Last week I visited Micosoft Reasearch in Cambridge, where I gave a talk on SAT solvers, Grobner basis algorithms, and Cryptography. It was nice to visit Microsoft, and in particular I enjoyed seeing how “the cathedral” works. The kind of software development carried out at Microsoft and almost all large software companies is called the cathedral model due to the book “The Mythical Man-Month“. It’s a very interesting book that talks about how software (and in particular, a “Programming Systems Product”) is developed. All Windows variants, and e.g. the IBM System/360 fall into the category of programming systems products: they are all tested, documented, extensible, have predefined inputs and outputs, come with predefined interfaces, etc. The development of these software is called the cathedral model simply because that’s the example the book uses, and it got stuck in the literature.
I have been developing CryptoMiniSat for quite some while now, and I must say it is exceedingly difficult to create a programming systems product — CryptoMiniSat doesn’t even scratch the surface of the requirements, but then almost no SAT solver does — maybe SAT4J is the sole exception. Documentation is usually scarce, and even though I have added a lot, it’s still not enough. The input is rarely checked for sanity, and resource needs are usually just ignored: if memory is not enough, then the program will probably just crash; don’t expect graceful degradation.
As for Microsoft, I must say I was pretty impressed. It’s a nice building, with large offices, free tea/coffee/orange juice, filled with some very kind and very bright people. The audience was astonishingly good too, asking questions when they felt like they haven’t understood something, instead of just getting disinterested and e.g. browsing email. All in all, it was very interesting to visit the place, though I still feel more inclined to release my code as Free Software ;)