update_columns doesn't support an empty hash

The following will return an error:


ArgumentError: Empty list of attributes to change

I think that’s surprising, because I see update_columns as an analogue to update_attributes, just without callbacks happening. An empty hash for the latter is fine:



So shouldn’t update_columns support an empty hash too? In the rare cases where callbacks should be avoided and this is needed, it would save having to make a special-case check to prevent the error.

Are not both cases invalid input? Why should we accept empty hashes?

Basically because http://en.wikipedia.org/wiki/Null_Object_pattern. It’s often the case that systems have to deal with “N” possible changes, where N is between 0 and 1. e.g. I might have a procedure to gradually build up a list of things to change, starting with {} and appending to it if certain conditions are met. It wouldn’t be an exceptional situation if none of those conditions are met.

It makes sense to me, you get passed a hash with the attributes to update,
which could be none.

This is a no-op edge-case similar to append an empty list to a list, etc.,
and could have use cases where the hash is built programatically.

The implementation generally uses iterators, which is the normal way to
support these edge-cases with no explicit handling, maybe the exception is
raised by update_all(attributes)



Bah I mean "N is between 0 and " (rather than 1 and ).

And what happen if users expect N to be between 1 and and we silently do nothing when 0 is sent to these methods?

In my opinion if you don’t have changes to be made in the database, just don’t call the method.

This behavior was added at https://github.com/rails/rails/pull/7133.

Also don’t assume that update_attributes and update_columns are analogues. The later is dealing directly with the database without any kind of interaction with the object instance, so they will behaves differently in a lot of ways, like type casting for instance.

Yes. It is raised by update_all(attributes).

I think the error makes sense with update_columns and not with update_attributes. Since the former is essentially a pass through to the database, and the database would raise an error without any columns to update, I’d expect an error when the method is used that way as well.