Re: Re: [RFC] [Discussion] [VOTE] Rounding Integers as int

Hey Marc, On 18.3.2024 08:53:01, Marc Bennewitz wrote:

Hi Bob, On 17.03.24 14:59, Bob Weinand wrote:

On 17.3.2024 13:23:04, Marc Bennewitz wrote:

Hello internals, I have opened the vote for the "Rounding Integers as int" RFC: https://wiki.php.net/rfc/integer-rounding Do to Easter weekend the vote will run for two weeks and two days until Tue the 2nd of April 2024. Best regards, Marc Bennewitz

Hey Marc, I've voted no; it should be just changed without any force_float parameter. Just always return int when possible (and the input was int). If users wish to have the old behaviour, they should just explicitly cast via (float). The effective BC break of that would be quite small if some things which return float today now would return int. I cannot imagine many cases where this would actually be unwanted. And as said, explicit (float) casts are always possible. I also dislike force_float, as it cannot just be added to a function in any code which shall be backwards compatible to 8.3 and older. It would just emit Uncaught Error: Unknown named parameter $force_float.

Changing the return type from float to int is a non trivial quite hard to find behavior change. Imaging code like this: $x = 800; $y = 800; round($x/$y) === 1.0; This will return false instead of true especially because we teach users to use strict comparison. Such behavior change should be done in a major version.

I see, here we disagree: - Strict comparison should be avoided when working with numbers. Strict comparisons are generally for strings and booleans. - There's no reason to artificially wait years here.

With the additional parameter it's possible to opt-in into the new behavior already in 8.4 while in PHP 9.0 the default behavior will change but previously opted in code does not need to get touched again. Just changing the behavior means waiting for PHP 9.0 without a way to opt-in in 8.4 already. If you are not interested in opting in in 8.4 already you can just ignore the additional argument as this will be the default in 9.0. I'm not interested in having an additional parameter I have to carry forward for quite some years.

To mimic the previous behavior in a fully BC way it's as simple as explicitly casting the value to float.

I would rather have preferred to see a static analysis solution how often round()ed results are compared strictly, assess the actual BC impact and possibly encourage tools like Rector to recognize such patterns.

Bob

Regards, Marc

Bob