[Haskell-cafe] Loading a texture in OpenGL

[L Corbijn]

On Tue, Feb 7, 2012 at 5:23 AM, Austin Seipp <mad.one at gmail.com> wrote:
> Just to clarify, this guarantee possibly could be made, ghc just doesn't do
> it now. In the past ghc never guaranteed a finalizer would ever be run.
>
> Regardless I would be wary of trusting finalizers to clean up very scarce
> resources. A malloc'd buffer is probably fine to have a finalizer for,
> texture objects or file descriptors are a different matter. Predictability
> matters in those cases.
>
>
> Sent from my iPhone^H^H^H^H^HPortable Turing machine
>
> On Feb 6, 2012, at 10:16 PM, Austin Seipp <mad.one at gmail.com> wrote:
>
> It's a precise GC of course (conservative collection would be madness
> considering how much memory Haskell programs chew through.) That still
> doesn't ensure your finalizer will run during the next GC even if all the
> references are gone by then.
>
> Sent from my iPhone^H^H^H^H^HPortable Turing machine
>
> On Feb 6, 2012, at 10:09 PM, Clark Gaebel <cgaebel at csclub.uwaterloo.ca>
> wrote:
>
> Is the Haskell garbage collector conservative, or precise?
>
> If it's conservative, then this will only usually work. If it's precise, it
> should always work.
>
> On Mon, Feb 6, 2012 at 10:56 PM, Ben Lippmeier <benl at ouroborus.net> wrote:
>>
>>
>> On 07/02/2012, at 2:50 PM, Clark Gaebel wrote:
>>
>> > I would be running the GC manually at key points to make sure it gets
>> > cleaned up. Mainly, before any scene changes when basically everything gets
>> > thrown out anyways.
>>
>>
>> From the docs:
>>
>> newForeignPtr :: FinalizerPtr a -> Ptr a -> IO (ForeignPtr a)Source
>> Turns a plain memory reference into a foreign pointer, and associates a
>> finalizer with the reference. The finalizer will be executed after the last
>> reference to the foreign object is dropped. There is no guarantee of
>> promptness, however the finalizer will be executed before the program exits.
>>
>>
>> "No guarantee of promptness". Even if the GC knows your pointer is
>> unreachable, it might choose not to call the finaliser. I think people have
>> been bitten by this before.
>>
>> Ben.
>>
>>
>
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>
>
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>

You have to be really careful with automatic cleanup using finalizers.
Not only to the mentioned not cleaning up of resources but also to
deleting it too early. What could happen is that you code doesn't
explicitly use (= call a GL function with it) an object after binding
it. Then it could be seen as a dead pointer and be deleted by it's
finalizer. But thereby it might unbind the object from the binding
point.
Maybe a (more) realistic example is using only one shader program, you
create it once, call usePorgram with it and never change it after
that. As there is no other program to use you don't have to reactivate
the program with useProgram or have to change anything about it. So in
effect it's not used by Haskell anymore and the finalizer is run for
it deleting the program. For a program this is not a big problem as
the OpenGL spec tells you that it isn't deleted immediately so you can
use it afterwards.

With textures it's different, the are deleted immediately, so you may
think you have a texture bound but in reality the finalizer might have
run and deleted the texture for you. So watch doing the OpenGL memory
management using the references in Haskell, as you might accidentally
delete objects ahead of time.

Lars,

P.S. To make it worse, the bound objects (programs, textures, etc.)
can also be queried and thereby there are non dead objects
automatically, but there is no Haskell reference to them so the GC
cannot now this.