Should I feel wary about creating clojure keywords which have non-existent namespaces?
An example would be :foo/bar, where namespace foo doesn't actually exist. This seems to be possible because these keywords behave like literals. I couldn't find any problems doing this in the REPL, but I'm concerned about possible problems with AOT compilation.
A namespace will in fact not be created simply because a keyword or symbol is encountered which would "belong" to it, as the following interaction at a fresh REPL illustrates:
; SLIME 2010-05-06
user> (-> (.getNamespace :user/foo) symbol)
user
user> (-> (.getNamespace :user/foo) symbol the-ns)
#<Namespace user>
user> (-> (.getNamespace :bar/foo) symbol the-ns)
; java.lang.Exception: No namespace: bar found
However, this is no cause for worry. A keyword's or symbol's "namespace" field is just an interned string; there is no reference back to the corresponding namespace object involved even if one exists. In fact, as can be seen above, the .getNamespace method of keywords and symbols returns a string and one has to jump a few hops to get to the actual namespace from that.
Trying to resolve a namespace-qualified symbol with the resolve function is safe too. That's regardless of whether the namespace actually exists; if it doesn't, nil is returned, as in the case where it does exist, but holds no Var of the given name. ns-resolve, in contrast, will throw an exception like the one mentioned in the snippet from the REPL above if it can't find the given namespace.