<< Back to previous view

[CLJ-1387] reduce-kv on hash map ignores reduced objects in large maps Created: 18/Mar/14  Updated: 22/Mar/14  Resolved: 22/Mar/14

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

Type: Defect Priority: Major
Reporter: Jozef Wagner Assignee: Jozef Wagner
Resolution: Completed Votes: 0
Labels: None
Environment:

JRE 7


Attachments: File clj-1387.diff     File clj-1387-v2.diff     File clj-1387-v3.diff    
Patch: Code and Test
Approval: Ok

 Description   

Larger hash maps have nested INodes. As kvreduce implementations in INodes dereference reduced objects, parent INodes continue to reduce.

user=> (defn test-reduce-kv [m] (reduce-kv (fn [_ k v] (when (== 1 k) (reduced :foo))) nil m))
#'user/test-reduce-kv
user=> (test-reduce-kv (zipmap (range 3) (range 3)))
:foo
user=> (test-reduce-kv (zipmap (range 300) (range 300)))
nil

Dereferencing reduced objects should happen only PersistentHashMap/kvreduce - intermediate nodes should pass the Reduced object along.

Patch: clj-1387-v3.diff
Screened-by:



 Comments   
Comment by Alex Miller [ 18/Mar/14 5:11 PM ]

I updated the patch to use a generative test that will try many combinations of map size and the reduced index to bail out on. This test failed before applying the source patch and passes with it.

Comment by Rich Hickey [ 21/Mar/14 7:33 AM ]

if(root != null){ - return root.kvreduce(f,init); + init = root.kvreduce(f,init); + if(RT.isReduced(init)) + return ((IDeref)init).deref(); }

Turns code that always had a return into code that sometimes does.

Comment by Alex Miller [ 21/Mar/14 9:07 AM ]

Added new version of patch that retains the return flow and doesn't fall through.

Generated at Sun Nov 23 01:35:41 CST 2014 using JIRA 4.4#649-r158309.