Admin does not correctly prefill DataTimeField from URL

[gilhad]

I was not able to format URL for Admin interface to prefill DateTimeField with given value.

It worked in 0.96, but does not work in 1.0 ( I used ../admin/MyApp/MyTable/add/?box=359&datum_date=2008-12-01&datum_time=17:30:27)

After some looking on source code and testing i find a solution:

• in /django/contrib/admin/options.py before line 520 add

    if isinstance(f, models.DateTimeField):
           initial[k] = initial[k].split(",")
  

• use this format: ​https://../admin/MyApp/MyTable/add/?box=359&datum=2008-12-01,17:30:27

For reference - the model of MyTable is such :

class MyTable(models.Model):
        box = models.ForeignKey(Boxes)
        datum =  models.DateTimeField(null=True, blank=True)

(plus some other insignificant fields.
The "datum" field should be prefilled with some date, which is computed by long way (not simple now()) and the use must be able to edit it BEFORE saving it)

---

The problem arises from DateTimeField be treated by MultiWidget, but not properly broken when got by URL (GET)

---

Patch:

--- options.py.old      2008-12-01 19:56:34.000000000 +0100
+++ options.py  2008-12-01 19:40:34.000000000 +0100
@@ -517,6 +517,8 @@
                     continue
                 if isinstance(f, models.ManyToManyField):
                     initial[k] = initial[k].split(",")
+                if isinstance(f, models.DateTimeField):
+                    initial[k] = initial[k].split(",")
             form = ModelForm(initial=initial)
             for FormSet in self.get_formsets(request):
                 formset = FormSet(instance=self.model())

[gilhad]

Typo - (Admin does NOT correctly prefill DataTimeField from URL (GET) + patch)

[Malcolm Tredinnick]

Edited description for clarity.

Please attach a proper patch to the ticket, rather than mixing the proposed changes into the description of the problem. It makes things a lot easier for reviewing and further improvements (if subsequent patches are made).

[gilhad]

Added patch for code AND documentation

[Alexander Herrmann]

attached a new diff for the current source and added test

[Gandalfar]

patch against r17364

[Gandalfar]

I've tested that patch works correctly with current trunk and cleaned-up docs syntax.

[Aymeric Augustin]

I dislike the idea of blessing hand-crafted URLs as a public API.

[Tim Graham]

In any case, the patch needs to be updated to apply cleanly.

[Ridley Larsen]

Replying to timo:

In any case, the patch needs to be updated to apply cleanly.

I have attached a patch that applies cleanly and passes testing.

[Claude Paroz]

Ridley, would it be possible to turn the patch into a Github pull request?

[Tim Graham]

​PR

[Markus Holtermann]

It feels to me like using a single white space instead of a comma is more intuitive. 2015-04-05 12:34:56 vs 2015-04-05,12:34:56

[Ridley Larsen]

patch against a2e56db

[Ridley Larsen]

I've attached a new patch with a different approach. Instead of special-casing the SplitDateTimeWidget, I modified its' decompress method to account for string values. Thanks Tim for the advice on the isinstance call.

[Tim Graham]

#19431 is a duplicate with an alternate approach that might be worth looking at to see if any bits could be incorporated.

[Gilhad]

I updated to Djago 1.10 on more sites, the problem still remains (cannot prefill admins datetime field) and the original fix still works (just the lines moved after line 1404 now).

Would be possible to somehow resolve it and incorporate it in main stream, so I would not be forced to manually patch it each and every version? Please, prettty, pretty please ...

[Baptiste Mispelon]

Just a quick update to note that this issue still happens in the current master (e8fcdaad5c428878d0a5d6ba820d957013f75595) with the following error message:

'str' object has no attribute 'utcoffset'

However there's been a bit of progress as there's now a documented way to populate a form using the data from request.GET with ModelAdmin.get_changeform_initial_data() [1] (introduced in 1.7).

Interestingly, the current implementation of get_changeform_initial() has a special case to handle ManyToManyField [2]but that special case is neither documented nor tested (removing it doesn't trigger any test failure).
I also find it curious that the current implementation doesn't seem to interact at all with the form layer (it only interacts with the model layer, checking that the keys of request.GET correspond to existing fields). Forms (and widgets) tend to be good at converting plain strings to useful python objects so maybe it could help resolve this ticket without hardcoding a bunch of special cases?

[1] ​https://docs.djangoproject.com/en/2.2/ref/contrib/admin/#django.contrib.admin.ModelAdmin.get_changeform_initial_data
[2] ​https://github.com/django/django/blob/e8fcdaad5c428878d0a5d6ba820d957013f75595/django/contrib/admin/options.py#L1495

[Fabian Binz]

I've looked at this ticket for quite some time during and after the recent sprint at the Cologne Django Meetup.

For now, I've come to the conclusion, that - while there are definitely ways to build the functionality that's requested in the ticket - it's actually quite hard to do that in a manner that's general enough to cover all cases.

What follows is a lengthy discussion of my current understanding of the problem, partly to just have a record of this for myself, but maybe we could also discuss a way forward:

1. Handling of initial data for multi-value fields
2. Parsing of localized datetime strings
3. Deviation from the way HTML forms normally work

The source of the "bug" is actually pretty simple: The SplitDateTimeWidget.decompress method expects its input to be a datetime object, which is not the case when the initial data is passed via a query parameter. According to the docs, this is a reasonable behavior, since the decompress method actually should only receive valid inputs (i.e. a datetime object). So, I guess it's not unreasonable to crash if passed a plain string.

So, why is this contract violated in this case? Because the initial data specified via query parameters is passed verbatim - via a bunch of indirections - to the widget's decompress method, without any form of validation or parsing.

Now, how could we actually parse or validate the input properly? In the case of a datetime field, I think we'd either have to settle on some standard way of formatting datetimes (datetime.isofromat seems reasonable as opposed to a comma- or space-separated representation discussed above) or we could take into account the DATETIME_FORMAT setting that the user has set. However, this doesn't really seem feasible.

Fundamentally, I think the issue is that POST and GET data are not handled in a similar manner for MultiValueFields/MultiWidgets:
Let's say we have a datetime field named "created_at". The SplitDateTimeWidget would render HTML similar to the following:

<input name="created_at_0" ...>
<input name="created_at_1" ...>

Normally, we'd use query parameters like "created_at_0=2024-01-02" and "created_at_1=15:31:23" to pre-fill both the date and the time. For multi-value fields however, we expect that we can set both values using a single query parameter.
When data is POSTed, we don't have that problem, since both inputs are simply separate values and can easily be passed to the MultiValueField.compress method.

For my taste, having this difference between POST and GET does not make much sense and it would seem more consistent to just use the individual input names to pre-fill fields. This also enables partially pre-filling inputs, which is much harder when expecting to have a single representation of a value.

But I also see the convenience of having a single URL parameter to set multiple fields. There are even test cases which kind-of test this behavior (though not for the SplitDateTimeWidget). So maybe an alternative would be to make the serialization/deserialization to/from string a part of the interface of the MultiValueField/MutliWidget classes, so that those methods can be used where needed.