Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:102199
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain gmail.com designates 209.85.223.195 as permitted sender)
MIME-Version: 1.0
References: <0d108361-f5fa-fe76-b6ad-66e23bfa88f5@loot.at> <3c36b189-108d-bce7-bea0-7a3d6a432607@gmail.com>
 <CAHWSxWTc0pUSdhXj_=+vb6XjVpsMG3C4JFRXLXGU7vUApONTHg@mail.gmail.com> <CALKiJKoGOaDh2zcq0SZzXSusni=PAZru5ZoVB078AqRy6OqUrQ@mail.gmail.com>
In-Reply-To: <CALKiJKoGOaDh2zcq0SZzXSusni=PAZru5ZoVB078AqRy6OqUrQ@mail.gmail.com>
Date: Fri, 8 Jun 2018 01:29:38 -0600
Message-ID: <CAHWSxWQqLojt6-dOo_2WoGOtXcseprJ9ZKrjSGFPFyYivyUWFw@mail.gmail.com>
To: Rowan Collins <rowan.collins@gmail.com>
Cc: Internals <internals@lists.php.net>
Content-Type: multipart/alternative; boundary="00000000000059031f056e1c5e17"
Subject: Re: [PHP-DEV] Better types without runtime checking
From: ryan.jentzsch@gmail.com (Ryan Jentzsch)

--00000000000059031f056e1c5e17
Content-Type: text/plain; charset="UTF-8"

Thanks Rowan for the detailed explanation.
I thought that leveraging `__get` / `__set` internally would have addressed
many of the type checking issues that you mentioned.
I've used __get/__set in userland to enforce property types and hoped that
an internal C based solution would something that is reasonable to code and
implement.

Also, can anyone in internals respond to this inquiry?
https://github.com/php/php-src/pull/1797#issuecomment-394434391


On Wed, Jun 6, 2018 at 4:00 AM Rowan Collins <rowan.collins@gmail.com>
wrote:

> On 6 June 2018 at 07:14, Ryan Jentzsch <ryan.jentzsch@gmail.com> wrote:
>
>> Why would something like this not work?
>>
>> strict class MyClass
>> {
>>     protected int $foo = 1;
>>     public string $bar = "strict keyword in front of class
>> allows/enforces strict properties";
>>     private string $isItReallyThatDifficult = "to implement this?";
>> }
>>
>
>
> The problem is not in defining what code has type annotations, it's what
> to do with them afterwards.
>
> The performance hit comes from the fact that *every* time you run:
>
> $foo->bar = $baz;
>
> The engine now needs to check whether $foo is a class with type
> constraints, and what the type constraint of bar is, and whether the type
> of $baz matches that constraint. The answer will be "no constraint" most of
> the time, but it doesn't know that until it checks, at run-time, every time.
>
> Nor is it just straight-forward assignments that would need to change,
> there are knock-on effects across the language, such as assignment by
> reference:
>
> strict class MyClass
> {
>     public int $foo;
>     public string $bar;
> }
>
> $a = new MyClass;
> $inst =& $a->foo; // If this is allowed...
> $inst = 'bad'; // ...then this needs to error
>
> $s = 'good';
> $a->bar =& $s; // And if this is allowed...
> $s = 42; // ... then this needs to error
>
> The previous proposal for typed properties simply raised errors when you
> tried to use reference assignment with any typed property, which feels like
> an arbitrary restriction to me. The alternative is to allow *every variable
> in the language* to carry type constraint information. I mused some more on
> this here: http://rwec.co.uk/q/php-type-system
>
> In short, yes, it really is that difficult.
>
> Regards,
> --
> Rowan Collins
> [IMSoP]
>

--00000000000059031f056e1c5e17--