Blue Lotus Blog

I have spent quite a few hours with the iPhone SDK now and must say that I am quite impressed. The experience of bringing some of my code written for Mac OS X over to the iPhone with no changes and having it just work is quite a pleasant experience. I REALLY enjoy being able to use Objective-C 2.0 synthsized properties and fast enumerators in my code (garbage collection would have been nice but hey, this is a mobile device). I maintain Tiger support for Language Aid and develop purely for Leopard in Razor and it is kind of an interesting mix of both to code for the iPhone. I am really excited to to flesh out a game I have been thinking of for the iPhone that would take advantage of a lot of its unique features.

So the posted requirements say that this beta version of the SDK requires an Intel Mac but so far I have not run in to much trouble developing with it on my G5 (the only Intel mac in the house right now is a Core Solo mini which is already pulling duty as a source server, build server, DVR and media center). Inspecting the simulator and the xcode plugins show that they are all universal binaries. The only gotcha that I ran into is that I had to make sure that in my build settings that the “Architectures” setting is set to “Native Architecture of Build Machine” so that the binaries produced are ppc so that the simulator (I have heard a surprising number of people errantly referring to it as an emulator) can load and run them. For some reason this reset itself to i386 one time and I had to track it down to that setting.

Other small quirks that may or may not be related to me running on ppc are that it seems like every 10th click event or so on the simulator window seems to fall through and hit the window behind it. Also, for some reason the simulator is also not responding to the “Rotate” commands whereas it would correctly respond on my Intel mini. Of the example code on the developer website there are a lot of projects that will not run in the simulator due to a pragma that intentionally errors itself out if built targeting the simulator:

#if TARGET_ASPEN_SIMULATOR
#error This application cannot be run in the Aspen Simulator. Please change Active SDK to “Device” and ensure a device is connected.
#endif

Simply build the sample using Xcode and run it on the device (OpenGL ES is not supported in the simulator).

It is inevitable that I will plunk down the $99 to be able to perform remote debugging/use Instruments/use Shark with a real iPhone. However, for now while I am busy bringing up the core game engine and getting used to the new platform, the simulator is doing just fine.

Launch Services is one of those things that does a lot of magic for Mac OS X but is not really visible on the surface and is somewhat vague and ethereal. Launch Services is responsible among other things, for associating file types with applications, associating icons, associating system services and many other wonderful things that make your experience just work. Fortunately there is a tool to poke at the Launch Services database buried in the system at:

Most notably the -dump command basically outputs the database so that you can see exactly why a certain file type wants to be opened by which app and why certain files get their icons and association hijacked by some other random new app. The -f command will let you update the database with any new changes you have made to an application or bundle.

Some of things managed by this database are the system services that show up in the “Services” option under the application menu (these are outlined in the NSServices array of a bundle’s Info.plist). Language Aid provides one of these as an alternative to the lookup trigger and contextual menu as a way to lookup text. Unfortunately, it appears that just tickling the database with the new information about which options you want available in the menu (dictated by which plugins you have selected), is not enough to rebuild the Services menu. A reboot seems to do the trick, but it would really be nice if I could figure out what causes that menu to rebuild.

Ugh, this bug was so nasty to track down… Apparently if you call this family of methods a couple of times it screws up whether NSBundle can find resources in the future. My specific situation is that I would call this method to find what bundles are in a directory. I would then run it one more time (because that is how the code is laid out). Then when kqueue tells me that something has changed in that directory (because I added a new bundle to the target directory) I call it one more time. It finds the new bundle just fine, but for some reason it completely ignores the stuff in Resources/ when trying to query the contents of said new bundle! I don’t know if this is the result of some errant caching or what but it is consistently reproducible. The old bundles work fine but there is nothing that can make Cocoa see the contents of this new bundle. This is a big problem because it means that [NSBundle loadNibNamed:owner:] now can’t find any nibs in the new bundle (which was what first tipped me off to the problem). In the current Developer Tools Documentation under the entry for [NSBundle pathsForResourcesOfType:inDirectory:] there is a statement:

Note: This method is best suited only for the occasional retrieval of resource files. In most cases where you need to retrieve bundle resources, it is preferable to use the NSBundle instance methods instead.

I hope that is not implying this weird (caching?) behavior. Also, after the damage has been done NSBundle instance methods no longer work either. The solution to this has been to instead use a UNIXy way like scandir() to find out what bundles are in a directory. Not nearly as graceful, but at least I don’t have weird peripheral behavior. So anyway…fair warning to all against using these routines.