Re: NSView caching of isFlipped

[Quentin Mathé]

Hi Derek,

Le 31 juil. 2010 à 00:10, Derek Fawcus a écrit :

> I came across an interesting effect due to the above,  namely a
> flipped view with the incorrect behaviour.
>
> This was in part defined as:
>
> @interface TextView: NSView
> {
>        BOOL _flipped;
>         /* other fields ... */
> }
> - (id) initWithFrame:(NSRect)frameRect andStorage:(NSTextStorage *) 
> storage isFlipped:(BOOL)flipped;
> @end
>
> @implementation TextView
> - (BOOL)isFlipped
> {
>        return _flipped;
> }
>
> - (id)initWithFrame:(NSRect)frameRect
>         andStorage:(NSTextStorage *)storage
>         isFlipped:(BOOL)flipped
> {
>        if (![super initWithFrame:(NSRect)frameRect])
>                return nil;
>
>        _flipped = flipped;
>
>         /* other stuff ... */
> }
>
> Then due to the effect of +alloc (zeros memory) and the above,
> any view which was supposed to be flipped was not.
> After spending quite a bit of time trying to figure out why the
> program did not operate correctly on GS,  whereas it did no OSX
> and OPENSTEP,  the workaround was eventually simple,  namely to
> move the assignment before the [super initWithFrame].

You could alternatively call the private methods - 
_invalidateCoordinates then -_rebuildCoordinates. But that's not  
better I admit.
To support -setFlipped: method or similar that changes the flipping on  
the fly, there is no other way though.

Here is the not-so-pretty trick I use:
- (BOOL) isFlipped
{
#ifdef GNUSTEP
        return _rFlags.flipped_view;
#else /* Mac OS X */
        return _flipped;
#endif
}

- (void) setFlipped: (BOOL)flag
{
#ifdef GNUSTEP
        _rFlags.flipped_view = flag;
        [self _invalidateCoordinates];
        [self _rebuildCoordinates];
#else /* Mac OS X */
        _flipped = flag;
#endif
}

> So - how much of a performance gain do we get from caching the
> flipped status in NSView?  If it is significant,  then is there
> any way that the value could be sampled upon the first message
> sent _after_ init?  Or should the caching simply be removed?

I submitted a patch (which needs improvements iirc) a while ago about  
this -isFlipped issue: http://savannah.gnu.org/patch/?6630
I'm in favor of removing the caching completely since there is nothing  
in Cocoa doc that states -isFlipped return value must be constant. In  
my tests, Mac OS X reacts transparently to -isFlipped value change at  
any time.

I don't know what the performance impact would be. But I have been  
writing drawing code that abuses message sending a lot and it isn't  
that slow on GNUstep. Although GNUstep drawing is a lot slower than  
Cocoa and I don't know why precisely.

In future, libobjc2 and clang should also be able to easily inline  
methods such as -isFlipped.

Cheers,
Quentin.