Hector Martin

You'd think that having such a basic feature broken would cause most software to crash and burn, but actually the game I was testing with, Wind Waker, booted fine. Ish.

It got to the title screen, but instead of loading the title screen cinematic behind the logo, it loaded the first map, with a playable Link. You could move around, but the Y axis of the stick was "folded" so you could only ever move backwards, not forwards.

So yeah, software damage is far from guaranteed to segfault. This is how you get really odd "undebuggable" problems from memory corruption. And how Intel gets away with still crippling desktop CPUs by disabling ECC RAM support.

Your desktop probably flips a few bits in RAM weekly, but you just don't notice. RAM sizes are too huge for memory to be close to 100% reliable. It just isn't possible without error correction support.

Interesting story, but I don't think this is true.

Result from a large scale study on memory error rates: “an average of one single-bit-error every 14 to 40 hours per Gigabit of DRAM.”

We would see that right away with pointer corruption on architectures that don't allow unaligned access though?

How many pointers in a given system, as a fraction of total memory? How many random unsolved app hangs or crashes?