Re: [AD] aRts "pure virtual function call"

[<tjaden@xxxxxxxxxx>]

> Selon Eric Botcazou <ebotcazou@xxxxxxxxxx>:
>> > I think this came up before, but couldn't find a reference to it.
>>
>> Yes, we had this kind of problem with the SVGAlib module, but we
>> "fixed" it  by not unloading the module on exit.  In light of your
>> diagnostic, I can't  figure out the rationale anymore...
>

The rationale was that SVGAlib registers its own shutdown function with
atexit(), and so if the SVGAlib module was unloaded, the atexit mechanism
would be left with a dangling function reference.

> The rationale for not unloading the module on exit ? I don't know. It
> sure is  *much* easier. In the case of the arts driver, I considered
> doing this, but  what if the user calls remove_sound() in the middle of
> the program ? The sound  wouldn't really be removed... this could be
> reasonable, though.

The sound system could be shut down whilst leaving the module loaded in
memory.

>
> I have found a problem with the patch I proposed: each time
> install_sound is  called, a new function is registered with atexit. This
> doesn't break anything,  and my solution still works, but this could be
> considered a memory leak... I'll  think some more about it tonight.

If it's the same problem as the SVGAlib module, then your patch won't work.

>
> Argh! This atexit stuff is a real mess :-)

Worst. Design. Ever.