Cleaned form data should not be deleted if other data is invalid.

[Ben Slavin]

Currently, when validating a newforms Form object (.full_clean()), the cleaned_data attribute is deleted if any of the fields on the form are invalid.

Data that has been cleaned should be retained, rather than deleted.

Some sample use cases are:

1. Allowing partial previews (discussed in #5153)
2. Maintain file-uploads across submits of a form
3. Prevent having to perform expensive operations twice (hash the value and compare on resubmit)

Discussed here ​http://groups.google.com/group/django-developers/browse_thread/thread/a23807d32cc3042d/58a4ea6a2e3e8a77.

[Ben Slavin]

cleaned_data is persisted for fields that do not have validation errors. (corrected typo in previous)

[Ben Slavin]

Now with docs and tests.

[Malcolm Tredinnick]

Is the change to the _data() method really necessary?

I'd tend to lean towards displaying the data exactly as the user entered it when we are redisplaying the form after an error. This part of the patch changes what the user entered to the normalised form, which may be more confusing than helpful. I'm not going to reject it out of hand in case there's a subtlety I'm missing, but more explanation required for that part, please.

[Ben Slavin]

Replying to mtredinnick:

Is the change to the _data() method really necessary?

I'd tend to lean towards displaying the data exactly as the user entered it when we are redisplaying the form after an error.

You're absolutely right on this. The use-case I'm using as a reference is #2 from the ticket description (preserving file uploads across submits). In my own solution to that problem, I had to hack data to pull the information from the normalization process. Looking more closely, this was the wrong way of handling things since it has unintended (and potentially harmful/confusing) results.

I'm attaching a new patch which introduces a cleaned_data property to the BoundField class. This means that normalized data will be easily accessible, but the behavior of _data() will remain unchanged.

What do you think of this approach?

I should have thought about this when I had to go changing so many tests.

[Ben Slavin]

Spoke with Malcolm, adding 1.0 milestone.... he doesn't think it's likely to be backwards incompatible, but we should give it a bit more thought.

[Malcolm Tredinnick]

Pushing to after 1.0 because it adds features. We've just got too much to do to nurse the bugs that might be caused by even little additions at the moment. Sorry, Ben.

[(none)]

Milestone post-1.0 deleted

[Colin Copeland]

this ticket is somewhat related to #10828

[Julien Phalip]

The tests would need to be rewritten using unittests since this is now Django's preferred way.

[Claude Paroz]

Refresh the patch

[Claude Paroz]

I just attached a new patch for current trunk. I didn't include the cleaned_data for the bound field, as it might be a separate issue, if I understand it well. Comments welcome.

[Thomas Güttler]

I applied 5524-new.diff and all of my 280 unittests passed. I read the patch and think it can be committed.

[ris]

I often find myself using the pattern

form = BananaForm ( request.GET )

if not form.is_valid ():
	# probably very APIFRAGILE, but we need the remnants of the cleaned_data
	# containing the bits of form that did validate.
	form.cleaned_data = {}
	form._clean_fields ()
	form = BananaForm ( initial = form.cleaned_data )

so I can supply arbitrary pre-filled-out form elements in a GET string to a view that only usually POSTs to itself. Doing it this way allows just the supplied (and I suppose valid) elements to be filled in without nasty error messages for e.g. required fields that weren't.

I would love to be able to remove such flaky bits from my code.

I would have personally suggested moving/renaming cleaned_data to something else (partial_cleaned_data?) when overall validation fails rather than just not deleting it. Some existing code may be implicitly relying on the existence of cleaned_data to signify the form as being valid.

[Claude Paroz]

Replying to ris:
(...)

Some existing code may be implicitly relying on the existence of cleaned_data to signify the form as being valid.

The docs clearly specify that this behaviour may change in the future, so they have been warned (sorry for them).

[Aymeric Augustin]

This change should be explained in the release notes. Besides this the patch looks pretty good.

[Claude Paroz]

Added release notes section

[Aymeric Augustin]

I'm uploading a new patch with the following changes:

• updated patch against latest revision of trunk
• kept cleaned_data when a modelform fails a uniqueness check or a date uniqueness check (and I'm not totally confident on these changes)
• replaced a hasattr(form, 'cleaned_data') with form.is_valid() in the formtools tests
• updated docs/ref/forms/validation.txt to account for this change
• rephrased the docs slightly

[Thomas Güttler]

I applied 5524-new-3.diff and all of my 296 unittests passed. I read the patch and think it can be committed. But the release notes should be updated, too.

[Simon Charette]

The documentation should be changed to reflect that it has been changed in 1.5 and not 1.4

[Simon Charette]

I'll update the patch.

[Simon Charette]

Doc update and unicode_literals related changes

[Simon Charette]

Since the patch was originally written by a core developer and marked as RFC by another one and it's really just a doc update to reflect the changes are taking effect in 1.5 I'm moving back to RFC. Feel free to move back to accepted. All tests still pass on Python 2.7.3rc1 sqlite3.

[Claude Paroz <claude@…>]

In [121fd109de09ece4e263e508f9034df9b583da46]:

Fixed #5524 -- Do not remove cleaned_data when a form fails validation

cleaned_data is no longer deleted when form validation fails but only
contains the data that did validate.
Thanks to the various contributors to this patch (see ticket).