Disaster recovery planning is famously difficult, and not just becauseit's a hard subject. In fact there seems to be a kind of repulsion fieldthat makes people either shy away from thinking about it or dismiss itas something that they can't do anything about anyways. (The latter isthe 'if we have a serious fire the organization's dead anyways, so whyworry about the backups?' mindset.)

I've come to think that one significant reason for this is that when you

On Bots is a fascinating reporton a large scale experiment to see how web search bots would explorean almost limitless set of linked pages. Whether or not the resultsgeneralize (or are still applicable), it's got a bunch of prettypictures.

(From Tim Bray ,rather belatedly.)

I have a small defect in the Python code I produce: I rarely botherto make my classes printable or to give them a repr(). Most of theclasses will never be printed, and the default repr value is goodenough to distinguish two instances from each other.

But this is a mistake, nicely illustrated by my grump about assert'sweakness as a debugging tool . Objects having a usefulstring value makes it much easier to dump out information about thestate of things when a problem comes up.

A fundamental problem with CR is that it implicitly assumes that yourcorrespondent wants you to read their email more than you do. Just lookat who's doing the work: your correspondents are doing work so that youdon't have to.

(Of course, observe that the people for whom this is the most true are the spammers.)

When this is not true, when you want to read the mail more than thesender wants you to, is exactly where CR systems break down. You mightthink