<< Back to previous view

[CLJ-1318] Support destructuring maps with namespaced keywords Created: 06/Jan/14  Updated: 23/Feb/14  Resolved: 31/Jan/14

Status: Closed
Project: Clojure
Component/s: None
Affects Version/s: Release 1.5
Fix Version/s: Release 1.6

Type: Enhancement Priority: Major
Reporter: Alex Miller Assignee: Unassigned
Resolution: Completed Votes: 0
Labels: None

Attachments: File clj-1318-1.diff     File clj-1318-2.diff     File clj-1318-3.diff     File clj-1318-4.diff     File clj-1318-5.diff     File clj-1318-6.diff    
Patch: Code and Test
Approval: Ok


Current :keys destructuring expects symbols and creates local bindings based on those symbols. This works fine with maps that use non-namespaced keyword keys. This enhancement is to add support for destructuring maps with namespaced keyword keys.

;; typical key destructuring for keyword keys without namespaces
(let [{:keys [a b]} {:a 1 :b 2}] (+ a b))

;; WANT some way to destructure map with namespaced keys
(let [{:keys [????]} {:x/a 1 :y/b 2}] (+ a b))

Approach: Allow keywords (with or without namespaces) in :keys destructuring. Destructure to bindings with the name of the keyword (namespace is ignored).

;; this now works
(let [{:keys [x/a y/b]} {:x/a 1 :y/b 2}] (+ a b))

;; add support for putting keywords into :keys as well to support ::keywords
(let [{:keys [:x/a :y/b]} {:x/a 1 :y/b 2}] (+ a b))
(let [{:keys [::a]} {:user/a 1}] a)

;; syms will also now support namespaced symbols
(let [{:syms [x/a y/b]} {'x/a 1 'y/b 2}] (+ a b))

Patch: clj-1318-6.diff

Screened by: Stuart Sierra. See comments, below.

Doc TODO: Will need to update http://clojure.org/special_forms#binding-forms with new binding form.

Comment by Nicola Mometto [ 06/Jan/14 11:58 AM ]

Why {:keys [:a/b]} and not {:keys [a/b}}?
Also, this should probably be extended to :syms for consistency

Comment by Alex Miller [ 06/Jan/14 12:28 PM ]

Good questions both. For the first question, we want to make locally namespaced keywords (::foo) work and there is no way to say that as a symbol.

I am waiting to hear back from Rich whether support for namespaced :syms is desirable. I think the change to support it is identical to the change to support namespaced keywords as symbols. I'm going to proactively update the patch to support that too.

Comment by Alex Miller [ 06/Jan/14 12:50 PM ]

Added new patch - now supports namespaced symbols or keywords in :keys and namespaced symbols in :syms.

Comment by Rich Hickey [ 06/Jan/14 1:00 PM ]

Should (also) support symbols for names, e.g. {:keys [a/b]}, only limitation is you can't get ns alias resolution. :syms support makes sense, but may seem weird to provide keywords for local names (where it doesn't as much for keywords), but would allow reaching aliases. My preference is no keyword names support for :syms, i.e. {:syms [a/b]} ok, {:syms [:a/b]} not.

Comment by Nicola Mometto [ 06/Jan/14 1:10 PM ]

To me {:syms [:a/b]} doesn't feel any more weird than writing {:keys [:a/b]}.
If this is going to be added, I think it should be consistent for :keys and :syms.
I understand that :syms is rarely used and this should not be an issue realistically, but I would expect everything that works for :keys to work for :syms too and adding only half a feature to :syms might cause unnecessary confusion.

Comment by Nicola Mometto [ 07/Jan/14 2:16 PM ]

With this patch this will now work:

user=> (let [:a/b 1] b)

I don't think this is desiderable.

Comment by Alex Miller [ 07/Jan/14 3:52 PM ]

Right, that is a consequence of allowing keywords in the :keys. At a glance this seems hard to address without significant changes unless we catch it prior to processing. Will consider.

Comment by Alex Miller [ 07/Jan/14 4:40 PM ]

Added new patch variant that catches keywords as let binding keys and throws an Exception.

Comment by Alex Miller [ 10/Jan/14 2:24 PM ]

Added one test in -4 showing example of auto-resolved keywords in :keys.

Comment by Stuart Sierra [ 10/Jan/14 3:00 PM ]

Screened. A few comments:

1. The examples in the tests use {:keys (a b)} with lists instead of {:keys [a b]} with vectors. Both forms are accepted both before and after the patch, but the docs at Clojure - special_forms only show vectors.

2. I would like this to work, but it would add some complexity:

(ns com.example.myproject.foo)

  (def data
    {::a 1 ::b 2})

  (ns com.example.myproject.bar
    (:require [com.example.myproject.foo :as foo]))

  ;; I would like this to work:
  (let [{:keys [foo/a foo/b]} foo/data]
    [a b])
  ;;=> [nil nil]

  ;; This is good enough, however:
  (let [{:keys [::foo/a ::foo/b]} foo/data]
    [a b])
  ;;=> [1 2]

3. This doesn't produce an error, which is logically consistent but perhaps not desirable:

(let [{:a ::foo/a} foo/data]
Comment by Rich Hickey [ 24/Jan/14 10:11 AM ]

please change the tests to use vectors

Comment by Alex Miller [ 24/Jan/14 10:28 AM ]

Added new -5 diff that uses vectors instead of lists in :keys tests.

Comment by Alex Miller [ 24/Jan/14 11:07 AM ]

And also fixing :syms [] in -6 diff.

Comment by Alex Miller [ 24/Jan/14 11:08 AM ]

Changed examples in description to use [].

Comment by Fogus [ 07/Feb/14 2:23 PM ]

A potential point of confusion here is illustrated by the following:

(let [m {:x/a 1, :y/b 2, :x/b 3000}
        {:keys [x/a y/b x/b]} m]
  (+ a b))

//=> 3

To get the answer 3001 one needs to remove the conflicting binding :y/b. Maybe this is not a big deal, but expect questions for the next 100 years.

Comment by David Nolen [ 23/Feb/14 5:01 PM ]

Ported to ClojureScript with CLJS-745

Generated at Tue Jul 26 17:07:09 CDT 2016 using JIRA 4.4#649-r158309.