How to detect auto namespace in keyword ?

Question

I'm serializing a data structure with pr-str that may have keywords with auto namespace

I would like to reduce the data structure not to contain auto keywords before serialization, but I don't know how to detect it?

(def x ::outbound)

(pr-str x) => "::outbound"

(edn/read-string "::outbound") => exception

(namespace x) => "current-namespace" 

so is there a way to know that x has auto namespace ?

thank you,

Sorry - wrote the question without running it on repl - a proper example here

(keyword ":out") => ::out

(pr-str (keyword ":out")) => "::out"

(edn/read-string (pr-str (keyword ":out"))) =>

Execution error at user/eval93457 (form-init751239350220000799.clj:1).
Invalid token: ::out

Answer 1

Auto-resolved namespacing is not an attribute of keywords. Keywords have just a namespace and a name. Autoresolved (::) keywords are resolved at read time into a fully-qualified keyword with a namespace. The Clojure printer will never print a keyword with ::.

If you wanted to customize the printer or use some other function to detect that a keyword's namespace is the same as your current namespace and print it as an autoresolved keyword, that's possible (but almost certainly not a very good idea).

edn/read-string won't read auto-resolved keywords as the edn spec does not include auto-resolved keywords.

Comments

Thanks, I thought so, but for some reason I'm able to reproduce this, I'll dig deeper into it and will ask more once I find the culprit.

Thanks,

Timo

Answer 2

Auto-namespaced keywords are expanded to normal namespaced keywords during read time, so once you have them in a data structure, they are printed with namespace:

$ clj
Clojure 1.10.0
user=> (pr-str ::out)
":user/out"

How did you manage to make (pr-str ::out) print ::out?

Edit:
What you are observing is the fact that clojure allows you to contruct keywords that are not readable when they are printed.
(keyword ":out") constructs a keyword that has no namespace and ":out" as a name, so when it is printed as "::out", it's not that it's printed as "autonamespaced", it just prints it's name (":out") after colon ":" that identifies it as a keyword. You can go wild with constructing keywords from strings:
(keyword "heh [{:a 1}]" "//::! :: ::") will produce a keyword with namespace "heh [{:a 1}]" and name "//::! :: ::", it's printed like that :heh [{:a 1}]///::! :: ::.

Since in your case there is no way to clean your data, I'd suggest using serialization that allows custom readers and writers for different types instead of edn. Take a look at transit, for example

Comments

(keyword ":out") => ::out

(pr-str (keyword ":out")) => "::out"

(edn/read-string (pr-str (keyword ":out"))) =>

Execution error at user/eval93457 (form-init751239350220000799.clj:1).
Invalid token: ::out

obviously in our case the input is dynamic - the problem for us now is that it is stored in Datomic history and cannot be removed - thus we would need a way to fix either the serialization or deserialization somehow.

clojure.core/read-string evaluates the string and resolves the keyword correctly, but it opens up security problems so it should not be used

---

Oh, that's what's going on. What you are observing is the fact that clojure allows you to contruct keywords that are not readable when they are printed.
`(keyword ":out")` constructs a keyword that has no namespace and `":out"` as a name, so when it is printed as `"::out"`, it's not that it's printed as "autonamespaced", it just prints it's name (":out") after colon ":" that identifies it as a keyword. You can go wild with constructing keywords from strings:
`(keyword "heh [{:a 1}]" "//::! :: ::")` will produce a keyword with namespace `"heh [{:a 1}]"` and name `"//::! :: ::"`, it's printed like that `:heh [{:a 1}]///::! :: ::`.

Since in your case there is no way to clean your data, I'd suggest using serialization that allows custom readers and writers for different types instead of edn. Take a look at [transit](https://github.com/cognitect/transit-clj), for example