To address some performance issues, I started recycling some view controllers. However, the performance benefit of re-using a recycled controller's view is only present if that view has been drawn. If, for, example, I want to pre-populate the recycle queue with a controller but never place its view on screen, I get no such benefit.
How can I force the controller's view to be 'pre-rendered' and added to my queue such that when it is recycled I receive the performance benefit I am seeing from my other recycled controllers? I know that a controllers's view is created when first needed, but even adding the view and immediately removing it (before the parent view is displayed) doesn't seem to do it.
My first thought on a quick and dirty way to do this is to just set the controller's view to hidden, and still have it "drawn" to the screen.
That bring said, I'm curious as to what the "performance issues" are that you're observing. Depending on whether the creation of the UIViewController itself, or the UIViews that are causing a problem, then there may be a better overall solution to the problem.
Related
Suppose I have 2 views. In the first view, I allocate memory to displaying many UI components such as an UILabel, UIImages, etc.
Suppose the user navigates to the next view (via UINavigationController)
Is it OK to deallocate memory assigned to displaying UI components in the first view and then initialize them again once the user goes back to the first view (in viewFirstLoad or the appropriate function)?
It seems to me if you don't do this, then memory will keep on increasing the longer the user uses your app in that particular session.
Is this not allowed? frowned upon? impossible?
It is perfectly normal and in fact, that functionality is built in standard UIViewController - when controller is not displayed its view may be released from memory and you can release all its subviews (e.g. retained through IBOutlet references) in controller's -viewDidUnload method.
When controller needs to display again it reloads its view again.
It depends. Generally, the rule of thumb is that you should free objects that you don't need. If your view is just a view, then yes, I'd release it and all of its subviews. If your view has data that was obtained through a lengthy retrieval process (e.g. a web service call), I'd probably hold onto the data somewhere so that I don't have to go back out and retrieve it when the user goes back to the first view.
To clarify a little: Apple recommends you display data specific to a view in it's -viewDidLoad method, such as setting text on labels. Then, in -viewDidUnload you should release (or nil outlets of) the view objects you setup in -viewDidLoad. It's critical you implement -viewDidLoad, as the base UIViewController code checks that it's subclass actually implements -viewDidLoad before it assumes it can unload the view (and therefore call -viewDidUnload). Failing to implement -viewDidLoad results in the controller thinking it can't recreate your view at a later time, and so it doesn't unload the view from memory. A developer I know experienced this same problem, took forever to track down.
I want to hear developers opinions on the best way to swap views on the iphone.
For example, I have a tab bar and one of its tabs defaults to a login view. When The user logs in the view changes to a logged in view.
I was going to just use one view controller and have all the content in one xib hiding and showing content as needed but this seems in no way elegant.
Secondly I was considering having one viewcontroller and simply swapping the xib. I'm a litle reluctant to try this as I've read in an article or 2 that it can lead to memory leaks.
Finally I was considering using 2 view controllers with with 2 seperate xibs. My gut tells me this would probably be the "proper" solution but I so far have failed to hunt down any sample code on the correct way to do it.
Can you offer advice on the best way to solve this problem?
Is there a technique that I have not listed?
Thanks.
I would keep the logic for which view to show in the view controller. The XIB is the view itself, and should have no objects in it that are transient or not always visible for that particular view.
Your second approach (of swapping the views) seems to be the right approach to me, and is always something I, personally, do in these situations. I am not aware of any memory issues if you do it right (remove from superview, followed by loading the new view as a subview of the controller's view). You could perform any custom initialization once the new XIB has been loaded and before showing it to the user.
Multiple view controllers just seems superfluous as then you would ideally require another top level controller to manage the two view controllers.
I'm getting strange navigation bar behaviour, for example when I hit back button the screen displayed is the previous screen, however the Navigation Bar items do change. So I'm left with screen A, but with nav bar buttons for screen B.
Could this be due to memory leaks? I do note with my app still:
This behavior seems to happen:
immediately if I trigger memory
warning via the simulator menu, or
on a device after it has been on
for a while [without being killed
and then restarted as an app].
I do have some memory leaks I'm
trying to clean up (i.e. Profiler
highlights items in "leaked blocks"
section)
Any tips on fault finding root cause of why pushing a back button would end up in a weird state? e.g. screen on previous parent view, but nav bar items don't change...
UPDATE - I have finally removed the memory leaks in my app, however I note the nav bar issue still remains. I guess this doesn't confirm the answer to my question is NO in general, but in my specific case it wasn't the memork leak...
From Apple:
The navigation controller updates the
navigation bar each time the top view
controller changes. Thus, these
changes occur each time a view
controller is pushed onto the stack or
popped from it. When you animate a
push or pop operation, the navigation
controller similarly animates the
change in navigation bar content.
Based on this, I would start by looking for a bug or misconfiguration in your view definitions. Check for any InterfaceBuilder warnings if you defined your views via NIBs. Make sure your view hierarchies are correct in both UIViewControllers. Also check for possible bugs in your view life-cycle methods: viewWillAppear:, viewWillDisappear:, etc,.
Actually, it would be nice if you could post some screenshots and/or code. Thanks!
Any view that is not currently visible and only retained by it's view controller (as a part of the view property of that view controller) will be released (along with any non retained subviews) when a memory warning occurs.
Chances are you are creating the view as a part of init, and not retaining it in the controller (simply letting the view socket hold it from releasing). One way around this is to create properties for the views you create (nonatomic, retain), and after creating them and autoreleasing, assign to those properties, don't forget to assign those properties to nil as a part of dealloc to avoid leaking. Another way is to create your custom view elements in viewDidLoad as opposed to init.
Hard to say without code example from the offending views :)
I've seen something like this happen after calling -[UINavigationController setViewControllers:]. You might try not doing any programatic manipulation of the navigation controller's view controller beyond calling -[UINavigationController pushViewController:animated:].
The root question is "how many UIViewControllers can you push on the navigation stack?" without causing memory warnings or earning a watchdog termination.
Suppose I have an application that is basically a database for three entities where each can have a relationship with some other entity, and the relationship is shown on a UIViewController. Users can follow those relationships and each one brings up a new controller - If the entities are A, B and C and A->B->C->B->C->A then each kind of view is on the stack twice. I understand how to push and pop, how to push back to a particular controller, and I think rather than just extend the navigation stack indefinitely it might be best to reuse a view controller in the navigation stack.
To do this, every time I wanted a FirstEntityViewController I could scan the navigation stack to find an object where [self isKindOfClass:[FirstEntityViewController class]]; and then call methods designed to rejig that view for what I currently want to see - just refreshing the data in the same way you do when reusing a UITableViewCell.
This is fine except for the effect it might have on the NavigationController. If I use UINavigationController:popToViewController:animated: I think it is going to discard everything above the view I'm popping to, including the view which the user expects to find on tapping "Back" in the navigation bar. So user taps a relationship, taps back and goes "huh?"
If I remove the matching controller from the navigation stack and then pop it on to the top of the stack the back behavior remains OK as long as the user doesn't go back as far as the instance of FirstEntityViewController that was moved or else again navigation will seem inconsistent.
Is the right solution to remove the controller from the stack, and somehow hold a place in the stack so that when the reused controller is popped it can be replaced back where it came from? Should I maintain my own list of view kind and data display so that when popping I can replace the view below the view about to pop to, staying one step ahead of back navigation?
Or is that just getting too complicated? Is there no need to even worry about this situation because the OS reuses much of the view controllers in the same way as UITableViewCells are reused, and there is no real memory or performance impact in having a 50-deep navigation stack?
ViewController instances remain in the UINavigationController's stack, but any view except for the top view may be unloaded at any time (the view controller is notified via the viewDidUnload message).
In other words, the views underneath the top view do not hang around and will eventually be unloaded in low-memory conditions, so there's no need for you to attempt to re-use your view controllers.
Last I checked you can't push a viewcontroller that's already on a navcontroller stack back onto it again. You'll have to create a fresh viewcontroller and push it onto the stack and each back button will pop that one off the stack. The best you can do is make a cache of viewcontrollers and dole them out as-needed -- as long as they're popped off the navcontroller stack. But it probably won't buy you much by way of memory savings.
UITableViews are a bit different in that there's only a relatively small number of cells in view at any given time and as soon as the cell goes offscreen it's removed and returned back into the pool. If you can guarantee that the maxdepth of your chain is fixed, then you can pull a similar windowing scheme. If not, you may have to stick with going deep and be vigilant about releasing memory as soon as you can.
At the beginning of my app, I am initializing a UIViewController with a NIB.
At some point, I am removing the view from the screen, then releasing it, so it no longer takes up memory. (At this point I am also saving the state of the view in the UIViewController)
At a later point, I want to recreate the view and bring it back on screen. I will then restore its state. (Using the same UIViewController, not a new one, since it saved the state)
My question, is when I recreate the view, how do I do so from the NIB, or is this not possible?
To me, the obvious remedies are:
Don't save state in the UIViewcontroller (Where does convention dictate that I do save state?)
Don't release the view (maybe just release all of its subviews?)
Don't load my view from a NIB, create programmatically (seems to go against using IB for everything)
Note:
I am not using a UINavigationController, I am handling the swapping of the views myself, since thee are only 2 of them.
If you need that level of control, you dont really want to use IB. Because, as you noted, if remove the view and then later recreate it, you would have to do it in code then anyway. Just design most of the views in IB, then write some code that generates just this view. Then you can call that same method again later to recreate that view when you need it.
You may be able to archive it and later turn it back into an object, but that seems like an inelgant solution. IB does not allow for dynamic creation of controls at runtime, even if they used to exist but don't anymore. There is no shame in leaving IB out of loop for this. In fact it's probably a good idea.
OR
If its a complicated view with a lot of pieces, put the view in it's own nib, and make a view controller for it. Then you can simply instatiate the view controller with the nib name, and add the controllers view as a subview to you main view. Then your view controller handles loading of the nib, and you get to design it in IB. Nothing says the view of a view controller has to take up the entire screen either.
self.otherController = [[OtherController alloc] initWithNibName:#"Other" bundle:nil];
[self.view addSubview:otherController.view];
Don't create the view controller and the view in the same nib file; either create the controller in code or put the view in a separate nib file. If you later nil out your controller's view property, the controller should recreate the view.
But why are you so worried about this? There's already a mechanism to automatically free up the view if you're low on memory: the low memory warning. It will only destroy and recreate the view if you actually need to do so, and it's built in to the system so you don't have to do anything. Remember the saying about premature optimization?