Gary King on testing

Riffing on Drew McDermott's essay about the drawbacks of Literate Programming:

I find this happening to me about three-quarters of the time that I try test-first development. I don't know exactly what I'm doing (cf. Paul Graham's wonderful introduction to On Lisp or ANSI Standard Lisp (I can't remember which one)) and writing tests often turns out to be silly. Perhaps I'm just not getting out or can't rid myself of other bad habits or perhaps test first isn't something that makes sense all of the time.

I've noticed this — for an infrastructural component, like a serialiser, it makes sense to have thorough tests, and even to write the tests first. For an application doing exploration in the space above that component, writing tests is terribly premature; anything apart from very quickly throwing together code, in fact, is usually premature, because you don't even know which problem you're solving. The least-credible testing arena is that of output; imagine testing the XHTML output from a page component function that is constantly in design flux. It's a spectrum, I suppose.

Posted at 2006-05-13 14:58:08 by Richard • Link to Gary King on testi…
Comments, trackbacks.