On Mon, Oct 1, 2012 at 2:57 PM, Dirkjan Ochtman <[email protected]> wrote: > On Sat, Sep 29, 2012 at 10:48 AM, Benoit Chesneau <[email protected]> wrote: >> Voilà, hopefully this thread can be a good start for merging rcouch, >> bigcouch & apache couchdb and will ease merge of the other features imo. > > As a distributor, the most important deficiency in today's release > tarballs is the fact that pretty much all the dependencies are > bundled, with no convenient way to use system copies. IMO this should > be done especially for C libraries like snappy and spidermonkey, but > preferably also for the Erlang parts with an alive upstream. I haven't > tracked it to closely, but I always get the impression that CouchDB > just vendors upstream and isn't that good about making sure patches > end up upstream as well (and tracking what versions correspond to the > dependencies shipped with a given CouchDB tarball).
Not sure what you mean. spidermonkey is choosed on the system while snappy is build statically if it's what you mean. My proposal for package maintainer is to use the autotools for that. Which would solve all of their needs. The autootols here would detect or use the path choosen by the package maintainer. While we keep the possibility to build static releases. About upstream version, I don't think it's really a problem. We shouldn't stop ourself to improve couchdb just because a distributions is slow to upgrade to latest stable release. Which doesn't mean we should always choose an upstream version. It should be choosed because it will improve the user experience. - benoît
