m2m_changed signal is unaware if .set() method is being called

[Leif Kjos]

When a related_m2m_field.set() method is called, the m2m_changed signal is fired four times for action: pre_remove, post_remove, pre_add, and post_add (or pre_clear / post_clear).

This poses a problem if the signal is supposed to run validation on the final state of the model. For example, let's say I have a model called Customer with a many-to-many relation to SubscriptionPlan and I have a m2m_changed signal that is supposed to validate if their SubscriptionPlan is valid.

If I call Customer.subscription_plans.set([some_plans]), the many-to-many manager will first call subscription_plans.remove() then .add() inside an atomic transaction. If my signal validates on the first .remove(), it could be in an invalid state, even though it won't be by the time the .add() completes. However, I still need to be able to validate on the signal if the standard .remove() method is called.

To get around this, I had to create a custom ManyToManyField with a custom RelatedManager that set an instance variable on the model when the .set() method was called. I'd like to propose adding a feature to the RelatedManager or to the signals to make the m2m_changed signal aware of if the .set() method was called when running. This could be a private attribute on the instance or extra information passed to the signal receiver.

If this feature already exists or there's a decent workaround, let me know and i'll close the ticket! Otherwise, I have a patch in mind that I can raise a PR for and attach.