On Wed, May 27, 2015 at 11:55 AM, Joan Touzet <[email protected]> wrote: > I've skimmed the entire thread, and it seems like Ilya is trying to > solve a much bigger problem than Paul has outlined. Ilya's approach is > more all-encompassing but involves a lot of "make-work" just to get the > system back to its current state. Meanwhile, Paul is trying to respond > to some requests for enhancement (presumably coming thru IBM/Cloudant) > and is suggesting just enough functionality to fix that. > > Unless there are more developers outside of IBM willing to contribute > code towards Ilya's proposal my guess is that IBM won't approve the > resources for such a vast change, and I'm not seeing people rushing to > help implement this approach who are well acquainted with the couch > codebase. > > Paul, consider this a vote of support in favour of your proposal, subject > to the concerns already raised. What does Bob have to say? > > -Joan
This is actually in response to the CouchDB Community's concern that the number of extension points being defined in the config files was growing to the point where it was confusing and misleading since they're a mostly different type of configuration. This lead to the vendor specific plugin proposal that was accepted as a good approach at solving that issue. Ilya did some work on this via plugerl. When reviewing that I realized that we have at least three places (couch_stats, chttpd handlers, plugerl, and I keep thinking I'm forgetting one more) where we have roughly the same issue of trying to connect various parts of the code base in a configurable manner (at the release level). So rather than have three different approaches I thought it'd be useful to take a step back and try and design something that could solve each of these cases in a single spot rather than having multiple implementations of roughly the same functionality. So no, this isn't motivated by IBM. Its motivated by not wanting to have three implementations of basically the same thing in CouchDB.
