Is there any way to force a GC instance in a DLL to free the heap pages allocated by gc_init? gc_term()is just an alias to fullCollectNoStack().

the following DLL leaks around 1,000-2,000K each time the dll is reloaded.

extern (Windows):
BOOL DllMain(HINSTANCE hInstance, ULONG ulReason, LPVOID pvReserved)
{

    g_hInst = hInstance;

    switch (ulReason){
    case DLL_PROCESS_ATTACH:

        gc_init();
        _minit();
        _moduleCtor();

	activate();
	break;

    case DLL_PROCESS_DETACH:

        _moduleDtor();
        gc_term();
        break;

    case DLL_THREAD_ATTACH:
    case DLL_THREAD_DETACH:

        return false;
    }

    return TRUE;
}

> Is there any way to force a GC instance in a DLL to free the heap pages
> allocated by gc_init? gc_term()is just an alias to fullCollectNoStack().
Tango's gc_term() call's a destructor which frees the calloc'ed heap.

But sorry, I don't know much more than that..

 - Paul

Paul Findlay wrote:
>> Is there any way to force a GC instance in a DLL to free the heap pages
>> allocated by gc_init? gc_term()is just an alias to fullCollectNoStack().
> Tango's gc_term() call's a destructor which frees the calloc'ed heap.
> 
> But sorry, I don't know much more than that..
> 
>  - Paul

Alright, well i added this to src/phobos/internal/gc.d and rebuilt phobos.lib.

extern (C) void gc_destroy(){ _gc.Dtor(); }


And added gc_destroy() to the unload section of DllMain.


Looks like it solved the problem somewhat - it only leaks 100kb now when the DLL reloads instead of 1.5mb. That 100kb could easily be the fault of my program too; i'm not sure atm.


Should i file a bug report about this or what? Its possible gc_term()'s behavior has been the same ever since the start.

I'm using DMD2.02 if i haven't mentioned it already.