Fixed #34421 -- Fixed QuerySet.update() on querysets in descending order by annotations.

[HB6H057]

I often use admin to manage data, so I wrote an action to update data in bulk, but sometimes it reports an error

FieldError: Cannot resolve keyword 'message_length' into field. Choices are: message, id, text, x

And the error is reported in descending order, similar to the execution of the following query statement

Model.objects.annotate(message_length=Length('message')).order_by('-message_length').update(text="Can I come on board?")

Through debug found that the programmer ignored the case of descending order, so wrote this patch to fix the problem, see the code for more details.

First submission, so if there are any questions please guide me.

[charettes]

@felixxm, not sure how we want to treat this one.

It's clearly a case missed by #15747 / ticket-28897 which is part of 4.2 but with and without the patch the test fails with the same signature

django.core.exceptions.FieldError: Cannot resolve keyword 'abs_id' into field. Choices are: foo, foo_id, id, m2m_foo, x

So it's technically not a regression but certainly a bug in newly introduced feature that allows update to be ordered by annotations. cc @David-Wobrock

[felixxm]

So it's technically not a regression but certainly a bug in newly introduced feature that allows update to be ordered by annotations. cc @David-Wobrock

ticket-28897 is classified as a bug (not a new feature) and #15747 is a bugfix so I don't think it's a release blocker for Django 4.2 🤔

[felixxm]

So it's technically not a regression but certainly a bug in newly introduced feature that allows update to be ordered by annotations. cc @David-Wobrock

ticket-28897 is classified as a bug (not a new feature) and #15747 is a bugfix so I don't think it's a release blocker for Django 4.2 thinking

... but I'm happy to backport it, if we will have a reasonable patch before Monday.

[charettes]

yeah I'm on the fence as well, whatever works best here.

[felixxm]

yeah I'm on the fence as well, whatever works best here.

I noticed that 3ef37a5 was backported to Django 4.1 (in alpha) so it definitely doesn't qualify for a backport. I'd treat it as a separate bug.

@HB6H057 Can you create a new ticket in Trac? All bugfixes require an accepted ticket.

[HB6H057]

@felixxm I created a ticket and added the tests to the list of skipped tests, but now I have a little problem with the suggestion. Please see if there are any other problems, thanks.

[felixxm]

@HB6H057 Thanks for updates 👍 Welcome aboard ⛵

I pushed small edits.

[HB6H057]

@felixxm Hi, it occurred to me that you used the new python 3.9 feature removeprefix in your modified code, and the django documentation shows that version 4.x supports python 3.8, so is that okay?

[felixxm]

@felixxm Hi, it occurred to me that you used the new python 3.9 feature removeprefix in your modified code, and the django documentation shows that version 4.x supports python 3.8, so is that okay?

On the main branch we're developing Django 5.0 which support Python 3.10+.