Association after_add callback for new records

Hi everyone

Please help me to identify the issue here. Either with the code or the documentation, not sure where the issue lies at the moment. Code first, then question.

I have replicated the issue with this simple test on edge:

def test_building_macros_callbacks
post = :title => ‘Work in progress’ )
assert post.new_record?
assert_equal [“before_adding”], @david.post_log # <- ASSERTION FAILS HERE

@david.equal ["before_adding<new>", "after_adding#{[](}"], @david.post_log


Test failure:
<[“before_adding”]> expected but was
<[“before_adding”, “after_adding”]>.

I have some functionality that depends on #after_add only getting called when the new association member is saved. The docs doesn’t state what would happen with new records, but my logic assumed that #after_add would only be called once the new association member has been saved. I suspect the later holds, cause it is only then when it truly becomes part of the association…

Now, do I fix the docs and work on another solution, or do I attempt to change the behavior to the one outlined above?

Kind regards

Hi Kenneth,

Could you try if this regression was introduced after this commit: 5cda000bf0f6d85d1a1efedf9fa4d0b6eaf988a1



Hi Eloy

Thanks, went back to dec91a2e0655d288bd5184c981e2f4c60d9b3cf1 (one previous to the one you suggested) and the bug is still visible. I also branched off the v2.2.2 tag, and the bug is visible back then as well.

I’ll try my best blame/bisect combos to see if I can find the offending commit that broke this, seems to be way older than thought.

Kind regards

Thanks for investigating Kenneth!
At least now we know I didn’t break this for 2.3 RC2 :slight_smile:


This getting very strange now… Went back all the way to 2.0.5, and it still didn’t work. So I git-log’ed the callback test case to find its “origin” as a starting point.

This lead me to r9224 (36b8073ff553e) and ticket #11374 ( The test still fails here. At this stage I’m beginning to wonder if it ever really work at all…

So I jumped back one commit, to r9223 (50538fb5249), and the test still fails.

Should I carry on with this trip down memory lane, or branch off and attempt to fix it as I mentioned in my first email?


Looking closely at the patch provided on #11374 I see the test was probably just copied from somewhere else where the behavior was already specified to call #after_add before saving the child association.

So, this makes me think that I inferred the wrong behavior from the documentation. But I also want to add that I strongly feel that #after_add should only be called once the association is binded permanently in the database, since that is the behavior of all the other ActiveRecord callbacks.

Any thoughts?

I would go for the latter at this point.


Cool, I’ll kick off an attempt to get the expected behavior. If anyone wants to stop me, please do it quickly :slight_smile:


(Starting a new thread)

OK, some feedback as to where I’ve gotten with this one.


I gave some sample code in the ticket, and it’s part of the second patch, which highlights an issues with has_many :through’s and the association callbacks. The after_add callback gets called, but the association is never setup correctly. I’m not exactly sure how to proceed with this one now and need some advice. Did I blow open another issue here?