the importance of the timer rollover bug in Win9x

Warren Young
Thu Sep 11 20:18:00 GMT 2008

Matthew Woehlke wrote:
> Oy, hadn't run across that before :-).

The fact that the problem wasn't fixed until 2000 or so made this more 
than just an embarrassment for Microsoft.  It was an inflection point.

What you had was a bug that was absolutely deterministic, which affected 
hundreds of millions of machines over many years.  Multiply it out and 
you come to something like 100 billion times the bug could have 
happened.  Sounds like a programmer's dream, right?  A bug you can count 
on to happen that reliably with such a huge installed base....yet it 
took ~5 years to diagnose and fix.

For such a bug to last so long, you're looking for probability of 
discovery down around 1 in 10 million.  It takes a lot of explaining to 
get from 1e11 to 1e-7.  I tried.  The "good reasons" got me down to 
about 1e2.  Maybe you can get down to 1e0.  You're still left with so 
many zeroes as to constitute objective evidence that Win9x boxes 
experience...erm, unscheduled restarts...*a lot*.

To this point, you had all kinds of anecdotal evidence of Win9x's 
instability.  The arguments raged on, as those based only on anecdotal 
evidence will.  This incident provided objective proof of the sort you 
don't see ignored outside of politics and religion.

More information about the Cygwin-talk mailing list