range function wrong in 1.7.0-beta?

204 views
Skip to first unread message

Mathias De Wachter

unread,
Apr 18, 2015, 3:32:40 PM4/18/15
to clo...@googlegroups.com
Hi,

this looks like quite a serious bug to me (at least it messed up my project):

First taking the code taken from grimoire:

clojurechess.position> (defn range
  "Returns a lazy seq of nums from start (inclusive) to end
  (exclusive), by step, where start defaults to 0, step to 1, and end to
  infinity. When step is equal to 0, returns an infinite sequence of
  start. When start is equal to end, returns empty list."
  {:added "1.0"
   :static true}
  ([] (range 0 Double/POSITIVE_INFINITY 1))
  ([end] (range 0 end 1))
  ([start end] (range start end 1))
  ([start end step]
   (lazy-seq
    (let [b (chunk-buffer 32)
          comp (cond (or (zero? step) (= start end)) not=
                     (pos? step) <
                     (neg? step) >)]
      (loop [i start]
        (if (and (< (count b) 32)
                 (comp i end))
          (do
            (chunk-append b i)
            (recur (+ i step)))
          (chunk-cons (chunk b) 
                      (when (comp i end) 
                        (range i end step)))))))))
WARNING: range already refers to: #'clojure.core/range in namespace: clojurechess.position, being replaced by: #'clojurechess.position/range
#'clojurechess.position/range
clojurechess.position> (range 0 11 2)
(0 2 4 6 8 10)

which is what I'd expect and relied on.

Now, looking at the new code:

clojurechess.position> (clojure.repl/source clojure.core/range)
(defn range
  "Returns a lazy seq of nums from start (inclusive) to end
  (exclusive), by step, where start defaults to 0, step to 1, and end to
  infinity. When step is equal to 0, returns an infinite sequence of
  start. When start is equal to end, returns empty list."
  {:added "1.0"
   :static true}
  ([]
   (iterate inc' 0))
  ([end]
   (if (instance? Long end)
     (clojure.lang.LongRange/create end)
     (clojure.lang.Range/create end)))
  ([start end]
   (if (and (instance? Long start) (instance? Long end))
     (clojure.lang.LongRange/create start end)
     (clojure.lang.Range/create start end)))
  ([start end step]
   (if (and (instance? Long start) (instance? Long end) (instance? Long step))
     (clojure.lang.LongRange/create start end step)
     (clojure.lang.Range/create start end step))))
nil
clojurechess.position> (clojure.lang.Range/create 0 11 2)
(0 2 4 6 8 10)
clojurechess.position> (clojure.lang.LongRange/create 0 11 2)
(0 2 4 6 8)
clojurechess.position> (clojure.core/range 0 11 2)
(0 2 4 6 8)

So the culprit is clojure.lang.LongRange/create.

Nelson Morris

unread,
Apr 18, 2015, 5:00:28 PM4/18/15
to Clojure
I was trying to write a test.check property for this, and it seems to have a found a different bug around `count` and `range`.

```
*clojure-version*
;;=> {:major 1, :minor 7, :incremental 0, :qualifier "beta1"}

(require '[clojure.test.check :as tc])
(require '[clojure.test.check.generators :as gen])
(require '[clojure.test.check.properties :as prop])

(def range-count-is-ceil-of-space-divided-by-step
  (prop/for-all [start gen/int
                 end gen/int
                 step (gen/such-that #(> % 0)
                                     gen/nat)]
                (= (count (range start end step))
                   (long (Math/ceil (double (/ (max 0 (- end start)) step)))))))

(tc/quick-check 100 range-count-is-ceil-of-space-divided-by-step)
;;=> {:result false, :seed 1429389499284, :failing-size 5, :num-tests 6, :fail [-2 -1 3], :shrunk {:total-nodes-visited 9, :depth 3, :result false, :smallest [-1 0 2]}}

(range -1 0 2)
;;=> (-1)
(count (range -1 0 2))
;;=> 0

(tc/quick-check 100 range-count-is-ceil-of-space-divided-by-step)
;;=> {:result false, :seed 1429389621085, :failing-size 4, :num-tests 5, :fail [-3 4 5], :shrunk {:total-nodes-visited 11, :depth 5, :result false, :smallest [0 1 2]}}

(range 0 1 2)
;;=> (0)
(count (range 0 1 2))
;;=> 0
```

--
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to clo...@googlegroups.com
Note that posts from new members are moderated - please be patient with your first post.
To unsubscribe from this group, send email to
clojure+u...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
---
You received this message because you are subscribed to the Google Groups "Clojure" group.
To unsubscribe from this group and stop receiving emails from it, send an email to clojure+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Beau Fabry

unread,
Apr 18, 2015, 5:08:33 PM4/18/15
to clo...@googlegroups.com
Ouch. Suspect this is the problem https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/LongRange.java#L161

Pretty sure that boolean should be round the other way.

Nelson Morris

unread,
Apr 18, 2015, 6:46:26 PM4/18/15
to Clojure

Alex Miller

unread,
Apr 18, 2015, 8:33:27 PM4/18/15
to clo...@googlegroups.com
Thanks all! Will definitely fix before release.
Reply all
Reply to author
Forward
0 new messages