Add attribute 'step' to FloatField.

[Jacob Rief]

If someone wants to use the step attribute as provided by the HTML field
<input type="number" ...> , she/he has to specify that using for instance
FloatField(widget=NumberInput(attrs={'step': 0.5})).

Since the HTML standard offers a step attribute on input fields of type="number",
this feature shall be reflected by Django's FloatField and optionally DecimalField,
rather than having to parametrize the widget.

Min- and max-values are already supported by the FloatField, so the step-value
would make sense here as well. It furthermore would require to revalidate the
step-value by Django's Form validation, rather than by HTML alone.

Patch: ​https://github.com/django/django/pull/14162

[Markus Holtermann]

Thank you. Sounds like a perfectly valid use case.

[Tim Graham]

​django-developers discussion

[Kapil Bansal]

​PR

[Adam Johnson]

On ​the forum, Kapil wrote:

Hi,
I was working on ticket 32559 to add step in FloatField but how to write validation check for this.
Due to python floating point issues, I am not able to validate whether field value is of given step_size or not

The loss of precision in floating points is a real issue. The normal way to validate would be to use the modulo operator to check there is no remainder after dividing by the step, but this isn't possible in floating points with common decimal steps like 0.1:

>>> step = 0.1
>>> 1 % step == 0
False
>>> 1 % step
0.09999999999999995

This problem cannot be solved for FloatField - I suggest we move the ticket to modify DecimalField, as Decimal objects do not have the same problem:

>>> from decimal import Decimal
>>> step = Decimal('0.1')
>>> Decimal(1) % step == 0
True
>>> Decimal(1) % step
Decimal('0.0')

We'd want to enforce that step is given as a Decimal.

The float problem can also occur when constructing a Decimal from a float:

>>> Decimal(0.1)
Decimal('0.1000000000000000055511151231257827021181583404541015625')

Perhaps we should have a check that the given step does not have more than 10 digits after the decimal point?

[Kapil Bansal]

Hi,
I got a reply on mailing list to use math.isclose function. Pull Request is already opened to be reviewed

[Jacob Rief]

I reviewed that pull request and after some testing I came to the conclusion, that math.isclose with a tolerance of 1e-9 is the best solution to fix the floating point rounding errors. All other approaches did not work properly or were far too complicated.