Add range for float

[Khartir]

Feature request

Add support for float-ranges similar to what already exists for int-ranges.

I noticed this when working with ceil() (https://phpstan.org/r/80293019-a510-47c5-9f73-36af8786ac3f). I assumed it was the ceil-function that stripped the range from the input but then I noticed that there is no support for float-ranges in PHPStan. Is there a reason for this, or is this just something that hasn't come up yet?

Did PHPStan help you today? Did it make you happy in any way?

It's very satisfying to slowly remove errors from our huge baseline.

[VincentLanglet]

I take a look and at first implementation seems easy by copying IntegerRangeType.

But an issue I see with FloatRange is the fact we might need to exclude the limit sometimes.

If float<0.0, 10.0> represents all the float between 0 and 10, how do I exclude 0 for instance ?

float<0.0, 10.0> ~ 0.0 is not a great syntax and float ~ 0.0 is already not supported...
https://phpstan.org/r/5d999b51-465b-41d0-8340-91e9ad7d8b27

I should be able to have

• [0.0, 10.0]
• ]0.0, 10.0]
• [0.0, 10.0[
• ]0.0, 10.0[

Not sure if you had something in mind @ondrejmirtes @staabm for this

[ondrejmirtes]

I would not start by copying IntegerRangeType. Floats are different. In int<1, 5> you have a finite number of integers but in float<1, 5> you gave an infinite number of floats.

[ondrejmirtes]

Yeah, difference between closed and open interval should also be explored.

[mvorisek]

Floating numbers are not exact, for example 0.1 cannot be represented exactly [1]. The question is what would be the meaning of open vs. closed interval.

If there would be a syntax for (0.1, 0.2) (open interval, ie. without endpoints), what would be the meaning? Will it exclude the "nearest representable 0.1", even if such number is greater than 0.1?

[1] https://3v4l.org/XEvdH

[VincentLanglet]

Floating numbers are not exact, for example 0.1 cannot be represented exactly [1]. The question is what would be the meaning of open vs. closed interval.

If there would be a syntax for (0.1, 0.2) (open interval, ie. without endpoints), what would be the meaning? Will it exclude the "nearest representable 0.1", even if such number is greater than 0.1?

It mostly useful for #13859
positive-float would be ]0, max[

but indeed it's weird for most of other range...