Does Microsoft Software Evolve Well?

Ralph Johnson comments on my piece about Microsoft's O/R mapping fiasco...

I think that Microsoft has perfected the notion that software has to evolve. They release early and often. The joke is that their software is never ready until version 3, and I've heard people from Microsoft tell that joke. I don't know the issues behind what you are saying about ADO.NET and Longhorn, but whatever the problem, it is not because Microsoft in general thinks that software will not evolve. When I talk with people from Microsoft, they seem much more aware than most that everything they do is transitory.

Maybe we'd have to discuss specific examples. I don't think Object Spaces is a good example of "release early and often". As I wrote previously, Microsoft is very late to the table with an O/R product. Even now that product is being pushed out, apparently to 2007 at the earliest.

Taking the subject of database connections more broadly, the Microsoft picture looks even worse. They have a history of introducing replacement connection technologies rather than finding a suitable abstraction that can evolve.

These various API's are transitory but they don't seem to be evolutionary. I won't fault Microsoft for packaging software and selling it successfully, and that includes getting software out early and improving it. But when it comes to software design, abstraction, and loose coupling of components, they seem to be getting worse rather than better.

Maybe that's the right approach: put everything into Longhorn, and force and upgrade. Or maybe it's just an accident of bad planning and design. Or maybe it's something else altogether. I don't know, but it doesn't seem evolutionary.