Re: CA broken; how do I differentiate NSValues for size and point?

[Fred Kiefer]

I spend some time to understand this issue. But to be honest it took me more 
time to understand why this ever worked :-)

The problem here is that base uses two different ways to provide concrete 
subclasses for NSValue. One being GSValue, which supports generic data and the 
other mechanism is to have a specific subclass for types like NSPoint and 
NSSize. For the later base only generates specific methods for that types. 
Resulting in different classes for NSPoint and NSSize, which are incompatible 
although they have the same underlying data size.
In your demo application that animation for shadowOffset tries to set the 
offset size with a suitable default value of class GSSizeValue. But the code in 
GSObjCSetVal only compares the type information without looking at the type 
names. That way NSSize is regarded as the same as NSPoint and the pointValue 
get executed, resulting in the error you reported. Either GSObjCSetVal has to 
be more careful and use a check for the actual type first. Or we need to make 
the NSValue subclasses more general.

But why did it work before? Most likely because at that time CGSize and 
CGPoint, where different from NSSize and NSPoint so we did not get the specific 
optimisation in NSValue.

Hope this helps,
Fred

> Am 20.05.2018 um 14:03 schrieb Ivan Vučica <address@hidden>:
> 
> Hi,
> 
> Pretty much all Core Animation demos are currently broken under GNUstep with 
> a variation on the following:
> 
> 2018-05-20 12:54:25.464 QuartzCoreDemo[13476:13476] Problem posting 
> notification: <NSException: 0x15b53b8> NAME:NSInvalidArgumentException 
> REASON:[GSSizeValue-pointValue] should be overridden by subclass INFO:(null)
> 
> CA is accessing -pointValue method if it determines that the passed NSValue 
> matches the -objCType of NSPoint. It does not check for size values.
> 
> Clearly, sometimes it is trying to interpolate size values, which will match 
> the signature and it will incorrectly attempt to access -pointValue which is 
> then not implemented by GSSizeValue. I am not sure where might it be 
> interpolating size values, but it seems to be doing so. (Alternative bug is 
> that NSValue is incorrectly ingesting points as sizes, then complaining when 
> the point provided is being interpreted as a point. I can try checking this 
> later.)
> 
> I cannot check -respondsToSelector: because the class /does/ in fact respond 
> to -pointValue; it just throws an exception.
> 
> Adding a try/catch in this kind of situation would make for some very poor 
> code hygiene, in my opinion.
> 
> - Is NSValue supposed to be a class cluster like this? (Not on Mac at this 
> time, can't check.)
> - Is there a way out?
> - Would it make sense to extend GSSizeValue and add -pointValue to it? 
> (They're both 2d vectors.)
> 
> Thanks
> _______________________________________________
> Gnustep-dev mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/gnustep-dev