Everything

php

Make Closure covariant to callable

php.internals

Gina P. Banyard

1 year ago

Hello internals,

Ondřej Mirtes (author of PHPStan) remarked to me in August that Closure is not a subtype of callable.
The implementation of this is quite straight forward and can be seen on GitHub. [1]
The question is if there are any concern or not and if this requires an RFC or can just be merged as is.

Best regards,

Gina P. Banyard

[1] https://github.com/php/php-src/pull/15492

Rob Landers

1 year ago

On Tue, Nov 12, 2024, at 14:53, Gina P. Banyard wrote:

> Hello internals,
>
> Ondřej Mirtes (author of PHPStan) remarked to me in August that Closure is not a subtype of callable.
> The implementation of this is quite straight forward and can be seen on GitHub. [1]
> The question is if there are any concern or not and if this requires an RFC or can just be merged as is.
>
> Best regards,
>
> Gina P. Banyard
>
> [1] https://github.com/php/php-src/pull/15492
>

Hey Gina,

I think saying “Closure should be a subtype of callable” would require an RFC. To say that “Closure should satisfy the callable type check” is just an oversight/bug. They are functionally the same IMO, just different perspectives.

— Rob

Claude Pache

1 year ago

> Le 12 nov. 2024 à 14:53, Gina P. Banyard <internals@gpb.moe> a écrit :
>
> Hello internals,
>
> Ondřej Mirtes (author of PHPStan) remarked to me in August that Closure is not a subtype of callable.
> The implementation of this is quite straight forward and can be seen on GitHub. [1]
> The question is if there are any concern or not and if this requires an RFC or can just be merged as is.
>
> Best regards,
>
> Gina P. Banyard
>
> [1] https://github.com/php/php-src/pull/15492

Hi Gina,

That seems reasonable.

But personally, I would be more interested in having a type equivalent to `callable&object`, that is all objects that implement the `__invoke()` magic method (this is a strict super-type of \Closure).

—Claude

Larry Garfield

1 year ago

On Tue, Nov 12, 2024, at 10:43 AM, Claude Pache wrote:

>> Le 12 nov. 2024 à 14:53, Gina P. Banyard <internals@gpb.moe> a écrit :
>>
>> Hello internals,
>>
>> Ondřej Mirtes (author of PHPStan) remarked to me in August that Closure is not a subtype of callable.
>> The implementation of this is quite straight forward and can be seen on GitHub. [1]
>> The question is if there are any concern or not and if this requires an RFC or can just be merged as is.
>>
>> Best regards,
>>
>> Gina P. Banyard
>>
>> [1] https://github.com/php/php-src/pull/15492
>
>
> Hi Gina,
>
> That seems reasonable.
>
> But personally, I would be more interested in having a type equivalent
> to `callable&object`, that is all objects that implement the
> `__invoke()` magic method (this is a strict super-type of \Closure).
>
> —Claude

I am in favor of Closure being covarant to callable, and to a new pseudo-type for Closure|Object-with-__invoke.  (Maybe Invokable?)  I do think they'd require brief RFCs, but both would have my support.

--Larry Garfield

Gina P. Banyard

1 year ago

On Tuesday, 12 November 2024 at 16:43, Claude Pache <claude.pache@gmail.com> wrote:

> But personally, I would be more interested in having a type equivalent to `callable&object`, that is all objects that implement the `__invoke()` magic method (this is a strict super-type of \Closure).
>
> —Claude

An Invokable interface, similar to Stringable would be the solution to this.
And this is something that Ondřej also brought up in the PR.
I would be in favour for such an interface (considering the cat is already out of the bag of auto-implementing things with Stringable).
But I don't think I have the time/bandwidth to work on this, so if anyone wants something "easy" to work on that would be it.

Best regards,

Gina P. Banyard

Levi Morrison

1 year ago

There are some weird cases with `callable` which is why they aren't
allowed as a property type. I don't remember all the edges, but if we
are sure that we cannot hit those with simple co-/contra-variance,
then I don't think this needs an RFC and we can just merge the PR.

Gina P. Banyard

1 year ago

On Wednesday, 13 November 2024 at 00:46, Levi Morrison <levi.morrison@datadoghq.com> wrote:

> There are some weird cases with `callable` which is why they aren't
> allowed as a property type. I don't remember all the edges, but if we
> are sure that we cannot hit those with simple co-/contra-variance,
> then I don't think this needs an RFC and we can just merge the PR.

The reason that callable is not allowed as a property type is due to scope/method visibility issues.
See the discussion section of Nikita's Deprecate partially supported callables RFC: https://wiki.php.net/rfc/deprecate_partially_supported_callables#discussion
The callable type is allowed as a parameter and return type, and Closure is "better" defined than just callable.

Best regards,

Gina P. Banyard

Levi Morrison

1 year ago

> > There are some weird cases with `callable` which is why they aren't
> > allowed as a property type. I don't remember all the edges, but if we
> > are sure that we cannot hit those with simple co-/contra-variance,
> > then I don't think this needs an RFC and we can just merge the PR.
>
> The reason that callable is not allowed as a property type is due to scope/method visibility issues.
> See the discussion section of Nikita's Deprecate partially supported callables RFC: https://wiki.php.net/rfc/deprecate_partially_supported_callables#discussion
> The callable type is allowed as a parameter and return type, and Closure is "better" defined than just callable.

Then personally I see no reason to make this an RFC and would
recommend merging it in as a gap in existing features (not really a
new feature).

Niels Dossche

1 year ago

On 12/11/2024 14:53, Gina P. Banyard wrote:

> Hello internals,
>
> Ondřej Mirtes (author of PHPStan) remarked to me in August that Closure is not a subtype of callable.
> The implementation of this is quite straight forward and can be seen on GitHub. [1]
> The question is if there are any concern or not and if this requires an RFC or can just be merged as is.
>
> Best regards,
>
> Gina P. Banyard
>
> [1] https://github.com/php/php-src/pull/15492

Hi Gina

Makes sense to me.

Kind regards
Niels