Converted from http://www.assembla.com/spaces/clojure/tickets/374

stu said: If we are going to use #=, then how about modify record to take an optional metadata map before the keyvals:

(defn record
  "Construct an instance of record-type from keyvals.

   record-type is a keyword naming the record.
   keyvals can include *any* keyword/value pairs, and
   can be preceded by an optional metadata map."
  {:added "1.2"}
  [record-type & keyvals]
  {:pre [(keyword? record-type)]}
  (let [meta (when (map? (first keyvals)) (first keyvals))
        keyvals (if meta (rest keyvals) keyvals)]
    (record-implementation-detail-multimethod record-type meta (apply hash-map keyvals))))
</code></pre>

and then the read syntax can be

<pre><code>
#=(record :user.Foo {:some-meta 1} :a nil :c 3)

richhickey said: I think you need to distinguish between print and print-dup contexts. I don't like the switch-on-type-of-first-arg in general. We don't do optional args preceding variadics. Also, normally we don't see metatadata in print unless print-meta is true.

I think a #= ctor call works for print-dup:

#=(Foo. 1 2 3 {:my :meta} {:other :field})

eliding the meta and extmap when neither is present:

#(Foo. 1 2 3)

not sure about ordinary print yet

stu said: In the above, why isn't print-dup's effect recursive, e.g.

#=(user.Foo 1 2 3 #=(clojure.lang.PersistentArrayMap/create {:my :meta}) 
                                 #=(clojure.lang.PersistentArrayMap/create {:other :field}))

print-dup on maps certainly behaves that way, including nested maps.

Also, isn't a print-dup version that includes metadata only correct when print-meta is false?

stu said: Updating tickets (#370, #366, #374)

Patch scoped to the discussion in this ticket and at http://dev.clojure.org/display/design/defrecord+improvements

Added IRecord patch file. Sorry.

Using primitive type hints seems to break the map->Record constructor:

(defrecord Bar [^long a b])
(map->Bar {:a 1 :b 2})
=> NullPointerException   clojure.lang.RT.longCast (RT.java:1008)

The patch does not address deftype. It gets (as a freebie) all the new stuff available to arbitrary Java classes, but no create method, positional fn, etc. I think we should do deftype as a separate patch.

RE: longCast

I will check that out. Thanks

RE: deftype

deftype does get a create and getBasis method, but the create is breaking because I am assuming a record ctor. I can fix that if we think that types should have a create that takes a map.

(deftype T [a])
(user.T/getBasis)
;=> [a]

(user.T/create {:a 1})
;; NoSuchMethodError

This is because I'm building a call to the record ctor that takes metadata and an extension map. Does it make sense to generate such a create that takes a map for types?

Thanks
:F

minor fixes from last patch set.

deftypes shouldn't get any map-taking support as they are not mappy (mappish?). They should get positional support. defrecord and deftype stuff should go in together.

Latest patch includes changes to support deftype literals and removal of RecordExpr.

This patch has already been applied. Updating status of this ticket.

https://github.com/clojure/clojure/commit/ac1e8ad9f182dc2e8a5254f3e4b7b77c0258353d