Test_ChildVisitor fails with llvm 3.2

[chetant]

Some name strings in the test have the first four chars replaced with '\0' (null) chars.
Here's the diff from the C and Haskell version outputs.

58c58
< Name:mode_t, Kind:TypedefDecl

---
> Name:_t, Kind:TypedefDecl
143c143
< Name:_G_uint16_t, Kind:TypedefDecl

---
> Name:int16_t, Kind:TypedefDecl
344c344
< Name:clang_equalCursors(CXCursor, CXCursor), Kind:FunctionDecl

---
> Name:g_equalCursors(CXCursor, CXCursor), Kind:FunctionDecl
420c420
< Name:clang_getSpecializedCursorTemplate(CXCursor), Kind:FunctionDecl

---
> Name:g_getSpecializedCursorTemplate(CXCursor), Kind:FunctionDecl
455c455
< Name:clang_defaultCodeCompleteOptions(), Kind:FunctionDecl

---
> Name:g_defaultCodeCompleteOptions, Kind:FunctionDecl

[sethfowler]

I don't know enough to fix this, but consider me interested in a fix. I see segfaults and other strange behavior on the LLVM 3.2 branch. I suspect they're caused by the same issue described here.

[sethfowler]

Now I do know enough to fix this, and I fixed it. =) See issue #18.

[chetant]

Thanks for the changes, merged!
I see the problem, the ForeignPtr instance was gc'd before we were done with the Ptr we got out of it(unsafely), triggering the finalizer.
There are other places where I've used the same method (all the places in FFI.gc where I've used "fptr (ptr c)", I'm extracting the pointer.
I'm going to think about how to get this cleaned up..