[hinkley]

My CI/CD pipelines have always been a minmaxed affair of doing the math on how useful a piece of information is and how expensive it is to get.

As such, it's not unusual for the OS X tests to be farther down the list and not to trigger at all if earlier stages failed. It's easier to scale up multiple subprojects if you run the first couple phases on commodity hardware and then ramp up to the more peculiar stuff only if everything else already looks good. That way, your broken build can't slow down my green build very much.

It's not so much that I like doing this, as that it prevents a number of things I most definitely do not like at all.