django/contrib/localflavor/usa/forms.py USStateSelect missing blank options

.... it stripped off the blank value ticktick

(Fixed description formatting)

I'm not really sure what to do about this. The blank entry (if we include it), should be inserted by the widget, since it's not part of the data. And possibly only inserted if no choice is selected. We should also do the same thing for every other localflavor widget; nothing special about the US version here.

Patches welcome, certainly. Something to look at might make it easier to decide.

I was looking at this and there are 20 Select widgets which would need to be changed. It feels like there should be a new base widget which deals with the empty choice (The same way there is for the ModelChoiceField.

I could just be over thinking things and making them more complex than necessary.

from itertools import chain

class NullSelect(Select):
    def __init__(self, attrs=None, choices=(), empty_label=u"---------"):
        empty_choice = ((u'', empty_label),)
        super(NullSelect, self).__init__(attrs, chain(empty_choice, choices))

With this addition, the newforms.models.ModelChoiceField would no longer need the empty_label, and would use the newforms.widgets.NullSelect widget instead.

Working on a patch

Didn't notice that it was claimed. Sorry. Is there going to be a new base widget?

I am making two different patches. One w/o the new base, and one with a new base, updated docs, and a regression test.

Simple patch which does not include a new base widget, but adds an empty_label to each localflavor Select widget

NOTE: holding off on setting the 'has_patch' flag until I get the second extended patch done, then will open up the discussion to the developer list.

New NullSelect? widget which is used as the base for all the localflavor select widgets, and replaces the ModelField? default widget,.

The nullselect patch includes documentation and regression tests. It is as up to date as I can make it.

The older simple patch is now out of data and does not include required test fixes. I can easily make a new simple patch if the NullSelect widget is rejected.

This looks pretty good. I'm not 100% happy with the name "NullSelect?", but I'm not sure if there's a better color for that particular bikeshed.

Oh, and this needs a test for NullSelect? itself (there are tests for subclasses, but no the widget itself).

Updated patch with better NullSelect implementation which accepts empty_label on render method, and has full tests

I really don't care for NullSelect either. I thought of other names (EmptySelect, BlankSelect, etc) but NullSelect fit the meme of NullBooleanSelect and others.

I also considered adding the empty_label to the !Select widget it's self, but there is no good way of doing this without breaking most implementations out there. The nice part about NullSelect? is that it does not change any existing behavior.

One option would be to add it to !Select, defaulting to None, and providing a EMPTY_LABEL global for people to use (so that u'--------' is defined somewhere). This would be a rather verbose way of doing it though.

Without this patch I can't make an optional address in my model using USStateField. This should be committed soon.

Hmm, why is this still marked as new? It's been assigned and has a patch. Wierd.

Tim Saylor: if you want to help advance the ticket, feel free to fix the problems that were mentioned above. Complaining adds no value.

What complaints? The only things complained about recently were the name NullSelect?, and Jacob said he can't think of anything better, and a lack of tests for the base NullSelect? class, which dougn appears to have fixed. I'd be glad to help, but I don't know what's needed.

Since this hasn't been touched in 11 months, I'm assuming that the original assignee isn't interested in completing this ticket anymore. For now, I will assign it to myself to test and polish up this patch to be compatible with the current trunk. If dougn would like to continue working on this, feel free to reassign it to him.

Updated patch to make it compatible with r9912. Remove modifications to forms.models

I have updated dougn's patch so it is now compatible with r9912. I removed modifications to forms.models since I don't think we want all model generated choicefields to have this type of widget. Eliminating the empty label option produced failures in the forms test cases. If someone wants to use this widget with their ModelChoiceField?, they can override the widget for the ChoiceField? in kwargs.

What further needs to be done on this issue in order to get this patch accepted into SVN?

#10969 is also an alternate solution to dealing with this issue.

Reverting the summary. This issue was fixed as a part of #10969.