A Day of Programming

Sometimes days are like this:

Our testers — who are great — report a regression I caused but that I didn’t notice. Why is sub-pixel anti-aliasing not working in this view? The text looks bad. (Note to self: test with a non-retina display before comitting.)

I investigate, and find, to nobody’s surprise, that the culprit is a layer-backed ancestor scroll view. I turn off layer-backing — I just uncheck that particular box.

Then run the app. It’s good to see sub-pixel anti-aliasing back. But look at how things have gone wonky.

Wonky? Let’s be precise: the layout inside outline view cells is incorrect. If I check the box (turn layer-backing back on), then layout is fixed. Uncheck it, and layout is broken.

(This all sounds straightforward, but I’m glossing over a bunch of things I tried and researched. The above is a couple hours of work.)

The wonky view uses auto layout, but it’s not entirely constraint-driven: there’s a manual -layout method. When an ancestor is layer-backed, then that method is called at the appropriate times. When it’s not, then that method isn’t called often enough, or not at all the right times.

Or… but it also looks like something else might be moving those views, giving them a zero origin when it shouldn’t be. Exactly as if it’s doing constraint-based layout and not calling the -layout method. Sometimes.

I click around a while, add NSLogs and breakpoints and all the usual things. I can’t figure out what’s happening. I bring in resizeSubviewsWithOldSize: (the old friend). I accidentally introduce an infinite loop at one point.

To reassure myself, I check that checkbox again (turning on layer-backing) — and layout works again as expected.

Time to get pragmatic.

If it seems as if those views are sometimes getting laid-out by their constraints, and -layout isn’t getting called, then maybe that’s really what’s happening. So one possible solution is to give all the views their necessary constraints, so that -layout isn’t needed at all.

I tried it. Success! (I think. At least so far.)

And that was a day of work. In the end I don’t know why there was a problem, but I have a solution. I should probably file a Radar, but it seems like exactly the kind of thing that will be hard to duplicate in a simple case (though I could be wrong).

There’s nothing wrong with days like that. In the end the thing is solvable. What worries me is when I have too many days like that strung together, when I have too many Radars to file.

16 Oct 2014

Archive