more involved example – using the SHA1
function from openssl
. It takes an input buffer of type unsigned char*
, a size_t
denoting the length of input and an output buffer of type unsigned char*
and length 20. Suppose you have a list of Luna Int
s and want to compute the SHA1 digest of this list, as another list of Int
s. This is how this can be done with Luna's FFI:
Last updated