[mark-r]

I once fixed a bug caused by a faulty printer driver from HP. The driver changed the floating point control word and didn't change it back. Our program crashed while doing a completely innocuous operation, but only if you had printed to an HP printer earlier in the session.

[mikestew]

I remember that, because I worked on the MS FoxPro team at the time and report printing was crashing for a lot of users, but not all of them. Took forever to finally pin it down because, as parent comment points out, the culprit could have long left the building by the time the crash happens. Stupid driver sets the FPU to say "math errors like divide-by-zero are software's problem now, not mine" without telling software. IIRC (and this was over 20 years ago), operations had to be wrapped in/with the one line of code that flipped it back.

But here's the thing: it wasn't just HP, it was a lot of print drivers. I suspected that there was some printer driver boilerplate out there, possibly even published by MS, that included this bug.

(And, wow, did I swerve sharply into the off-topic lane for story time. Sorry.)

[mark-r]

That's funny, we had a very diverse user base but never saw the problem with any printer but HP. And 20 years ago sounds about right for my time frame too. I think that was about the point where Windows started providing more isolation between drivers and user programs.