Clear F expressions after a model instance is saved

[codefisher]

I would have thought these two expresses would be exactly the same, except the second would be more efficient.

A: some_model.some_field += 1
B: some_model.some_field = F('some_field') + 1

I hit some unexpected behaviour when, I did this

some_model.save()
# and somewhere far far away
some_model.save()

With A, the field is incremented once, as I expected. But with B, it gets incremented twice. I would have expected the F query to effect the next database call, and nothing more. It might be intended, but then it is not documented.

I could write some come to demonstrate this. I only noticed it because I was using an import script that was creating lots of objects, and I had A and B in the model's overridden save method. They were causing vastly different results.

[Tim Graham]

It doesn't seem intuitive and should definitely be documented. Probably a good idea to recommend fetching the object from the database after saving in order to get the new value if you are still using the object after save().

[codefisher]

You say "Probably a good idea to recommend fetching the object from the database after saving in order to get the new value if you are still using the object after save()."

That is already done, at least on the latest dev docs (see the note): ​https://docs.djangoproject.com/en/dev/ref/models/queries/#f-expressions

But the effect of calling save more than once is no where to be seen, and really not that expected.

[codefisher]

Adding the needed note into the documentation

[codefisher]

Just thinking about this, I am really more inclined to say how it works should be changed. I would suggest that after save() is called, it would be like the object had been called with defer() on that field. I don't think it would be hard to implement, I don't know the internals of the ORM, but defer() is already there, so what ever it does could be used again.

I can't see it causing backwards compatibility changes, this is undocumented, and not very intuitive. I would even suggest that after saving, that having the field affected by the F() expression not giving a sensible value is a bug in itself, and should be fixed - by treating as though marked by defer().

But there is always a change of causing backwards compatibility problems, that would be very hard to track down. Which needs to be weighed against how many problems the current counter intuitive behaviour would cause.

[Josh Smeaton]

I would be inclined to call this a bug - so I don't think backwards compatibility would be much of an issue. It's totally unexpected (for me at least) that the F() "persists" across a model.save() call. When the model is saved, the attribute should be converted to the relevant field/descriptor, rather than maintaining the proxy of the F() expression. I have no idea how hard or easy this would be without looking into it though.

[Tim Graham <timograham@…>]

In fc4b4fd:

Refs #23386 -- Documented that F() expressions are applied on each model.save()

[Tim Graham <timograham@…>]

In be03ce2:

[1.10.x] Refs #23386 -- Documented that F() expressions are applied on each model.save()

Backport of fc4b4fd5850989458d6e54de12a29b2e40e94ce8 from master

[Tim Graham <timograham@…>]

In 67c60cce:

[1.9.x] Refs #23386 -- Documented that F() expressions are applied on each model.save()

Backport of fc4b4fd5850989458d6e54de12a29b2e40e94ce8 from master

[Tim Graham]

Reclassifying the ticket to evaluate changing the current behavior.

[Tim Graham]

This will be fixed by #27222.