Test on destroy fails after 3.0.5 upgrade

After upgrading 3.0.4 -> 3.0.5 one of my tests now fails on an
assertion like 'assert !@product.destroy'.

This tests a :before_destroy callback which simply returns false to
prevent deletion. I verified that the callback is firing and haven't
been able to locate a difference between what's returned in the 3.0.4
and 3.0.5 source code. A console session shows destroyed? returning
false as expected.

Has this happened to anybody else? It's a simple test so I'd expect
others to have the same result if there's an actual problem.
Lighthouse shows a recent issue with before_destroy but it appears
unrelated.

Is this using a habtm association? There was a change to how destroy
happens with those (which I personally think is broken) to do with
when the before_destroy callbacks run relative to the clearing out of
associated rows. It looks to me like a side effect of that change
could be that the return value of destroy could change

Fred

Hey Fred,

The model in question does have a HABTM association, but based on
reading the ticket I thought it was unrelated because I don't have a
dependent destroy set up and am only operating on the parent. Maybe
not.

I had looked at the new and old versions of destroy() but it appeared
they were returning the same value.

destroy() calls freeze(). In a console session with a sample object
frozen? returns false with my callback enabled, and true with it
commented out, so that appears to be working.

I guess I'd need to set up 3.0.4 again and check the output to see if
it was returning a different (falsy) value...