07 November 2007

Link: an analysis of Excel's "65535" bug

From Chris Lomont: An Analysis of the Excel 2007 "65535" bug. (This is the bug that made certain floating-point calculations with results near 65535 or 65536 appear as 100000 or 100001.)

Towards the end, Lomont writes:
[A]n amazing number of people guessed the bug had something to do with the 65536 row limit, showing the flaws in belief in numerology.

Um, it's a power of 2? Somehow it's not surprising that powers of 2 occur in a lot of distinct places with binary computers. (Hence the "correlation is not causation" tag on this entry -- the number of rows didn't cause the bug, nor vice versa, but both were caused by the fact that the internal architecture of the computer is binary.)

1 comment:

I. J. Kennedy said...

After reading the explanation of the problem, it was only "sort of" a binary bug. The programmers used FFFF as an end-of-table marker, and the bug centers around the test for that marker. Granted, FFFF is a common list sentinel because it's at the edge of the 16-bit universe, but another, non-binaryish value might have been used.