namespace should be nil for unqualified idents

[frenchy64]

;; jank
clojure.core=> (namespace 'a)
""
clojure.core=> (namespace :a)
""

;; clojure
user=> (namespace 'a)
nil
user=> (namespace :a)
nil
user=> (namespace (symbol "" "f")) ;;subtle
""


;; via clojure-test-suite
Testing clojure.core-test.namespace

FAIL in (nil) (nil:nil)
expected: (= nil (namespace :abc))
  actual: (not (= nil ""))

FAIL in (nil) (nil:nil)
expected: (= nil (namespace (quote abc)))
  actual: (not (= nil ""))
Exception: not nameable: nil

[frenchy64]

I think changing get_namespace() to return an option<native_persistent_string> will emulate Clojure the closest, where is_none() means unqualified and is_some() means qualified.

The main wrinkle here is that (qualified-symbol? (symbol "" "f")) is true in Clojure.

We might want to more carefully manage ns to ensure it's not empty. It might be a good idea when creating unqualified symbols and keywords to avoid the 1-arity ctor and use make_box<symbol>(none, name). It's probably faster too.

[jeaye]

Hm, that sounds like it'll be very easy to get wrong. So Clojure actually distinguishes "" from nil for a symbol's namespace? That's absolutely bonkers.

[frenchy64]

My take is that the behavior for a "" namespace is probably unspecified, so it might be ok to co-opt it as unqualified.

Assuming that, a much less invasive change would simply to add a not-empty to the return of namespace.

[jeaye]

My take is that the behavior for a "" namespace is probably unspecified, so it might be ok to co-opt it as unqualified.

Assuming that, a much less invasive change would simply to add a not-empty to the return of namespace.

Yep, I like that.