forms.TypedChoiceField validation problem

[Mikhail Mitrofanov]

I get ValidationError on this code:

POST_DAYS_CHOICE = (
    (7, '1 week'),
    (14, '2 weeks'),
    (30, '1 month'),
    (60, '2 months'),
    (90, '3 months'),
)

def days_to_till_date(val):
    print val
    try:
        till_date = date.today() + timedelta(int(val))
    except ValueError:
        till_date = date.today() + timedelta(30)
    return till_date

class PostForm(forms.ModelForm):
    till_date = forms.TypedChoiceField(label='Period', choices=POST_DAYS_CHOICE, coerce=days_to_till_date, required=True, initial=30)

This is was removed from TypedChoiceField class:

    def validate(self, value):
        pass

[Claude Paroz]

This was changed in [9883551d50e].

Interesting use case, which is currently not tested in Django. Having the coerced value not in choices is a bit of a corner case (and not clearly documented), but I admit this might be considered as a regression.

I'll see if I can come up with a fix which is not too invasive...

In any case, the documentation will have to be clarified on this point.

[Claude Paroz]

Attached a patch with a possible fix. Test suite passes with SQLite. I'll let other core developers judge if this warrants to be added in a stable release.

[Claude Paroz]

Thinking a bit more about this, I think that it's too risky to push my patch in the 1.6 stable branch. We can document the regression in the 1.6 release notes, but you will probably have to create a custom field to workaround the issue until the next release.
Please, next time try to test your project with beta or at least RC releases, so we can catch those issues before the official release.

[Mikhail Mitrofanov]

Hi, no problem. I just test my project with 1.6 release. I create custom field.

[Claude Paroz]

Assigned to me, but would love to receive another core dev opinion.

[Simon Charette]

While you did a great job at putting a non-invasive patch together I'm not sure we should commit to backport it.

First because it doesn't meet any of the policy criterias and second because it concerns an undocumented and corner use case.

IMHO the small risk of introducing another regression, there's a small one given the original to_python override, is not worth the backport.

On the other hand -- apart the missing a release note -- the patch looks RFC for master.

[Claude Paroz]

Proposal for a 1.6 release note addition (misc backwards incompatibilities):

* Due to a change in the form validation workflow,
  :class:`~django.forms.TypedChoiceField` ``coerce`` method should always
  return a value present in the ``choices`` field attribute. That limitation
  should be lift again in Django 1.7.

Does it look sane?

[Simon Charette]

LGTM

[Claude Paroz <claude@…>]

In 4a00f132e0cc5246f7e7bd04b6d84a9d9ea4a0c1:

Added release note for TypedChoiceField coerce limitation

Thanks Elec for the report and Simon Charette for the review.
Refs #21397.

[Claude Paroz <claude@…>]

In 9f59149cfe97b3b0026bbc4ff1d2121e4cf04d0e:

[1.6.x] Added release note for TypedChoiceField coerce limitation

Thanks Elec for the report and Simon Charette for the review.
Refs #21397.
Backport of 4a00f132e0 from master.

[Claude Paroz <claude@…>]

In a0f3eeccf3d5a90f9914bfe20b15df05673ea59d:

Fixed #21397 -- Re-added flexibility to TypedChoiceField coercion

Thanks Elec for the report and Simon Charette for the review.