execute_after_commit triggers unexpected after_commit callbacks

[sevaorlov]

Hey, @magnusvk, I just realised that this code https://github.com/magnusvk/after_commit_action/blob/master/lib/after_commit_action.rb#L23
leads to triggering all after_commit callbacks on the model that don't have on option, which might not be an expected behaviour.

Checking all open transaction if they include current record can be slow, not sure what is the best solution here.

[sevaorlov]

maybe not so slow, so one possible solution I see is

def execute_after_commit(&block)
  transactions = ActiveRecord::Base.connection.transaction_manager.instance_variable_get(:@stack)
  records = transactions.map(&:records).flatten.uniq

  if ActiveRecord::Base.connection.open_transactions == 0 || !records.include?(self)
    block.call
  end
  ...
end

[magnusvk]

Hm, well that kind of defeats the purpose, though, right? In that case that code doesn't actually run after commit, it runs right away. The point of adding the record is so that this code can run after the commit. That side effect seems worth it?

[sevaorlov]

I think that is ok. When we call execute_after_commit we want to execute the code after current record transaction, but if there is no transaction for that record, that means record will not change and then there is no purpose of waiting some other transactions, we can execute the block straight away.

[magnusvk]

I don't think that's always true. What if I specifically warp a block in a transaction, for example? I'd expect after_commit hooks to run after the completion of that block, wouldn't I?