On 01/08/2013 11:03 PM, Peter Rosin wrote: > > [BIG SNIP] > >> Then again, in the longer term, wouldn't it be better to provide a >> (GNU or non-GNU) package meant to wrap all this MSVC incompatibilities >> in a secluded place, instead of having Automake chase all this >> intricacies with mixed fortune? After all, we don't have random users >> building on MSYS with MSVC -- the users interested in doing so should >> know what they are doing, so we could ask them to install this >> hypothetical "wrapping package" before trying any such compilation. >> It might also be made part of MSYS itself eventually, if it proves >> itself on field. >> >> Good idea, bad idea, or simple wishful thinking? > > I think it's a bad idea. Try finding a canonical cccl script on the > Internet... Trying to do this outside of autotools will just grow > forks. > Then what about doing it in the Autotools? That is, let's continue to carry the ar-lib, compile, etc. scripts in our tree, let's continue to distribute them in the client packages tarballs, but admit that the setups they cater for are today special and/or corner case enough that we don't need to "automagically" detect them; instead, let's the user explicitly ask for their use (with an environment variable? with a configure options? not sure). We'll have less time-wasting and complex probing logic, but still a tight integration, and no risk of proliferating forks.
Of course, this could create yet more compatibility and/or transition and/or deprecation headaches. So, all things considered, it might be a poor trade-off. Hmmm... Well, let's put this idea in the back-burner for now. We still have a confirmed bug to fix, before doing anything else ;-) Regards, Stefano
