> > I am still curious to know if the "linking" aspect applies to "import > xapian". > IMHO the linking aspect applies to "import xapian" and in general to any > kind of approach where the Xapian interfaces would have to be known from > CouchDb ASF provided code for an integration to work. > I would see establishing some kind of hard linkage and dependency between > the two technologies as "linking". > But I tend to consider linking in a technology neutral way, beyond the > interpretation of what linking means for the FSF or what it means a > specific > programming language, so take my words lightly.
Linking and interfacing with an API *are* different, or there'd be no need for the Affero GPL... With a modified BSD bridge like pyndexter it all becomes moot. If couch distributes an ASL or compatible reference indexer and the pyndexter bridge, if someone wants to use Xapian (in python) they'd have to *easy_install*xapian before they can *import *xapian (first they'd actually have to apt-get install xapian-core too). That's up to the end user but has nothing to do with couch. It's easy and everybody wins -- couch isn't distributing the bindings and there's no issue. Even better, as someone mentioned, would be an Erlang FTI implementation to distribute with couch, but that's a pipedream for now. Hyper Estraier, as an LGPL work, *could* have bindings shipped with couch, but it's just as simple to install these bindings as with xapian (in python at least) so it's not even necessary. And then there's Lucene (and CLucene). There are options a'plenty, and on top of it all indexing could work out of the box with pyndexter's (comparitively slow) pure-python implementation. So as soon as there's agreement on the index and query APIs I could wire pyndexter to them and we'd be in business.
