Lack of testing 'threatening stability of Linux'
This is a bit strong, but it never fails to amaze me the difficulty of doing things with computers that are seemingly really easy with real life.
You want to build a bridge (real life edition)? Pile some stone up. Measure the sheer, tension, torsion strength of the stone. Now you know how much the bridge can hold.
You want to build a bridge (computer program edition)? Pile some stone up. Figure out some automated ways to test the sheer, tension, torsion strength of the stone. Now you know absolutely nothing, because the second someone gets that bridge in real life you're going to have people driving boats into the pillars and throwing 65 million ton trucks on it and hitting it with tidal waves and getting pissed when it crashes because they weren't thinking of using it in the same way you were using it. Or, better yet, somebody will sit around and figure out the exact resonance frequency necessary to bring the thing down and then publish that. Or, even better, totally randomly someone will drive a car, the water will swirl and somebody 20 miles away will pound a jackhammer into the ground and, somehow, all these things will come together to cause a massive crack with no ability to predict it whatsoever.
It's my guess that it's this last thing that the Linux community will have the hardest time overcoming.
