I would guess that it's because def is a special form that gets special treatment in clj... I think the symbol in the syntax quote should be ~'myvar to ensure that the argument to def is an unqualified symbol (because syntax quote would typically try to qualify the symbol)

Apparently this is allowed in Clojure:

user=> (def user/foobar 1)
#'user/foobar

but I think officially it's not allowed, probably just under-checked

wow..thanks Bob. It works.

when i macroexpland on the original version, it did expand to a full qualified var in bb, but it blow up when I actually use it. werid

Again: https://clojurians.slack.com/archives/CLX41ASCS/p1690736359151019?thread_ts=1690735947.465629&cid=CLX41ASCS

Sorry I don't quiet follow, what is exactly allowed but not officially allowed? And how is that related to my case here?

I'm under a namesapce

user/foo is not a simple symbol but a qualified symbol

not user

yes... but macroexpand expand to myns/myvar

which is exactly right and what i want right?

no, you should quote it

~'myvar

according to <https://clojure.org/guides/learn/syntax#_def>, the argument to def should be a 'simple' symbol (no namespace), and def should create that var is the current ns. So, the fact that syntax quoting puts an ns on the symbol should probably 'break', but in the special case where the ns happens to be the current ns, it works in clj (but not in bb). So, the ~'myvar ensures that the def statement that gets created doesn't have a ns on the symbol.

Consider this:

(defmacro mydef
  []
  (let [some-vals-need-computing [1 2 3]]
    `(def myvar ~some-vals-need-computing)))
#'user/mydef
user=> (ns foo)
nil
foo=> (user/mydef)
Syntax error compiling def at (REPL:1:1).
Can't refer to qualified var that doesn't exist

So while your macro works, it only works if you use it from the same namespace in clj, which is pretty brittle. Best to avoid it

Thanks! I think I understand now.

I can relax the check in SCI to make it behave like JVM Clojure though, but it's not the case today and I doubt it is a good relaxation

I just read this code: https://github.com/clojure/clojure/blob/501348deb1e49736f3c49f2c177bcff947fe3a17/src/jvm/clojure/lang/Compiler.java#L532-L543 and it does seem an intended choice. I'd also check with CLJS too see what they do

Apparently CLJS supports it too:

cljs.user=> (def cljs.user/dude 1)
#'cljs.user/dude
cljs.user=> (def user/dudex 2)
Unexpected error (ExceptionInfo) compiling at (<cljs repl>:1:1).
Can't def ns-qualified name in namespace user at line 1
cljs.user=> (ns user)
WARNING: user is a single segment namespace at line 1
nil
user=> (ns cljs.user)
nil
cljs.user=> (def user/dudex 2)
Unexpected error (ExceptionInfo) compiling at (<cljs repl>:1:1).
Can't def ns-qualified name in namespace user at line 1

I'll make a SCI issue to support it in SCI too then, but currently you have to use ~'my-var

Thanks Michiel. It's been a while since I touched any clojure macro, I'm pretty rusty at macro and comply forget about the unqualified symbol syntax, which is why I made the mistake. Now that you explained, I think it's good not to relax it actually. 😀

My name is Michiel :)

Opps, sorry (

no worries