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

update_or_create() methods for QuerySet? and Manager

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

added documentation and tests

removed the "required" wording in the documentation

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

update() from #3180 and update_or_create() together

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

As per Jacob's comments in the mailing list.

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__().

Updated to work with r9844.

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.

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.

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

added versionadded directive for update_or_create

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.