[RFC] Partial Function Application, take 2

3 years ago by Larry Garfield — view source

unread

Hi folks. After much off-list discussion, iteration, and consideration, we have a new draft of PFA ready for review.

The URL is the same:

https://wiki.php.net/rfc/partial_function_application

It's a bit long because we wanted to be as precise as possible. However, in short:

Partial application creates a closure object you can use like any other.
A ? indicates "exactly one required parameter here"
A ... indicates "zero or more parameters here"
If the pattern in the partial is compatible with the underlying function, everything "just works"
If it's not compatible, you get an error.
Named placeholders are not supported.
Splat (argument unpacking) is not supported.

That gives us the 3 most important use cases, plus many others:

arbitrary_function(...) works to reference any function safely. (This is what Nikita's mini-scoped proposal did.)

some_func(1, 2, ?, 5) works to reduce any function to single-argument, which makes it useful in many callback situations.

any_func($all, $params, ...) works to provide all the arguments to a function but not call it yet; you can then cal the resulting closure later with no arguments to actually invoke it.

We're reasonably confident that we have all the ins and outs covered, and in a performant way. Because it's such a heavy rework, though, we want to give it enough time for the discussion to settle, so won't call a vote for at least 2 weeks from today (give or take the feedback we get here).

Cheers!

--
Larry Garfield
larry@garfieldtech.com

3 years ago by Mike Schinkel — view source

unread

Hi folks. After much off-list discussion, iteration, and consideration, we have a new draft of PFA ready for review.

The URL is the same:

https://wiki.php.net/rfc/partial_function_application

Really excellent work, all!

It's a bit long because we wanted to be as precise as possible. However, in short:

Partial application creates a closure object you can use like any other.

A ? indicates "exactly one required parameter here"

A ... indicates "zero or more parameters here"

My only comment/request/suggestion is to consider Mark Randall's suggestion to use ...? instead, for the reasons he mentioned in his email to the list:

https://externals.io/message/114157#114666 https://externals.io/message/114157#114666

Plus Levi Morrison seemed to approve:

https://externals.io/message/114157#114667 https://externals.io/message/114157#114667

-Mike

3 years ago by Guilliam Xavier — view source

unread

On Wed, Jun 2, 2021 at 7:47 PM Larry Garfield larry@garfieldtech.com
wrote:

Hi folks. After much off-list discussion, iteration, and consideration,
we have a new draft of PFA ready for review.

The URL is the same:

https://wiki.php.net/rfc/partial_function_application

Hi, thanks all for the reworks! I just thought to something: does it
support nullsafe calls, e.g. $foo?->bar(?)? and if yes, what is the
signature of the Closure when $foo === null (and does it make a
difference if, inside a function, it was created as a local variable $foo = null; vs received as a typed parameter ?Foo $foo)?

Related, for $null === null, is $c = $null->bar(?); / $c = $null::baz(?); an immediate error, or only later when calling $c($arg)?

Regards,

--
Guilliam Xavier

3 years ago by Larry Garfield — view source

unread

On Wed, Jun 2, 2021 at 7:47 PM Larry Garfield larry@garfieldtech.com
wrote:

Hi folks. After much off-list discussion, iteration, and consideration,
we have a new draft of PFA ready for review.

The URL is the same:

https://wiki.php.net/rfc/partial_function_application

Hi, thanks all for the reworks! I just thought to something: does it
support nullsafe calls, e.g. $foo?->bar(?)? and if yes, what is the
signature of the Closure when $foo === null (and does it make a
difference if, inside a function, it was created as a local variable $foo = null; vs received as a typed parameter ?Foo $foo)?

I just checked, and it... sort of supports nullsafe. Rather, if you try to partial a method on a null object, you get null back for your closure. Then when you try to call it, you get a "cannot invoke null" error. Which... I think makes sense.

cf: https://3v4l.org/4XeB3/rfc#focus=rfc.partials

As a cute side effect, Joe pointed out you could do this:

https://3v4l.org/ERRdY/rfc#focus=rfc.partials

I wouldn't really call that a feature, but more of a side effect of the implementation. Please don't count on that behavior.

I also discovered while checking that nullsafe differentiates between null and undefined. That was unexpected, but not related to the topic at hand. :-)

Related, for $null === null, is $c = $null->bar(?); / $c = $null::baz(?); an immediate error, or only later when calling $c($arg)?

That dies with "call to member function on null" when trying to create the partial:

https://3v4l.org/E6MgK/rfc#focus=rfc.partials

--Larry Garfield

3 years ago by Guilliam Xavier — view source

unread

On Thu, Jun 10, 2021 at 4:34 PM Larry Garfield larry@garfieldtech.com
wrote:

On Wed, Jun 2, 2021 at 7:47 PM Larry Garfield larry@garfieldtech.com
wrote:

https://wiki.php.net/rfc/partial_function_application

for $null === null, is $c = $null->bar(?); / $c = $null::baz(?); an immediate error, or only later when calling
$c($arg)?

That dies with "call to member function on null" when trying to create the
partial:

https://3v4l.org/E6MgK/rfc#focus=rfc.partials

That makes sense (eager evaluation of non-placeholder "arguments",
including the called-on object).

does it
support nullsafe calls, e.g. $foo?->bar(?)? and if yes, what is the
signature of the Closure when $foo === null?

I just checked, and it... sort of supports nullsafe. Rather, if you try
to partial a method on a null object, you get null back for your closure.
Then when you try to call it, you get a "cannot invoke null" error.
Which... I think makes sense.

cf: https://3v4l.org/4XeB3/rfc#focus=rfc.partials

Well I find that unexpected :/

As a cute side effect, Joe pointed out you could do this:

https://3v4l.org/ERRdY/rfc#focus=rfc.partials

So there's a "workaround", although I wouldn't call it precisely "cute"...

I wouldn't really call that a feature, but more of a side effect of the
implementation. Please don't count on that behavior.

I'd rather not indeed ;) Since $null?->whatever(1, 'a') currently always
returns null without error, shouldn't $null?->whatever(?, 'a') return a
closure (with a signature built from the placeholders only) that will
return null when called (i.e. equivalent to fn (mixed $arg1) => null
here)?

Thanks,

--
Guilliam Xavier

3 years ago by Dan Ackroyd — view source

unread

Since $null?->whatever(1, 'a') currently always
returns null without error, shouldn't $null?->whatever(?, 'a') return a
closure (with a signature built from the placeholders only) that will
return null when called (i.e. equivalent to fn (mixed $arg1) => null
here)?

No.

The short circuiting should happen in the same place, no matter what
is to the right of the "?->".

Having the behaviour vary and sometimes not return null, would be
highly surprising.

cheers
Dan
Ack

3 years ago by Guilliam Xavier — view source

unread

On Thu, Jun 10, 2021 at 7:32 PM Guilliam Xavier guilliam.xavier@gmail.com
wrote:

Since $null?->whatever(1, 'a') currently always returns null without
error, shouldn't $null?->whatever(?, 'a') return a closure (with a
signature built from the placeholders only) that will return null when
called (i.e. equivalent to fn (mixed $arg1) => null here)?

Ah, we can also see $foo?->bar(?) as simply "desugaring" to $foo === null ? null : $foo->bar(?), so the current behavior makes sense too (and
maybe even "more")... I guess I was confused to get a "null-implying
error" despite using a "null-safe" call syntax :s (There's also the
possibility of not supporting it, but I guess that would be a lose-lose...)

--
Guilliam Xavier

3 years ago by Guilliam Xavier — view source

unread

Sorry, me again :s I have tested the examples from
https://wiki.php.net/rfc/partial_function_application on
https://3v4l.org/#focus=rfc.partials and several of them currently give an
error:

Ex 10: on the line $c = stuff(?, ?, f: 3.5, ..., p: $point);
=> Fatal error: Named arguments must come after all place holders
(typo I guess, $c = stuff(?, ?, ..., f: 3.5, p: $point); is OK)
(Ex 11: no error but a typo: 'hi' vs 'foo')
Ex 16: for the last call (four(..., d: 4, a: 1))(2, 3);
=> Fatal error: Uncaught ArgumentCountError: four(): Argument #2 ($b) not
passed
(on the function definition line)
((four(..., d: 4, a: 1))(2, 3, 5, 6); idem,
but (four(..., d: 4, a: 1))(b: 2, c: 3); throws an "Unknown named
parameter $b" Error on the call line)
(weird)
func_get_args() and friends: one the last line $f(1, 2); (after $f = f(?);)
=> Fatal error: Uncaught Error: too many arguments for application of f, 2
given and a maximum of 1 expected
(can make sense, e.g. https://externals.io/message/114532#114554 )
(Callable reference: no error but a typo: $f vs $foo)
Optimizations: on the line $boo = $baz(4, ...);
=> Fatal error: Uncaught Error: too many arguments and or place holders for
application of Closure::__invoke, 1 given and a maximum of 0 expected
($boo = $baz(?); throws the same error,
but $boo = $baz(4); throws a "not enough arguments for implementation of
foo, 4 given and exactly 5 expected" Error,
and $boo = $baz(...); makes the subsequent $boo(5); throw a "not enough
arguments ..." Error)
(weird, looks like $bar = $foo(2, ...); and/or $baz = $bar(3, ...);
dropped too many params)

Regards,

--
Guilliam Xavier

3 years ago by Joe Watkins — view source

unread

There's a couple of typos in the RFC, which Larry will fix when he has time.

There was also a typo in patch, and a fault in patch.

All fixed. pending tests, 3v4l won't update until tomorrow.

Cheers
Joe

On Fri, 11 Jun 2021 at 13:02, Guilliam Xavier guilliam.xavier@gmail.com
wrote:

Sorry, me again :s I have tested the examples from
https://wiki.php.net/rfc/partial_function_application on
https://3v4l.org/#focus=rfc.partials and several of them currently give an
error:

Ex 10: on the line $c = stuff(?, ?, f: 3.5, ..., p: $point);
=> Fatal error: Named arguments must come after all place holders
(typo I guess, $c = stuff(?, ?, ..., f: 3.5, p: $point); is OK)

(Ex 11: no error but a typo: 'hi' vs 'foo')

Ex 16: for the last call (four(..., d: 4, a: 1))(2, 3);
=> Fatal error: Uncaught ArgumentCountError: four(): Argument #2 ($b) not
passed
(on the function definition line)
((four(..., d: 4, a: 1))(2, 3, 5, 6); idem,
but (four(..., d: 4, a: 1))(b: 2, c: 3); throws an "Unknown named
parameter $b" Error on the call line)
(weird)

func_get_args() and friends: one the last line $f(1, 2); (after $f = f(?);)
=> Fatal error: Uncaught Error: too many arguments for application of f, 2
given and a maximum of 1 expected
(can make sense, e.g. https://externals.io/message/114532#114554 )

(Callable reference: no error but a typo: $f vs $foo)

Optimizations: on the line $boo = $baz(4, ...);
=> Fatal error: Uncaught Error: too many arguments and or place holders for
application of Closure::__invoke, 1 given and a maximum of 0 expected
($boo = $baz(?); throws the same error,
but $boo = $baz(4); throws a "not enough arguments for implementation of
foo, 4 given and exactly 5 expected" Error,
and $boo = $baz(...); makes the subsequent $boo(5); throw a "not enough
arguments ..." Error)
(weird, looks like $bar = $foo(2, ...); and/or $baz = $bar(3, ...);
dropped too many params)

Regards,

--
Guilliam Xavier