http://dev.clojure.org/jira This file is an XML representation of an issue en-us 4.4 649 25-07-2011 http://dev.clojure.org/jira/browse/CLJ-1106 Clojure <p>Equality for sets is not consistent with other persistent collections when the hashCode differs:</p> <p>(= {(Integer. -1) :foo} {(Long. -1) :foo})<br/> =&gt; true</p> <p>(= <span class="error">&#91;(Integer. -1)&#93;</span> <span class="error">&#91;(Long. -1)&#93;</span>)<br/> =&gt; true</p> <p>(= #{(Integer. -1)} #{(Long. -1)})<br/> =&gt; false</p> <p>Attached is what I believe to be a basic fix. It removes a hashCode check from APersistentSet.equals. A similar check was removed from APersistentMap in the following commit:</p> <p><a href="https://github.com/clojure/clojure/commit/b5f5ba2e15dc2f20e14e05141f7de7c6a3d91179">https://github.com/clojure/clojure/commit/b5f5ba2e15dc2f20e14e05141f7de7c6a3d91179</a></p> <p>With this patch, set equality works as expected and all tests pass.</p> <p>My understanding of the implications of this change are limited, so someone more knowledgeable would need to chime in about its correctness.</p> CLJ-1106

Broken equality for sets

Defect Major Closed Completed Aaron Bedra Justin Kramer Fri, 9 Nov 2012 09:42:48 -0600 Fri, 1 Mar 2013 09:49:21 -0600 Fri, 8 Feb 2013 07:31:35 -0600 Release 1.4 Release 1.5 Release 1.5 4 6 <p>Can you add some unit tests that fail with the current code but succeed with the change?</p> <p>Revised patch with unit test that fails in master and succeeds with patch</p> <p>vetting</p> <p>This came up when we were trying to diff two datasets by putting them in sets and comparing. We had Integers because they came back from JDBC that way.</p> <p>I'm going to take a look and try to get this shipped. It hit us and I would love to to see it in 1.5</p> <p>I applied the patch on master (ca465d6d) and it applied cleanly and fixes the issue.</p> <p>Instead of removing <tt>m.hashCode() != hashCode()</tt>, perhaps we should use `Util.hasheq()` instead. It already seems to special case numbers (<a href="https://github.com/clojure/clojure/blob/f48d024e97/src/jvm/clojure/lang/Util.java#L164-L172">https://github.com/clojure/clojure/blob/f48d024e97/src/jvm/clojure/lang/Util.java#L164-L172</a>) and won't have the potential performance impact that skipping hashCode checks could bring.</p> <p>Attaching a patch with an alternate fix for this issue that does not skip hashCode checking.</p> <p>It passes all existing tests and fixes this issue.</p> <p>I want to benchmark the difference, too.</p> <p>On further thought and comparison to <tt>APersistentMap</tt>, I'm not sure if that's the best use of <tt>Util.hasheq()</tt>. I can't find good reference on the semantic differences and am not familiar enough with the Clojure source to infer it, yet.</p> <p>Bo, I don't see you listed on the contributors list. Did you send in a CA?</p> <p>Both of the patches above are not complete. APersistentSet equiv calls equals. APersistentMap has two separate ways of handling this. This is the path that the fix should take.</p> <p>This patch addresses the difference between equals and equiv.</p> <p>The Set equality code currently in Clojure uses .hashCode to quickly bail out of comparisons in both .equals and .equiv. This feels wrong since .equals and .equiv presume different hashes. The map equality code avoids the problem by not checking any hash.</p> <p>Aaron's 2/13 patch makes set equality work like map equality, not checking the hash. (I think a much shorter patch that accomplishes the same thing merely drops the call to hash.)</p> <p>It seems to me a separate problem that both hash and set are broken for Java .equals rules, because of the equiv-based handling of their keys.</p> <p>I think this needs more consideration.</p> <p>Notes from discussion with Rich:</p> <p>1. Aaron's 2/2/13 patch, which mimics map logic into set, is the preferred approach.</p> <p>2. The broader issue I was worried about is the semantic collision between Map's containsKey and Associative's containsKey. This is out of scope here, and perhaps ever.</p> <p>Any chance this will make 1.5?</p> Approval Ok Global Rank Patch Code and Test