Is this an issue which users have reported? in my experience with Solr mailing list, I am yet to see a request where users wish to add arbitrary directories to classpath
I don't really feel like searching through the archives at the moment, but it has come up – i don't know if anyone has explicitly requested the ability to add arbitrary directories, but there have certainly been discussions about the annoyance of needing to copy and/or symlink jars.
If nothing else: I'm a user, and i'm requesting it.
How important is this feature to be in 1.4?
As i said in my first comment i don't know.
It would be nice to have, but i certainly don't think it's a blocker ... even with the testing i've done, and even with the new tests i added to the patch, and even though the behavior for existing ./lib users hasn't been changed, i still wouldn't consider committing unless other people try it out and gave it a thumbs up.
Users in general have a lot of problems with classloading. Even with the current support with one lib directory I have seen so many users having trouble with classloading . This can only add to that confusion
I don't really see how this will make confusion about classloading any worse. most problems i can think of where people have classloader difficulty in solr stem from not understanding where they are suppose to copy their jars – they tend to get confused about which "lib" directory, particularly with example/lib containing the jetty jars. Allowing people to put the jar any where they want and point at it by name in the config file should reduce confusion.
Besides which: they're still free to create a ./lib dir and copy jars – that still works, no configuration needed.
I agree that the original patch (with the order in the config mattering) would have been confusing for people, but with the more recent patches where all jars are in the same classloader i can't imagine any situation where this will cause more problems/confusion then forcing people to make a lib dir.
I am sure most of the users will be happy with the minimal solr. The rest of them will happily download the whole thing however big it is.
I REALLY don't want to argue the merrits of this issue as if it's purpose was to decrease the size of the distribution – it was not the purpose, it's just a possible additional benefit – but i HAVE to disagree with you on this ... most users may only need a minimal solr, but we should not passively discourage people from finding features that can make them happier by making those features more complex to get (via an alternate larger download).
Adding this feature, and using it to reduce the size of the current examples may not reduce the size of the current distribution enough to be worth worrying about, that's fine. But i'm trying to thing longer term: there have been multiple threads discussing the goal of adding many more example directories demonstrating cool use cases of solr via interesting permutations of features (DIH, clustering, solr cell, velocity, etc...). This patch (or something like it) is going to be necessary before we can do anything like that.