On Aug 19, 11:04 am, Steven D'Aprano <st...@remove-this- cybersource.com.au> wrote: > On Tue, 17 Aug 2010 19:15:54 -0700, Russ P. wrote: > > The convention of starting with zero may have had some slight > > performance advantage in the early days of computing, but the huge > > potential for error that it introduced made it a poor choice in the long > > run, at least for high-level languages. > > People keep saying this, but it's actually the opposite. Signpost errors > and off-by-one errors are more common in languages that count from one. > > A simple example: Using zero-based indexing, suppose you want to indent > the string "spam" so it starts at column 4. How many spaces to you > prepend? > > 0123456789 > spam > > Answer: 4. Nice and easy and almost impossible to get wrong. To indent to > position n, prepend n spaces. > > Now consider one-based indexing, where the string starts at column 5: > > 1234567890 > spam > > Answer: 5-1 = 4. People are remarkably bad at remembering to subtract the > 1, hence the off-by-one errors. > > Zero-based counting doesn't entirely eliminate off-by-one errors, but the > combination of that plus half-open on the right intervals reduces them as > much as possible. > > The intuitive one-based closed interval notation used in many natural > languages is terrible for encouraging off-by-one errors. Quick: how many > days are there between Friday 20th September and Friday 27th September > inclusive? If you said seven, you fail.
The error mode you refer to is much less common than the typical off- by-one error mode. In the far more common error mode, zero-based indexing is far more error prone. > One-based counting is the product of human intuition. Zero-based counting > is the product of human reason. I suggest you take that up with mathematicians, who have used one- based indexing all along. That's why it was used in Fortran and Matlab, among other more mathematical and numerically oriented and languages. -- http://mail.python.org/mailman/listinfo/python-list
