UITabBarController, MoreNavigationController and the Holy Grail of Device Rotation

Question

UPDATE: See my answer to this question first. This appears to be a bug. A minimal test case has been created and a report has been filed with Apple. (Fixed as of iPhone OS 3.1.)

Greetings! Here's a puzzler from the "I'm so close!" department.

I have a Tab Bar-based iPhone app. Each tab features a UINavigationController with the usual suspects (nav bar, table view ... which in turn can lead to another VC, etc.).

Now, one of those lower-level VCs is to be used in portait and landscape modes. But there's a problem. Our landscape-friendly VC's shouldAutorotateToInterfaceOrientation: won't get called out-of-the-box! What to do?

Here's what we do. In my Tab Bar Controller, which I have implemented in its own file, I have this:

- (BOOL)shouldAutorotateToInterfaceOrientation:(UIInterfaceOrientation)interfaceOrientation {
     return [self.selectedViewController shouldAutorotateToInterfaceOrientation:interfaceOrientation];
}

This ends up propagating the request to my landscape-friendly VC, which also responds to this message. All my other VCs don't implement this method, so they simply go with the default portrait orientation.

Problem solved!!! Yay!

Well, not quite. :(

Seems like things don't go so well when my landscape-friendly VC is invoked from within the depths of the tab bar controller's MoreNavigationController.

I decided to compare/contrast between a VC called from within one of the first four tab bar UINavigationControllers ... and that same VC called from within the MoreNavigationController. This is going to be a bit ultra-detailed, so bear with me. Hopefully the play by play proves useful for sleuthing things out.

When the app loads, there are several initial calls to the tab bar controller's shouldAutorotate... method. In these early cases, selectedViewController is nil. However, we eventually finish loading, the initial tab item is selected, and all is well.

Right. First, let's pick one of the first four tab bar items and drill down to our VC.

We'll pick third nav bar item, so that's the third nav controller. We drill down to our VC that supports rotation. A quick inspection confirms that the parent is indeed the third nav controller from our tab bar's view controller list. Good!

Let's rotate the device. The tab bar controller is asked to autorotate (see the above code). We observe that selectedViewController is also that third nav controller, plus the nav controller's top and visible view controllers are both set to our trusty VC that supports rotation.

Thus, the tab bar controller will forward the shouldAutorotate message over to the third nav controller ... but our rotation-friendly VC ultimately gets the message. (I'm not doing anything special here. Maybe the desired VC gets the message because it's the top and/or visible VC?) In any event, we rotate to landscape, things are resized, and all is well. "Great Success!"

Now let's hit that back button and pop the VC stack, leaving landscape mode in the process. The tab bar controller is queried again.

Time for a little aside here. The topViewController for our nav controller is still that rotation-friendly VC, but visibleViewController is now set to UISnapshotModalViewController! Heh. Never saw this one before ... but Erica Sadun has. Looks like it's for "disappearing view controllers" (which in this case is certainly true - it's disappearing alright).

As I keep stepping through, the visible VC stays as Snapshot, but the top VC eventually changes to the next VC on the stack, since my special VC is eventually gone. Fair enough.

So that's the scenario where everything works well.

Now let's try the same test, only this time we're going to go to the MoreNavigationController (the More tab bar item) and drill down to the same VC class as before. In my case it happens to be the 7th one in the tab bar controller's VC list.

We enter the rotation-aware VC and ... this time it gets asked to rotate directly! The Tab Bar Controller is not asked for permission to rotate at all. Hmm.

A quick check of the parent VC shows it is a MoreNavigationController. OK, that makes sense.

Now let's try to rotate the device. NOTHING GETS CALLED. None of our breakpoints get hit. Not in our VC. Not in our tab bar controller. (Huh?!?!)

O-kaaaay. Let's pop the stack, go back into the same VC and try to rotate again. Weird. NOW we get a call in the Tab Bar Controller asking for autorotation permission. Here, the selected Controller is our trusty Nav controller (#7), but this time its visibleViewController and topViewController are SET TO NIL!

Once we continue from here, a mysterious message appears in the debugger console:

Using two-stage rotation animation. To use the smoother single-stage animation, this application must remove two-stage method implementations.

Mysterious because I'm not using two-stage rotation animation! No SecondHalf method variants are in play anywhere in my source code.

Alas, my rotation-aware VC is never told that rotation is occurring (even though rotation does occur on-screen), so of course my view is all fouled up as a result. Mayhem and sadness ensue. :(

We won't even bother popping the stack at this point.

I think the View Controller doc hints at the possible problem:

If you want to perform custom animations during an orientation change, you can do so in one of two ways. Orientation changes used to occur in two steps, with notifications occurring at the beginning, middle, and end points of the rotation. However, in iPhone OS 3.0, support was added for performing orientation changes in one step. Using a one-step orientation change tends to be faster than the older two-step process and is generally recommended for any new code.

I wonder if MoreNavigationController is still responding to the two-step process, and is thus tripping up any attempts to use the one-step process? Note that, if you respond to the two-step messages, the one-step variant will not work (again, per the docs). I'm not responding to them, but I have a sneaking suspicion something behind-the-scenes IS.

In fact, if I comment out the single-step method, and try to respond to willAnimateSecondHalfOfRotationFromInterfaceOrientation:duration:, I do get the memo! But it still doesn't pop off the stack very cleanly (in terms of visuals). Even stranger: willAnimateFirstHalfOfRotationFromInterfaceOrientation:duration: is NOT called, even when I tried to sneak a call to self (using the FirstHalf message) in shouldAutorotateToInterfaceOrientation:. It returns immediately during a trace, as if I never even defined it. Sigh.

So that's the play-by-play.

In summary, has anyone successfully handled one-step device rotation for a VC invoked from within a Tab Bar Controller's MoreNavigationController?? Inquiring minds want to know! Thanks for reading.

Answer 1

It appears we have a bug. I have created a reproducible, minimal test case, and reported it via Apple Bug Reporter (RADAR Problem 7139857).

Update: This has been fixed as of iPhone OS 3.1.

The essential problem is:

View controllers already on a Navigation Controller stack do not receive willAnimateRotationToInterfaceOrientation:duration: messages when a Tab Bar Controller's 'More Navigation Controller' is in use.

This problem does not occur when the tab bar item view controllers are basic view controllers. Only when they are navigation controllers and only when the "More" navigation hierarchy is in use.

The console message (regarding two-stage rotation animation) suggests that something within the framework (the More Navigation Controller?) is still using a two-stage animation, even though single stage is now recommended as of iPhone OS 3.0.

That could explain why willAnimateRotationToInterfaceOrientation: is not called in that particular case. Per Apple's view controler documentation, this message will NOT be invoked when two-stage, first/second-half orientation messages are being responded to instead.

Answer 2

I have a tabbar application and a UIWebView on one of the tabs. I want to change the orientation from portrait to landscape. Is there a way to set the orientation mode to landscape?

Thanks Zeeshan

Answer 3

Apple advises against subclassing UITabBarController, so I found an easy way to handle autorotation using categories instead. It doesn't fix your bug with the More... view controllers, but I think it's a more Apple-friendly way of getting the job done (and means less subclassing for you).

To make every tab in my application autorotate properly, I've defined -shouldAutorotateToInterfaceOrientation: in my custom view controllers, but they are all inside UINavigationControllers within a UITabBarController, so the message won't get sent down the chain to my VC until those two also respond. So I added the following lines to my app delegate files:

Added to the bottom of MyAppDelegate.h

@interface UITabBarController (MyApp)
@end

@interface UINavigationController (MyApp)
@end

Added to the bottom of MyAppDelegate.m

@implementation UITabBarController (MyApp) 
- (BOOL)shouldAutorotateToInterfaceOrientation:(UIInterfaceOrientation)toInterfaceOrientation {
    return YES;
}
@end

@implementation UINavigationController (MyApp) 
- (BOOL)shouldAutorotateToInterfaceOrientation:(UIInterfaceOrientation)toInterfaceOrientation {
    return YES;
}
@end

Answer 4

Are you sure this is fixed in 3.1? I am having exactly the same issue as you describe and I can't get it working in 3.1 or 3.1.3 :-(