Good week at work

[dmaze]

Funny how, when you spend 40 hours a week at work, progress there affects your mood. For a change, I haven't been on multi-week soul-sucking hardware race condition bug hunting...I've been getting to do forward development, and some bug fixing, but the bug fixing is mostly of the form "huh, gdb says things crash here, what if I assign something besides NULL to this pointer?" Then Tuesday was the day of "it doesn't crash when I post a single XML file to it", Wednesday the day of "test driver A doesn't instantly explode", Thursday "test driver B works too, plus I can run this 2,000 test suite without dying from memory leaks". Getting to make lots of visible forward progress for little pain or effort is actually pretty satisfying.

Comments

[kvarko.livejournal.com]

Hardware race condition bug hunting?

[dmaze]

A typical bug I've had to deal with in the past year involves an interaction between a "host system" and a "coprocessor". There's some shared memory, and the program running on the host system can indicate to its driver that it's done with the shared memory and the coprocessor can reuse it. That's all fine until there's a bug in the program where a pointer into the shared memory is left around, and everything works fine until you're putting enough load on the combined system that the memory is actually getting reused by the coprocessor before the host system is actually done with it.