osx-app test is flaky because it is not hermetic w.r.t. software updates

[gregory-marton]

For the past day, and for many hours or days this month, pcg-osx-test has required an update of one sort or another. As soon as it displays needing an update, it prevents other clicks, including the simulated ones needed for testing, and so tests fail. When there was a genuine failure a week or two ago, it took me a long time to find/diagnose it because I had become so accustomed to its failing for reasons not to do with the code.

For now, I've added two entries to /etc/hosts in an attempt to get apple to stop updating us: 127.0.0.1 itunes.apple.com #where software updates actually come from. 127.0.0.1 apple.com

This is overly narrow. In principle there's very little reason for pcg-osx-test to touch anything outside of MIT, except perhaps for using mathjax, though indeed, there is reason to cache even that.

But mostly, I want to both socialize that temporary decision, and ask for feedback, either on better ways of accomplishing what I want, or at least on a better place to document the decision and its implementation, so that someone who actually wants to update the machine will not have quite as hard a time doing so.