NetNewsWire for iOS, currently in TestFlight beta, was getting a lot of crashes related to background refreshing.
They would happen when the user wasn’t actually using NetNewsWire — it happened when the app would download feeds and sync in the background.
The crash logs were not identical, but they had this same thing:
Namespace RUNNINGBOARD, Code 0xdead10cc
This meant that the system was killing the app because, after the background task was complete, the app still had references to a SQLite database (or sometimes another file).
We did everything we could to fix this over a course of several months, including various forms of major surgery — and the best we could do was make it worse.
I did a lot of research — including reading this blog post from The Iconfactory several times — and we still came up with nothing.
Finally I asked around.
Marco had already been through all this with Overcast, and he gave me this advice:
Don’t keep the SQLite database in the shared container. You’ll never get rid of all of those crashes. Instead, communicate with extensions via other means than having them read/write the DB directly, such as Darwin notifications or writing plist files in the shared container.
iOS will always terminate the app and generate that crashlog whenever your app or extension has an open file handle to a file in a shared container at suspension time.
And there are some cases where your app gets forcibly suspended without calling background-task completion handlers. (Especially in extensions, where they don’t exist.)
I tried wrapping every SQLite query in a background task once to avoid this. A standard Overcast session may issue hundreds or thousands of database queries. I later found that apparently each one generates a process power assertion, the OS wasn’t made for that level of usage, and after some time, Springboard would crash.
There’s also the NSProcessInfo background-task thing that allegedly works in extensions, except that it doesn’t.
The moral of the story ended up being: just don’t keep your SQLite database in the shared group container. There’s no way to avoid these crashes 100% of the time.
We were sharing our SQLite database with one of our extensions! Marco’s advice is, basically, don’t do that.
So we stopped doing that. We un-shared the databases and switched to a super-low-tech thing for communicating between the extension and the app (the extension writes to a plist file which the app later reads).
And — as of last night’s build — the dreaded
0xdead10cc crashes are gone!
That was the last big challenge on our list for shipping. We still have work to do, but we’re getting close — the milestone says we have just four bugs left. Good deal.
PS I’ve since heard from other developers: don’t-share-the-database appears to be the common wisdom, which it just took me longer to learn. 🐥