managing embedded structs

I want to write a Haskell interface to a C library that provides two data structures A and B, where B is contained in A. That is, if A is freed, then B is automatically freed, too. There are two ways to obtain a Ptr B. Either construct it separately or as part of A, i.e. b0 :: Ptr B b0 <- createB b1 :: Ptr B b1 <- getBofA =<< createA Now I want to let the garbage collector manage the resources for B in a ForeignPtr. The first case is easy: fb0 :: ForeignPtr B fb0 <- newForeignPtr deleteB =<< createB But how to create an fb1? fa1 :: ForeignPtr A fa1 <- newForeignPtr deleteA =<< createA fb1 :: ForeignPtr B fb1 <- newForeignPtr ??? =<< withForeignPtr fa1 getBofA Obviously, there is nothing I can insert for ???. fb1 must not have its own finalizer, because fa1 already has one. But I have to make sure that fa1 lives at least as long as fb1. How to do that?

On 27 August 2016 at 13:23, Henning Thielemann
I want to write a Haskell interface to a C library that provides two data structures A and B, where B is contained in A. That is, if A is freed, then B is automatically freed, too.
I don't quite understand. Do A and B live in the same buffer and problem in keeping buffer alive as long as there're pointers to A or B. If so it's easy problem and vector solves exactly this problem for storable vectors. Check implementation of basicUnsafeSlice. Idea is to share finalizer between ForeignPtr If they live in different buffers and B should not be collected as long as A lives... I don't know how to implement this.

On Sun, 28 Aug 2016, Alexey Khudyakov wrote:
On 27 August 2016 at 13:23, Henning Thielemann
wrote: I want to write a Haskell interface to a C library that provides two data structures A and B, where B is contained in A. That is, if A is freed, then B is automatically freed, too.
I don't quite understand. Do A and B live in the same buffer and problem in keeping buffer alive as long as there're pointers to A or B. If so it's easy problem and vector solves exactly this problem for storable vectors. Check implementation of basicUnsafeSlice. Idea is to share finalizer between ForeignPtr
Sharing finalizer between ForeignPtr A and ForeignPtr B might indeed be a solution. It would require dependence on the ghc package (and thus GHC), though.
If they live in different buffers and B should not be collected as long as A lives...
This would be true in another situation, where I create a finalizer by the LLVM-JIT and need to finalize the finalizer's code if it is no longer needed. https://ghc.haskell.org/trac/ghc/ticket/12547
participants (2)
-
Alexey Khudyakov
-
Henning Thielemann