Skip to content

removeObjectsForKeys doesn't properly remove objects #17

Closed

Description

@jfmorin

Hi,

There is a problem with the NSMutableDictionary returned by mutableObjectFromJSONString. A call to removeObjectsForKeys will remove the objects but not the keys. So we end up with nil objects. I also noticed side effects on other dictionary values that should not be affected.

Should it be overridden in the JKDictionary class like removeObjectForKey is?

Thanks,
JF

Activity

  1. johnezang commented on Apr 19, 2011

    @johnezang
    Owner

    Can you provide the JSON and some example code the demonstrates the problem?

    From your description, it's not clear what the problem is. When you say "A call to removeObjectsForKeys will remove the objects but not the keys. So we end up with nil objects.", this is the expected behavior- when a NSDictionary does not contain a key, -objectForKey: will return nil (aka NULL).

  2. jfmorin commented on Apr 21, 2011

    @jfmorin
    Author

    Instead of removing the key/value pair as expected, the key remains in the mutable dictionary but the value is replaced by a null value. If I print the resulting object I can see: "myKey"=(null).

  3. johnezang commented on Apr 21, 2011

    @johnezang
    Owner

    The behavior you describe is the expected behavior. The key does not "remain" in the dictionary, any key that does not exist in the dictionary returns NULL (aka nil). There are a number of internal checks in the form of NSCParameterAssert that ensure that every key in the dictionary contains a non-NULL object associated with it. If you did not compile JSONKit with NS_BLOCK_ASSERTIONS, these checks are enabled, and would have caused an exception, which would have been logged to the console.

    I suggest you try creating a NSMutableDictionary with a number of keys/objects. Remove one of the keys, and see if the NSMutableDictionary has the same behavior for the removed key.

  4. omarkilani commented on Apr 22, 2011

    @omarkilani

    Hey John,

    I think what the OP is seeing is this (with JSONKit 2e6ccf4):

    Code:

      NSMutableDictionary *jkDict = [@"{\"a\":\"1\",\"b\":\"2\",\"c\":\"3\"}" mutableObjectFromJSONString];
      NSLog(@"jkDict = %@", jkDict);
    
      NSMutableDictionary *objcDict = [NSMutableDictionary dictionaryWithObjectsAndKeys:@"1", @"a", @"2", @"b", @"3", @"c", nil];
      NSLog(@"objcDict = %@", objcDict);
    
      [jkDict removeObjectForKey:@"b"];
      [objcDict removeObjectForKey:@"b"];
    
      NSLog(@"jkDict = %@", jkDict);
      NSLog(@"objcDict = %@", objcDict);
    

    Output:

    2011-04-22 14:46:05.354 main[90231:903] jkDict = {
        a = 1;
        b = 2;
        c = 3;
    }
    2011-04-22 14:46:05.357 main[90231:903] objcDict = {
        a = 1;
        b = 2;
        c = 3;
    }
    2011-04-22 14:46:05.358 main[90231:903] jkDict = {
        a = 1;
        c = (null);
    }
    2011-04-22 14:46:05.358 main[90231:903] objcDict = {
        a = 1;
        c = 3;
    }
    
  5. omarkilani commented on Apr 22, 2011

    @omarkilani

    Hey John,

    I've tracked this down to the fix which fixed issue #8 :)

    524b9d7

    Specifically this line:

    524b9d7#L0R1014

    This comment:

    // If the key was in the table, we would have found it by now.

    Is incorrect if removeObjectForKey was used (like my example above).

    dictionary->entry looks like this with the example:

    [0] = {keyHash = 1062, key = @"a", object = "1"}
    [1] = {keyHash = 0, key = NULL, object = NULL}
    [2] = {keyHash = 1068, key = @"c", object = "3"}

    So the loop terminates with NULL due to [1], but we need to keep going to find [2].

  6. johnezang commented on Apr 22, 2011

    @johnezang
    Owner

    @omarkilani, the line you mentioned in your comment is one half of the problem. The real problem is in the removal logic- it (should have) "re-added" the items following the removed entry so that linear probe hash collisions didn't result in a NULL gap in the middle of the linear probe hash collision.

    Your example code was very useful- it happened to create a dictionary with 3 items with a capacity of 3 items, and a, b, and c all just happened to be ≡ 0 mod 3. Sort of a "perfect storm" to exercise this bug.

    I checked in a bug fix for this issue (used the new github issues feature to close this issue automagically with "Closes #17").

  7. jfmorin commented on Apr 22, 2011

    @jfmorin
    Author

    Thanks for the help reproducing the problem and fixing it.
    Good job!

  8. omarkilani commented on Apr 23, 2011

    @omarkilani

    @johnezang, Thanks for the fix. Yeah, the real problem became obvious once I started delving in a bit more. I seem to get pretty lucky with just the right dictionaries to trigger things. :)

    Thanks again.

  9. added a commit that references this issue on May 4, 2011
    85a4377
  10. added a commit that references this issue on Sep 23, 2011
    422c636
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions