v0.9.8 - 'Unlink' never removes previous link relation changes.

Love the package! It’s been very helpful of late, but I’ve noticed an issue 😃

I noticed that if you persist a parent model in memory, the relationChanges array just continues to grow as you link and unlink objects to the parent object.

It looks like there is a check failing in the unlink function for a couple of reasons.

In source, the function looks like this:

exports.unlink = function unlink(obj, options, callback) {

  if (typeof(options) === 'string') {
    options = {name: options};
  }
  var opts = h.$extend({
    name: 'default'
  }, options);

  callback = h.getCallback(arguments);
  
  var changes = this.relationChanges;
  for (var i in changes) {
    if (changes.hasOwnProperty(i) &&
        changes[i].name === opts.name &&
        h.checkEqual(changes[i].object, obj)) {
      delete this.relationChanges[i];
    }
  }
  
  this.relationChanges.push({
    action: 'unlink',
    options: opts,
    callback: callback,
    object: obj
  });
};

First, changes[i].name === opts.name will always evaluate false, as the relation name is actually at changes[i].options.name.

Second, delete this.relationchanges will leave an ‘undefined’ object in the array at the index, causing issues further downstream. You’re better off using this.relationChanges.splice(i, 1);

Third, in the h.checkEqual() function, there is a conditional as below:

else if (obj1.hasOwnProperty('modelName') && obj2.hasOwnProperty('modelName') &&
           obj1.modelName === obj2.modelName) {

This will exclude the entire second half of the function, as the modelName attribute is located at obj1.__proto__.modelName

I’ll prepare a PR with the appropriate changes to fix the issue. I’m not fussed if you would rather approach the issue in a different manner, but we’re keen to get this merged to avoid issues with patching at build.

[lp-belliot]

I’ve put the changes here:

https://github.com/lp-belliot/nohm/tree/feature/fix_bug_in_unlink_function

Not sure what to do given master is well past the v0.9.8 commit.

[maritz]

Sure, I just won’t spend much time on it in v1. 😉
But I can certainly read, accept/decline PRs and publish new versions.

The checkEqual is still the same in v2, the other 2 problems are already fixed in v2.

@lp-belliot thanks for the report & PR! Can you add some unit tests that fail without your changes? If possible one for each of the 3 problems you found.