M2M model field without `blank=True` should map to a serializer relationship with `allow_empty=False`.

[marccerrato]

The Django 1.8 system check framework displays the warning fields.W340 when the null parameter is defined on a ManyToManyField, since it has no effect.

This null field affects on whether a related field in a model serializer is considered required or not. We can see it here.

From my point of view, this behavior is correct for a ForeignKey but not for a ManyToManyField, that should be considered not required by default.

Adding the var to_many to the condition on the aforementioned code might do the trick.

[normanjaeckel]

See test in #2674.

[pdr]

+1

[tomchristie]

should be considered not required by default.

For both this and #2674 I'm unsure exactly what behavior you're expecting.
M2M fields default to required, because the input field itself is required, although it may be an empty list.

Are you expecting to be able to provide null as the input for the M2M field, or are you seeing an issue when providing an empty list?

As currently described the existing behavior seems correct to me - require the input, but allow it to be an empty list.

[normanjaeckel]

See my answer in #2674. If required=True (the default) the framework should not accept empty lists/arrays. This is because the framework should behave like Django forms and here we got an error if we try to send an empty list to a required form field.

(The other point (not this issue) is that the required flag should be controlled by the model field option blank which is not the case at the moment.)

[marccerrato]

I agree with #2674 (comment)

[emanuelschuetze]

+1

[tomchristie]

Retitled more appropriately.
The behavior that's being requested here as far as I understand it is to mirror Django's blank model field -> required input.

I'm not 100% convinced that we want ManyToManyField(Something) to always require >= 1 entry and ManyToManyField(Something, blank=True) to allow empty lists, and I'd really like to see that discussion pushed to the mailing list, but certainly we should at least allow either case to be specified.

If we do change the behavior we should do so in a major version bump and call it out prominently.

Requires #2250 as a prerequisite.

[ideallical]

+1