Random bug in the & (set intersection) operator in Arrays

ruby-talk
1

Hi

I think I have discovered a random bug in the & (set intersection) operator in Arrays.

It statistically appeared 37 times out of 200 executions of my program.

on version ruby 1.9.1p376 (2009-12-07 revision 26041) [i686-linux]
on version ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]

I cannot deliver this full program (a source to source compiler) but here is a example :

array1 = [#<While:0x8a15808>,
  #<Say:0x8b0c2c0 @str="2_1a">,
  #<Say:0x8b0c25c @str="2_1b">,
  #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

···

========================================
array2 = [#<Say:0x8b0c2c0 @str="2_1a">,
  #<Say:0x8b0c25c @str="2_1b">,
  #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>,
  #<While:0x8b0c02c,
  #<If:0x8b0bb2c @body=[#<Say:0x8b0baa0 @str="2_3">], @cond=true>,
  #<Else:0x8b0ba78 @body=[#<Say:0x8b0b94c @str="2_3">]>]

array1 & array2 sometimes gives things like :

[#<While:0x8a15808 >,
  #<Say:0x8b0c2c0 @str="2_1a">,
  #<Say:0x8b0c25c @str="2_1b">,
  #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

while the expected result is

[#<Say:0x8b0c2c0 @str="2_1a">,
  #<Say:0x8b0c25c @str="2_1b">,
  #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

How can I report that ? Thx
JCLL
(ps : using Set instead of Array, this works perfectly)

2

Just as you did, but with a code repro. Send it to ruby-core@ or file a ticket on redmine.

···

On Nov 6, 2010, at 9:08 AM, Jean-Christophe Le Lann <jean-christophe.lelann@orange.fr> wrote:

How can I report that ?

3

Well, Set's implementation and Array's are different. I believe Array does not use #hash while Set does to efficiently find equivalent instances. Can you show the code for Say#eql?, Say#hash and Say#== (same for all other classes whose instances are put into your Arrays)? I am not convinced yet that you found a bug.

Kind regards

  robert

···

On 06.11.2010 17:08, Jean-Christophe Le Lann wrote:

Hi

I think I have discovered a random bug in the& (set intersection)
operator in Arrays.

It statistically appeared 37 times out of 200 executions of my program.

on version ruby 1.9.1p376 (2009-12-07 revision 26041) [i686-linux]
on version ruby 1.9.2p0 (2010-08-18 revision 29036) [i686-linux]

I cannot deliver this full program (a source to source compiler) but
here is a example :

array1 = [#<While:0x8a15808>,
   #<Say:0x8b0c2c0 @str="2_1a">,
   #<Say:0x8b0c25c @str="2_1b">,
   #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

array2 = [#<Say:0x8b0c2c0 @str="2_1a">,
   #<Say:0x8b0c25c @str="2_1b">,
   #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>,
   #<While:0x8b0c02c,
   #<If:0x8b0bb2c @body=[#<Say:0x8b0baa0 @str="2_3">], @cond=true>,
   #<Else:0x8b0ba78 @body=[#<Say:0x8b0b94c @str="2_3">]>]

array1& array2 sometimes gives things like :

[#<While:0x8a15808>,
   #<Say:0x8b0c2c0 @str="2_1a">,
   #<Say:0x8b0c25c @str="2_1b">,
   #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

while the expected result is

[#<Say:0x8b0c2c0 @str="2_1a">,
   #<Say:0x8b0c25c @str="2_1b">,
   #<If:0x8b0c1d0 @body=[#<Say:0x8b0c054 @str="2_1c">], @cond=true>]

How can I report that ? Thx
JCLL
(ps : using Set instead of Array, this works perfectly)

--
remember.guy do |as, often| as.you_can - without end
http://blog.rubybestpractices.com/

4

Hello Robert

I have no such methods. I simply though that the object IDs would be used for the intersection of the two arrays. Am I wrong ? Thx for your time and explanations !

Regards
JCLL

···

Le 07/11/2010 11:40, Robert Klemme a écrit :

Hi

I think I have discovered a random bug in the& (set intersection)
operator in Arrays.

Well, Set's implementation and Array's are different. I believe Array does not use #hash while Set does to efficiently find equivalent instances. Can you show the code for Say#eql?, Say#hash and Say#== (same for all other classes whose instances are put into your Arrays)? I am not convinced yet that you found a bug.

5

As long as you define equivalence solely based on object identity you do not need specific implementations of #hash and #eql?. You can find a bit of explanation here

http://blog.rubybestpractices.com/posts/rklemme/018-Complete_Class.html

Still it would be good if you could provide a simple test case that reproduces the problem. You need that for a bug report anyway.

Kind regards

  robert

···

On 07.11.2010 12:06, Jean-Christophe Le Lann wrote:

Le 07/11/2010 11:40, Robert Klemme a écrit :

I think I have discovered a random bug in the& (set intersection)
operator in Arrays.

Well, Set's implementation and Array's are different. I believe Array
does not use #hash while Set does to efficiently find equivalent
instances. Can you show the code for Say#eql?, Say#hash and Say#==
(same for all other classes whose instances are put into your Arrays)?
I am not convinced yet that you found a bug.

I have no such methods. I simply though that the object IDs would be
used for the intersection of the two arrays. Am I wrong ? Thx for your
time and explanations !

--
remember.guy do |as, often| as.you_can - without end
http://blog.rubybestpractices.com/