> Spectre shows us that the building blocks provided to us by Intel, ARM, and all the rest are no longer "small parts understood entirely"; that instead now we

> Spectre shows us that the building blocks provided to us by Intel, ARM, and all the rest are no longer "small parts understood entirely"; that instead now we have to do "basic science" on our CPUs and memory hierarchies to know what they do. I object to the term "no longer" in that sentence. There exist old CPUs that are not vulnerable to these attacks, but their existence doesn't mean that when those CPUs were in widespread use they were "understood entirely". I imagine that most programmers affected by the F00F vulnerability in Pentiums didn't fully understood what it was about the chip/microcode that caused the lockup, either. The parts being assembled change over time, but I think you can always say "bah, {kids|engineers|scientists|etc} nowadays don't know how the things they're plugging together work" and be trivially "correct" without making any constructive point about how to go back to some nonexistent past in which things were understood. Even when computers ran on vacuum tubes and physical insects in machines as "bugs" ([apocryphal](https://www.computerworld.com/article/2515435/app-development/moth-in-the-machine--debugging-the-origins-of--bug-.html https://www.computerworld.com/article/2515435/app-developmen...), I know), people didn't know the composition of the hardware. Would one of those programmers, on average, be more likely to fully understand, say the heat range required for successfully making circuits in a tube or the effects of anode corrosion on the lifespan of a tube, than a modern-day program would understand how CPU microcode works? The amount of complexity per part might go up over time. But there's always a specialized threshold for "low level" understanding below which a) most people don't operate and b) dragons lurk.