I agree for the home-managed code (what scaries me is that I was sinking about this issue yesterday night and you post it today, are a brain spy or something ?).

But what about reusable framworks ?
I think it is a shame that HashSet[?] and HashMap[?] aren't pluggable like TreeSet[?] and TreeMap[?] are in java for example, this is unanticipated reuse. Or there is no other way to format tooltips other than HTML in swing.
Do you have good guidelines for this ? The best is having a (smart) user at home to feed us with requirements, but this is not always the case.

Ideally, frameworks should emerge as you eliminate duplication in client applications (you should start without a framework, and refactor to it), and framework APIs should not be solidified until they have been put to extensive use. You won't get by with a smart user, you need many users trying to do different things with the framework.

This will expose the real usage patterns of the API, not just the usage patterns you anticipate. It is more problematic to include features which will never be used than to not include features which will be needed. After all, it's a lot easier to add features to an API than to remove features from it.

The hard part of developing frameworks (to me) is managing the evolution of published APIs. How do you meet new requirements or usage patterns, and maintain backwards compatibility, without leading to API bloat? I'm by no means a framework expert; I'd be interested in gathering some opinions on this.

I doubt that there's anybody who disagrees with this sentiment and it works well within one team or a department in which different teams frequently communicate. But how do you scale it up to large organisations in which hundreds or thousands of developers are working on different projects around the world? This is what the reuse community is trying to address and knee-jerk exhortations to "just refactor" don't even begin to address the issues involved.

When it comes to large-scale frameworks with large user bases, I think Eclipse is a good example of how things can work. Erich Gamma shares some thoughts about frameworks and reuse in this article, which I found quite interesting:

http://www.artima.com/lejava/articles/reuse.html

Thanks Ryan, the interview with Eric Gamma is excellent. Bertrand Meyer provides additional advice for large-scale reuse, in "Object-Oriented Software Construction, Second Edition, Chapter 4."
He not only speaks of the technical and tool related issues, but he also speaks about the social, political, and organizational issues.