On 14 January 2011 10:28, Sim IJskes - QCG <s...@qcg.nl> wrote: > On 14-01-11 11:13, Dan Creswell wrote: > > I may be misunderstanding but I'm not real keen on rebuilding the build >> system on the basis that "it really will be better, really, even though we >> can't explain why in advance" - that is the mistake of many a dev team >> over >> the years. >> > > Even so, without knowledge of serious defects in the current build, it > feels a bit like: "lets rebuild river, in the xyz programming language, > because its newer, and better. And throw the old version away". > > I see a fundamental problem here. Peter puts a lot of effort into something > new here, without proof of the old one broken. He is completely free to do > so. If he wants to explore new things, thats absolutely fine, but in order > for us all to profit from his work, and that of Dennis, we need to have > consensus on our willingness to throw the old system away. Because thats > what we are going to do when we accept the new. > > > Yep, that is indeed a tension. I have a slight variation which is that the existing build mightn't be broken in the view of most but some want to make it "better". In that circumstance I think Dennis and Peter should be free to experiment but likely not on trunk (should disclose I've been working with Dennis on some of this stuff and actually it looks good, not sure I want to switch everything across all in one go though).
At some point in the future we need to debate the merits of one approach over another based on the experimentation thus far and figure out what to do next. However, that still leaves some chance that much work is done and then thrown out so to speak. That said, Linux has something like this kind of merit based system where stuff stays out of the mainstream kernel as patches until it's mature and clear that lots of people use it and want it in as a default (even then Linus sometimes refuses to accept something for design philosophy reasons and such). Thing is Linux has something of a benevolent dictator, the same sort of thing I believe is generally frowned upon for an Apache project. So the question is, when do we need consensus on such things?
