On 06/12/2010 20:44, Sean Leather wrote:
.NET and Java are used in billions of line of production code and have thousands of books written about them, but new versions often obsolete what would have previously been recommended practice. Breaking legacy documentation is generally the least of anyone's worries when moving a language forward.
As for breaking code, new versions of GHC often break a bunch of packages (see the build logs on hackage), although the breakage would be more pervasive in this case.
These two statements do not help your argument. "Other people do it, so it's okay if we do it, too" is not sufficient reason for breaking working programs and invalidating textbooks. This appeal to popularity[1] is false: Microsoft and Sun/Oracle certainly care about backwards compatibility, and the GHC developers do, too (cf. move from base-3 to base 4).
My anecdotal experience is that developers much prefer the .NET/Java approach in this regard, but I'm not aware of a formal survey. I would say that appeal to popularity is entirely the point here. We are not determining facts, but what users would prefer - introduce breaking changes (with a specified benefit), or stick to compatibility with old documentation.