Re: NSView draw order messed up

[Matt Rice]

On Wed, Oct 2, 2013 at 6:47 AM, Fred Kiefer <fredkiefer@gmx.de> wrote:
> I think this is an unrelated issue.

probably, just wanted to throw it out there as one possible avenue, as
I recall this one being not so fun to figure out.

> First off, I always thought the drawing order of sibling views is undefined. 
> Could you please point me to a document that states otherwise?

I know of no such document, it is undefined.


> Still if Apple followes the order in wich views wher added, we should do so 
> two. But as far as I remember this is already the case.
> And overlapping sibling views are a no-go in GNUstep, they might work for 
> drawing, but event handling will be screwed up.

Correct, on OPENSTEP/the osx's i've tested this was the case,
drawing was done in the order added (so back-most in the sibling views
array was on top) IIRC

event handling was done in the opposite order (so the back-most
top-displayed view had first shot at the event)

when using views that overlap it required care from the developer that
all sibling NSViews which overlap the rectangle needing display
-setNeedsDisplay at the same time, if you setNeedsDisplay a view which
is not top-most (while not setNeedsDisplay on the top-most),
display will become out of sync with event handling, thus it was
something which worked with the correct parameters, was technically
undefined, because its possible to get bad results.

similarly, if the top-most view doesn't respond to events, and the a
view it overlaps does respond to events you are likely screwed as
well.

its not something they want to define/advertise because honestly there
are more ways to screw it up than do it correctly.  but it has worked
given a careful implementation.



> Could you please provide an example that demonstrates the issue? After the 
> weekend I am Wiiling to look into this.

I don't really have a test case for it or a gnustep install around,
afaik until now DBModeler was the only thing using this feature and I
haven't compiled it in years, and it is undoubtedly broken.

hopefully Johannes can come up with something that exhibits the problem.