model instance update() method and QuerySet update_or_create() method

[Gary Wilson <gary.wilson@…>]

Method would call get_or_create(), updating the object's data if the object already existed.

Idea came up in:

​http://groups.google.com/group/django-developers/browse_thread/thread/49a6b99dbcea4364/

[Gary Wilson <gary.wilson@…>]

update_or_create() methods for QuerySet and Manager

[Gary Wilson <gary.wilson@…>]

I should note that this patch depends on the update() method in #3181.

[Gary Wilson <gary.wilson@…>]

added documentation and tests

[Gary Wilson <gary.wilson@…>]

removed the "required" wording in the documentation

[Gary Wilson <gary.wilson@…>]

Correction: I should note that this patch depends on the update() method in #3180.

[Gary Wilson <gary.wilson@…>]

update() from #3180 and update_or_create() together

[Gary Wilson <gary.wilson@…>]

Changed the title to reflect the merging of the two patches.

[Gary Wilson <gary.wilson@…>]

As per Jacob's comments in the mailing list.

[Gary Wilson]

Model instance update method should not use __dict__.update() as that won't work with fields like ForeignKey. Perhaps we need to loop through all the fields, using setattr and the other logic present in Model.__init__().

[Tai Lee]

Updated to work with r9844.

[Tai Lee]

Updated the patch to work with trunk (r9844). Updated docs to match new docs. Updated the Model.update() method to use setattr() as mentioned above so it works with ForeignKey fields. Also updated a few typos in the existing QuerySet.update() tests and model instance docs.

This is listed as a maybe / "second-tier" feature for 1.1 with a comment that they are likely to be included if done. I think this should be ready to go in now.

[Tai Lee]

Changed milestone to 1.1 beta, because this is listed as a 1.1 maybe feature that should be included if done, and it appears to be done.

[Gary Wilson]

took save() out of the assignment loop in update() and made a few tweaks to the docs.

[Gary Wilson]

added versionadded directive for update_or_create

[Jacob]

OK, this has gone around a few times on django-dev and in person, and it's clear that we're not reaching a consensus. There's a lot apathy, mostly -- few are arguing strongly for this feature, and few are arguing strongly against it. So, I'm going to do the BDFL thing and make a decision:

It's pretty clear to me that update() is a nonstarter; it's a single line of code::

def update(self, kw):

self.objects.filter(pk=self.pk).update(kw)

So that's out.

How about update_or_create()? I personally find the name counter-intuitive -- it's really get_and_then_update_or_create() -- but that's trivial; a better name could be found. A bigger problem is that it's nonatomic; we went to great lengths to insure that get_or_create() doesn't suffer from synchronization problems, but we can't do that here.

Given both these problems, and given that they won't exist if people are forced to write these methods in their own code, I'm marking this wontfix.

[George Sakkis]

A bigger problem is that it's nonatomic; we went to great lengths to insure that get_or_create() doesn't suffer from synchronization problems, but we can't do that here.

Given both these problems, and given that they won't exist if people are forced to write these methods in their own code, I'm marking this wontfix.

Sorry for digging this up a year later but I didn't find an answer in django-dev; can you please clarify these last two points ? Why can't save(force_update=True) be in a transaction just like save(force_insert=True) and why the problem doesn't exist if the method is written outside the framework ?

[Jacob]

Milestone 1.1 beta deleted

[Antti Kaihola]

Adds .update_or_create() to QuerySet and Manager, without adding .update() to model instances. I attempted to make this atomic. Patch against git HEAD as of Jun 8 2012.

[Tim Graham]

(#20429 is a duplicate of this ticket that has been accepted)