Reduce queries in uniqueness validation

Looks great! +1

How do I test if a query is being executed or not?

Thanks.

You could use assert_sql and/or assert_queries helpers.

Now with lasers, I mean, tests :P

This is interesting. I think in general the approach of not revalidating attributes that haven't changed is probably an optimization worth pursuing. Uniqueness is an odd duck tho, since it's relative to external state. I'm also not sure what you should do if the scope value has changed when the validated attribute has not. Keep working on it though!

Uniqueness is an odd duck tho, since it's relative to external state.

I think the validation for uniqueness should always be executed, even when the attribute hasn't changed for this reason.

I really don't think so. If the attribute was correctly saved (it was unique before), and it has not changed, then, doesn't matter the external state (that if was changed, was compared to the current object).

I just found a problem in my approach, that is when the scope changes (and I'm not validating this). I mean: I have a Product that belongs to a group, and the product name must be unique on that group. Then, I scope a validates_uniqueness_of :name to the group. If I change the group of the product, doesn't matter if I changed or not the :name, it must be validated.

Could you describe a concurrent situation where don't validate an attribute that didn't change can cause a problem (that doesn't include the situation above)?

How about adding support (to all validations which take a *list of columns) for :on => :changed (like :on => :create/:update).

Then people who want to reduce queries and are sure external changes won't screw things up can use it, and the current safe behavior remains for everyone else.

Carlos : I agree with everything Josh said. Also, partial updates can be turned off per model basis. So we should have tests for that, and always execute the query if partial updates are disabled.

not reproducible

-1 Tests fail when I applied the patch to 2-3-stable. The patch doesn't apply to master.

BugMash: +1

The patch applies and I don't see the failure in tests that Rizwan saw.

+1 verified.

The patches apply fine (Elad's on master and Carlos' on 2-3-stable) and all tests pass.

+1 verified Though I do wonder if *_changed? has the same issues mentioned at http://ryandaigle.com/articles/2008/3/31/what-s-new-in-edge-rails-d... (assignment/modification outside of attr=)

Elad Meidar's patch now does not get applied to master. Need to updated it. Elad Meidar can you update it for the master branch?

not reproducible in rails3 on current master.

Looks like validates_uniqueness_of was refactored pretty significantly in http://github.com/rails/rails/commit/44cd9e0e7132abe632664377f13f3e...

(which explains why Elad's patch doesn't apply to master any more)