Sorry for chiming in so late but here is the simple defined behavior:
On Tiger the linear gradient among others had a bug of reversed
parameters. To maintain binary compatibility, apps linked against
Tiger maintain the broken behavior on Leopard. Apps linked on
Leopard get the new correct behavior. If you link on Leopard and
need to support Tiger, you need to do a runtime check of the OS
Version.
Frank
On Mar 20, 2008, at 5:12 PM, Chris Meyer wrote:
Here's what I'm using to future proof this code (also attached to
avoid line break issues):
// return TRUE if we're using 10.4 behavor.
- (BOOL)linearGradient_10_4_Behavior
{
static BOOL checked = NO;
static BOOL behavior_10_4 = YES;
if ( !checked )
{
CIFilter *gradient_filter = [CIFilter
filterWithName:@"CILinearGradient"];
[gradient_filter setValue:[CIVector vectorWithX:0 Y:0]
forKey:@"inputPoint0"];
[gradient_filter setValue:[CIVector vectorWithX:32 Y:0]
forKey:@"inputPoint1"];
[gradient_filter setValue:[[[CIColor alloc] initWithColor:
[NSColor blueColor]] autorelease] forKey:@"inputColor0"];
[gradient_filter setValue:[[[CIColor alloc] initWithColor:
[NSColor redColor]] autorelease] forKey:@"inputColor1"];
CIImage *ci_image = [gradient_filter
valueForKey:@"outputImage"];
CIFilter *ci_crop = [CIFilter filterWithName:@"CICrop"];
[ci_crop setValue:ci_image forKey:@"inputImage"];
[ci_crop setValue:[CIVector vectorWithX:0 Y:0 Z:32 W:8]
forKey:@"inputRectangle"];
CIImage *final_image = [ci_crop valueForKey:@"outputImage"];
NSImage *image = [[[NSImage alloc]
initWithSize:NSMakeSize(32,8)] autorelease];
[image addRepresentation:[NSCIImageRep
imageRepWithCIImage:final_image]];
NSBitmapImageRep *bitmap = [[[NSBitmapImageRep alloc]
initWithData:[image TIFFRepresentation]] autorelease];
NSColor *color = [bitmap colorAtX:0 y:4];
float red, green, blue, alpha;
[color getRed:&red green:&green blue:&blue alpha:&alpha];
// should be blue. if it is not, then it is NOT the 10.4
(buggy) behavior.
if ( blue > red )
behavior_10_4 = NO;
checked = YES;
}
return behavior_10_4;
}
On Thu, Mar 20, 2008 at 4:57 PM, Adam R. Maxwell <email@hidden>
wrote:
On Mar 20, 2008, at 4:07 PM, Chris Meyer wrote:
> > I don't know exactly what they assumed or how this was
implemented,
> > but I do think that the behavior you're describing sounds like
a bug
> > in CoreImage.
>
> At this point, fixing it is only going break any code that I write
> to work around it. Not a great situation. Maybe I'll just write
code
> that makes a gradient, samples it, and chooses based on empirical
> evidence. Ugh!
FWIW, we noticed that the documented behavior was broken when 10.5
came out and worked around it using
if (floor(NSAppKitVersionNumber) > NSAppKitVersionNumber10_4) {
// assume gradients are in sensible order
}
else {
// assume gradients are in 10.4 order
}
If Apple fixes this to work as documented, it will probably break
numerous apps.
> > Most apps don't need to know what SDK they were built against;
they
> > need to know what the current OS version is. That's what
> > NSAppKitVersionNumber is meant to provide - what version of
AppKit
> are
> > you currently running against. It has _never_ been meant to
indicate
> > what version of the SDK you linked against.
>
> OK. No reason to bang my head against a wall even though this
> behavior seems to be of limited usefulness. But it is what it is.
It's useful for cases like this, where functionality changes based on
framework version instead of linkage (as it was apparently supposed
to).
--
adam
>
>
> On Thu, Mar 20, 2008 at 3:48 PM, Eric Schlegel <email@hidden>
> wrote:
>
> On Mar 20, 2008, at 3:45 PM, Chris Meyer wrote:
> > How would an application ever tell which SDK it was built
against?
>
> Most apps don't need to know what SDK they were built against; they
> need to know what the current OS version is. That's what
> NSAppKitVersionNumber is meant to provide - what version of
AppKit are
> you currently running against. It has _never_ been meant to
indicate
> what version of the SDK you linked against.
>
> > Before you dismiss this message (Eric), I think the case in
point is
> > how CILinearGradient works. It is NOT backwards compatible. The
Core
> > Image guys probably (correctly) assumed that
NSAppKitVersionNumber
> > would return the version of AppKit _linked_ against rather than
> > _available on the system_.
>
> I don't know exactly what they assumed or how this was implemented,
> but I do think that the behavior you're describing sounds like a
bug
> in CoreImage.
>
> -eric
>
>
> _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Quartz-dev mailing list (email@hidden)
> Help/Unsubscribe/Update your Subscription:
> http://lists.apple.com/mailman/options/quartz-dev/amaxwell
%40mac.com
>
> This email sent to email@hidden
<snippet.m> _______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartz-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartz-dev/email@hidden
This email sent to email@hidden
Frank Doepke
Senior Engineer
Core Image
Apple Computer, Inc.
M/S 302-3PG
2 Infinite Loop
Cupertino, CA 95014
phone: 408-974-1266
email: email@hidden
"I thought of this while riding my bicycle" Albert Einstein
