On Apr 19, 2017, at 15:49 , Charles Srstka <[email protected]> wrote:
>
> Cocoa automagically does its secret subclass thing to wrap the setter and
> call the didChange/willChange calls for any property you didn’t tell it not
> to do. It needs the property to be dynamic for this to work.
Yes, that’s almost exactly what I said**. But it still doesn’t make the
“dynamic” keyword an on/off switch. Again, the KVO notifications aren’t
activated *because* the method has the Swift-specific dynamic attribute, but
because the method uses Obj-C dispatching. The “dynamic” keyword [currently]
forces the method to use Obj-C dispatching, but the reverse isn’t true. In the
absence of the keyword, there’s nothing formally stopping the compiler from
using Obj-C dispatching if it chooses to.
At some level, though, I’m prepared to stipulate that it’s a distinction
without much practical difference.
> I should add that if a computed property needs ‘dynamic’ in order for its
> notifications to fire, the property is not properly KVO-compliant.
It’s impossible to say in general what counts as a change to a mutable
property. Indeed it’s perfectly possible for a property to exist for the
purpose of generating KVO notifications without having any meaningful value,
and there are plenty more un-generalizable considerations:
— It’s up to an individual property (with a meaningful value) to decide whether
KVO notifications are issued when the setter is called (in effect, the default
case) or when the value actually changes (as in Rick’s original code), or under
some other conditions (e.g. I can imagine a property limiting the rate of
notifications using a timer).
— There’s no general presumption whether the value returned by the getter is
synchronized with the value passed to the setter.
— There’s no general presumption whether the value returned by the getter is
synchronized with the notification, in a multi-threaded environment.
— There’s no general presumption that KVO notifications originate in the
property's setter at all, or in the setter only.
Etc. “keyPathsForValuesAffecting…”-style dependencies are just a convenient
short-cut to a specific, simple, typical behavior.
** Incidentally, when I was testing in Swift, I didn’t see any indication the
backtrace that a subclass was being used, or that any actual swizzling was
being done. It looked more like the dynamic dispatch was diverted to a
general-purpose function that generated the notifications around the actual
setter invocation. Here’s a partial backtrace when a breakpoint in the
“version” didSet accessor was hit:
frame #0: 0x000000010000190b
TestKVO`ViewController.version.willset(newValue="1", self=0x0000618000100900)
at ViewController.swift:16
frame #1: 0x0000000100001877
TestKVO`ViewController.version.setter(newValue="1", self=0x0000618000100900) at
ViewController.swift:0
frame #2: 0x00000001000017d8 TestKVO`@objc ViewController.version.setter at
ViewController.swift:0
frame #3: 0x00007fff91d6c897
Foundation`-[NSObject(NSKeyValueObservingPrivate)
_changeValueForKeys:count:maybeOldValuesDict:usingBlock:] + 848
frame #4: 0x00007fff91bf1c7d
Foundation`-[NSObject(NSKeyValueObservingPrivate)
_changeValueForKey:key:key:usingBlock:] + 60
frame #5: 0x00007fff91c5a55b Foundation`_NSSetObjectValueAndNotify + 261
frame #6: 0x000000010000310f
TestKVO`ViewController.buttonClicked(sender=some, self=0x0000618000100900) ->
() at ViewController.swift:57
frame #7: 0x0000000100003200 TestKVO`@objc
ViewController.buttonClicked(Any?) -> () at ViewController.swift:0
frame #8: 0x00007fffa5b903a7
libsystem_trace.dylib`_os_activity_initiate_impl + 53
frame #9: 0x00007fff8e475791 AppKit`-[NSApplication(NSResponder)
sendAction:to:from:] + 456
and here is a disassembly of the Swift code (frame 6) that invokes the setter:
0x1000030f2 <+434>: movq 0x3ebf(%rip), %rsi ; "setVersion:"
0x1000030f9 <+441>: movq %rax, %rcx
0x1000030fc <+444>: movq -0x20(%rbp), %rdx
0x100003100 <+448>: movq %rdx, %rdi
0x100003103 <+451>: movq %rcx, %rdx
0x100003106 <+454>: movq %rax, -0x80(%rbp)
0x10000310a <+458>: callq 0x1000047d2 ; symbol stub for:
objc_msgSend
That is, the objc_msgSend ended up in the Foundation function
_NSSetObjectValueAndNotify. I traced through the objc_msgSend, and it goes
straight to _NSSetObjectValueAndNotify from the method dispatch tables.
Just an interesting sidelight to this discussion. In the old days, you could
definitely see the secret subclass if you looked.
_______________________________________________
Cocoa-dev mailing list ([email protected])
Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com
Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com
This email sent to [email protected]