Re: NSViewController; loading views from nibs

[Fred Kiefer]

On 15.04.2011 22:05, Ivan Vučica wrote:
> On Mon, Jan 24, 2011 at 12:44, Fred Kiefer<fredkiefer@gmx.de>  wrote:
>
> > > I read a bit of Apple docs and figured out how loading from nibs is done.
> > I
> > > added support in Zcode, and  prepared a patch for GNUstep GUI. It's
> > clumsy
> > > but works for my needs.
> >
> > I submitted a patch similar to yours. This class will still need some
> > work to be truly useful.
> Hi Fred,
>
> It seems to me that your implementation of NSViewController's -loadView
> (which I tested only now) has a bug with retaincount. I never took a deep
> look at what NSNib is doing, but it seems to me that nib being released at
> the end of loadView pulls all other objects with it. That's just my guess.
>
> This bug is manifesting in Zcode. I have NSViewController subclasses for the
> editor, which are unloaded upon file switch and replaced with a different
> NSViewController. All of them are defined in nibs.
>
> Program crashes when I release the old subclass of NSViewController. The
> crash occurs in NSViewController's release, when _topObjects is released.
>
> This problem did not occur with my original submitted patch, which has used
> the same method that is used in NSWindowController's -loadWindow. Zcode used
> this method for a while, and it seems ok to me. Zcode also runs under Cocoa,
> so I guess I am Doing Things Right(tm).
>
> Here is a patch that uses the same method that I used in Zcode. It fixes the
> crash and exposes -loadView publicly (which I think is ok, since people who
> subclass NSViewController should know that -loadView exists).
>
> Please tell me if I introduced a memory leak; I don't know exactly what I'm
> doing and basically I'm learning from NSWindowController's source code.

Sorry, only today did I find this mail plus a lot more from you in my 
spam folder. Don't know why GMX put them there, but it hopefully 
explains why I didn't reply before.

You are definitely correct about the loadView method. It should be 
exported in the header file, probably that's why I added it there three 
months ago :-)
Looks like your header files don't match the implementation files as the 
later one seems up to date.

As for the NIB loading code, the method used in the current GNUstep code 
and the one that you suggest should be almost identical in the result. 
(Ignoring that your code will only load NIBs from the main bundle, that 
would be easily fixed)
If the current method results in a segmentation fault then we have a 
real problem here and just not just ignore it by using a different 
interface. Is the ZCode application freely available and how would I 
reproduce the problem with that code?
Most likely the problem has to do with topLevelObjects, but I cannot see 
the bug at the moment. Maybe it is that I wrote most of the code myself 
that keeps me from spotting the obvious. Anybody with a bit more insight 
out there?
Ah, I think finally I see it, topLevelObjects gets set to an 
autoreleased array and later the NSViewController releases it in its 
dealloc method. We should have retained it after the NIB loading!
I will try to correct this.

Fred