inline-c-0.9.1.3: Write Haskell source files including C code inline. No FFI required.

Language.C.Inline

Description

Enable painless embedding of C code in Haskell code. If you're interested in how to use the library, skip to the "Inline C" section. To build, read the first two sections.

This module is intended to be imported qualified:

import qualified Language.C.Inline as C

Synopsis

# GHCi

Currently inline-c does not work in interpreted mode. However, GHCi can still be used using the -fobject-code flag. For speed, we reccomend passing -fobject-code -O0, for example

stack ghci --ghci-options='-fobject-code -O0'


or

cabal repl --ghc-options='-fobject-code -O0'


# Contexts

data Context Source #

A Context stores various information needed to produce the files with the C code derived from the inline C snippets.

Contexts can be composed with their Monoid instance, where mappend is right-biased -- in mappend x y y will take precedence over x.

Instances
 Source # Instance detailsDefined in Language.C.Inline.Context Methodsstimes :: Integral b => b -> Context -> Context # Source # Instance detailsDefined in Language.C.Inline.Context Methodsmconcat :: [Context] -> Context #

Context useful to work with vanilla C. Used by default.

ctxTypesTable: converts C basic types to their counterparts in Foreign.C.Types.

No ctxAntiQuoters.

This Context adds support for ForeignPtr arguments. It adds a unique marshaller called fptr-ptr. For example, $fptr-ptr:(int *x) extracts the bare C pointer out of foreign pointer x. This Context includes a AntiQuoter that removes the need for explicitely creating FunPtrs, named "fun" along with one which allocates new memory which must be manually freed named "fun-alloc". For example, we can capture function f of type CInt -> CInt -> IO CInt in C code using $fun:(int (*f)(int, int)).

When used in a pure embedding, the Haskell function will have to be pure too. Continuing the example above we'll have CInt -> CInt -> IO CInt.

Does not include the baseCtx, since most of the time it's going to be included as part of larger contexts.

IMPORTANT: When using the fun anti quoter, one must be aware that the function pointer which is automatically generated is freed when the code contained in the block containing the anti quoter exits. Thus, if you need the function pointer to be longer-lived, you must allocate it and free it manually using freeHaskellFunPtr. We provide utilities to easily allocate them (see mkFunPtr).

IMPORTANT: When using the fun-alloc anti quoter, one must free the allocated function pointer. The GHC runtime provides a function to do this, hs_free_fun_ptr available in the h header.

This Context includes two AntiQuoters that allow to easily use Haskell vectors in C.

Specifically, the vec-len and vec-ptr will get the length and the pointer underlying mutable (IOVector) and immutable (Vector) storable vectors.

Note that if you use vecCtx to manipulate immutable vectors you must make sure that the vector is not modified in the C code.

To use vec-len, simply write $vec-len:x, where x is something of type IOVector a or Vector a, for some a. To use vec-ptr you need to specify the type of the pointer, e.g. $vec-len:(int *x) will work if x has type IOVector CInt.

bsCtx serves exactly the same purpose as vecCtx, but only for ByteString. vec-ptr becomes bs-ptr, and vec-len becomes bs-len. You don't need to specify the type of the pointer in bs-ptr, it will always be char*.

Moreover, bs-cstr works as bs-ptr but it provides a null-terminated copy of the given ByteString.

Sets the Context for the current module. This function, if called, must be called before any of the other TH functions in this module. Fails if that's not the case.

# Substitution

substitute :: [(String, String -> String)] -> Q a -> Q a Source #

Define macros that can be used in the nested Template Haskell expression. Macros can be used as @MACRO_NAME(input) in inline-c quotes, and will transform their input with the given function. They can be useful for passing in types when defining Haskell instances for C++ template types.

Given a C type name, return the Haskell type in Template Haskell. The first parameter controls whether function pointers should be mapped as pure or IO functions.

# Inline C

The quasiquoters below are the main interface to this library, for inlining C code into Haskell source files.

In general, quasiquoters are used like so:

[C.XXX| int { <C code> } |]


Where C.XXX is one of the quasi-quoters defined in this section.

This syntax stands for a piece of typed C, decorated with a type:

• The first type to appear (int in the example) is the type of said C code.
• The syntax of the <C code> depends on on the quasi-quoter used, and the anti-quoters available. The exp quasi-quoter expects a C expression. The block quasi-quoter expects a list of statements, like the body of a function. Just like a C function, a block has a return type, matching the type of any values in any return statements appearing in the block.

See also the README.md file for more documentation.

### Anti-quoters

Haskell variables can be captured using anti-quoters. inline-c provides a basic anti-quoting mechanism extensible with user-defined anti-quoters (see Language.C.Inline.Context). The basic anti-quoter lets you capture Haskell variables, for example we might say

let x = pi / 3 in [exp| double { cos($(double x)) } |]  Which would capture the Haskell variable x of type CDouble. In C expressions the $ character is denoted using .

### Variable capture and the typing relation

The Haskell type of the inlined expression is determined by the specified C return type. The relation between the C type and the Haskell type is defined in the current Context -- see convertCType. C pointers and arrays are both converted to Haskell Ptrs, and function pointers are converted to FunPtrs. Sized arrays are not supported.

Similarly, when capturing Haskell variables using anti-quoting, their type is assumed to be of the Haskell type corresponding to the C type provided. For example, if we capture variable x using double x in the parameter list, the code will expect a variable x of type CDouble in Haskell (when using baseCtx).

### Purity

The exp and block quasi-quotes denote computations in the Purity monad. pure denotes a pure value, expressed as a C expression.

### Safe and unsafe calls

unsafe variants of the quasi-quoters are provided in Language.C.Inline.Unsafe to call the C code unsafely, in the sense that the C code will block the RTS, but with the advantage of a faster call to the foreign code. See https://www.haskell.org/onlinereport/haskell2010/haskellch8.html#x15-1590008.4.3.

## Examples

### Inline C expression

{-# LANGUAGE QuasiQuotes #-}
import qualified Language.C.Inline as C
import qualified Language.C.Inline.Unsafe as CU
import           Foreign.C.Types

C.include "<math.h>"

c_cos :: CDouble -> IO CDouble
c_cos x = [C.exp| double { cos($(double x)) } |] faster_c_cos :: CDouble -> IO CDouble faster_c_cos x = [CU.exp| double { cos($(double x)) } |]


### Inline C statements

{-# LANGUAGE QuasiQuotes #-}
import qualified Data.Vector.Storable.Mutable as V
import qualified Language.C.Inline as C
import           Foreign.C.Types

C.include "<stdio.h>"

parseVector :: CInt -> Purity (V.IOVector CDouble)
parseVector len = do
vec <- V.new $fromIntegral len0 V.unsafeWith vec$ \ptr -> [C.block| void {
int i;
for (i = 0; i < $(int len); i++) { scanf("%lf ", &$(double *ptr)[i]);
}
} |]
return vec


## How it works

For each quasi-quotation of C code, a C function is generated in a C file corresponding to the current Haskell file. Every inline C expression will result in a corresponding C function. For example, if we define c_cos as in the example above in CCos.hs, we will get a file containing

#include math.h

return cos(x_inline_c_0);
}


Every anti-quotation will correspond to an argument in the C function. If the same Haskell variable is anti-quoted twice, this will result in two arguments.

The C function is then automatically compiled and invoked from Haskell with the correct arguments passed in.

C expressions.

Variant of exp, for use with expressions known to have no side effects.

BEWARE: Use this function with caution, only when you know what you are doing. If an expression does in fact have side-effects, then indiscriminate use of pure may endanger referential transparency, and in principle even type safety. Also note that the function might be called multiple times, given that unsafeDupablePerformIO is used to call the provided C code. Please refer to the documentation for unsafePerformIO for more details. unsafeDupablePerformIO is used to ensure good performance using the threaded runtime.

C code blocks (i.e. statements).

Emits a CPP include directive for C code associated with the current module. To avoid having to escape quotes, the function itself adds them when appropriate, so that

include "foo.h" ==> #include "foo.h"


but

include "<foo>" ==> #include <foo>


Emits an arbitrary C string to the C code associated with the current module. Use with care.

# Ptr utils

withPtr :: Storable a => (Ptr a -> IO b) -> IO (a, b) Source #

Like alloca, but also peeks the contents of the Ptr and returns them once the provided action has finished.

withPtr_ :: Storable a => (Ptr a -> IO ()) -> IO a Source #

class WithPtrs a where Source #

Type class with methods useful to allocate and peek multiple pointers at once:

withPtrs_ :: (Storable a, Storable b) => ((Ptr a, Ptr b) -> IO ()) -> IO (a, b)
withPtrs_ :: (Storable a, Storable b, Storable c) => ((Ptr a, Ptr b, Ptr c) -> IO ()) -> IO (a, b, c)
...


Minimal complete definition

withPtrs

Associated Types

type WithPtrsPtrs a :: * Source #

Methods

withPtrs :: (WithPtrsPtrs a -> IO b) -> IO (a, b) Source #

withPtrs_ :: (WithPtrsPtrs a -> IO ()) -> IO a Source #

Instances
 (Storable a, Storable b) => WithPtrs (a, b) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b) -> IO b0) -> IO ((a, b), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b) -> IO ()) -> IO (a, b) Source # (Storable a, Storable b, Storable c) => WithPtrs (a, b, c) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b, c) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b, c) -> IO b0) -> IO ((a, b, c), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b, c) -> IO ()) -> IO (a, b, c) Source # (Storable a, Storable b, Storable c, Storable d) => WithPtrs (a, b, c, d) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b, c, d) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b, c, d) -> IO b0) -> IO ((a, b, c, d), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b, c, d) -> IO ()) -> IO (a, b, c, d) Source # (Storable a, Storable b, Storable c, Storable d, Storable e) => WithPtrs (a, b, c, d, e) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b, c, d, e) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b, c, d, e) -> IO b0) -> IO ((a, b, c, d, e), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b, c, d, e) -> IO ()) -> IO (a, b, c, d, e) Source # (Storable a, Storable b, Storable c, Storable d, Storable e, Storable f) => WithPtrs (a, b, c, d, e, f) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b, c, d, e, f) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b, c, d, e, f) -> IO b0) -> IO ((a, b, c, d, e, f), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b, c, d, e, f) -> IO ()) -> IO (a, b, c, d, e, f) Source # (Storable a, Storable b, Storable c, Storable d, Storable e, Storable f, Storable g) => WithPtrs (a, b, c, d, e, f, g) Source # Instance detailsDefined in Language.C.Inline Associated Typestype WithPtrsPtrs (a, b, c, d, e, f, g) :: Type Source # MethodswithPtrs :: (WithPtrsPtrs (a, b, c, d, e, f, g) -> IO b0) -> IO ((a, b, c, d, e, f, g), b0) Source #withPtrs_ :: (WithPtrsPtrs (a, b, c, d, e, f, g) -> IO ()) -> IO (a, b, c, d, e, f, g) Source #

# FunPtr utils

Easily get a FunPtr:

let fp :: FunPtr (Ptr CInt -> IO ()) = [C.funPtr| void poke42(int *ptr) { *ptr = 42; } |]


Especially useful to generate finalizers that require C code.

Most importantly, this allows you to write newForeignPtr invocations conveniently:

do
let c_finalizer_funPtr =
[C.funPtr| void myfree(char * ptr) { free(ptr); } |]
fp <- newForeignPtr c_finalizer_funPtr objPtr


Using where possible newForeignPtr is superior to resorting to its delayed-by-a-thread alternative newForeignPtr from Foreign.Concurrent which takes an IO () Haskell finaliser action: With the non-concurrent newForeignPtr you can guarantee that the finaliser will actually be run

• when a GC is executed under memory pressure, because it can point directly to a C function that doesn't have to run any Haskell code (which is problematic when you're out of memory)
• when the program terminates (newForeignPtr's finaliser will likely NOT be called if your main thread exits, making your program e.g. not Valgrind-clean if your finaliser is free or C++'s delete).

funPtr makes the normal newForeignPtr as convenient as its concurrent counterpart.

## FunPtr conversion

$(mkFunPtr [t| CDouble -> IO CDouble |]  generates a foreign import wrapper of type (CDouble -> IO CDouble) -> IO (FunPtr (CDouble -> IO CDouble))  And invokes it. $(mkFunPtrFromName 'foo), if foo :: CDouble -> IO CDouble, splices in an expression of type IO (FunPtr (CDouble -> IO CDouble)).

\$(peekFunPtr [t| CDouble -> IO CDouble |]) generates a foreign import dynamic of type

FunPtr (CDouble -> IO CDouble) -> (CDouble -> IO CDouble)


And invokes it.