Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:106856
User-Agent: Cyrus-JMAP/3.1.7-186-gf4cb3c3-fmstable-20190904v1
Mime-Version: 1.0
Message-ID: <c4fa3f40-8270-4438-9705-ddd69da0da5b@www.fastmail.com>
In-Reply-To: 
 <CANS-=pco95mG_6AmZhFYJ-ig8UEagZoR-MZ65AVEAm9foDZdhg@mail.gmail.com>
References: 
 <CAJVEVObtRNX_1VWtd35bKKYm_Z9RY=kgx78VY3T3cuFhWMR=5g@mail.gmail.com>
 <CABdc3WpzBP7uYvsHPGeuW7+W+KSZpw7QCGwm706jz2LBS59hpw@mail.gmail.com>
 <CANS-=pco95mG_6AmZhFYJ-ig8UEagZoR-MZ65AVEAm9foDZdhg@mail.gmail.com>
Date: Wed, 04 Sep 2019 11:06:58 -0500
To: "Steffen via internals" <internals@lists.php.net>
Content-Type: text/plain
Subject: Re: [PHP-DEV] Silent Types
From: larry@garfieldtech.com ("Larry Garfield")

On Wed, Sep 4, 2019, at 2:21 AM, Lynn wrote:
> Hi,
> 
> Would this warrant a having mixed types? By making every variable of type
> "mixed" when no type is defined, regardless of what value it receives or is
> initialized with, this would make an opt-in system. This behavior would
> cater developers who prefer only strict types, only dynamic types, and
> those that want to type their code bit by bit without putting declares that
> affect whole files.
> 
> ```
> $foo = 'foo';
> // same as
> mixed $oneHundred = 100;
> // difference being:
> // $foo is accepted when 'string' is typed in the signature
> // $oneHundred is accepted when 'int' is typed in the signature
> 
> // no longer mixed, only accepts their defined type as value
> // and is accepted only as their defined type in signatures
> string $foo = 'foo';
> int $oneHundred = 100;

Yes, an implicit "mixed" type would be the only way it could be done, just like object properties that are not explicitly typed are implicitly "mixed".  Whether there is a "mixed" type for explicit declaration is an open question; if it were added, it would IMO make sense to add for properties and parameters/returns, as well, for consistency.  Whether or not that's wise in general I don't know.

> // does imply we might need custom type definitions in the future to not
> have to work with mixed
> typedef Stringable = string | MyToStringInterface;
> Stringable $foo = (bool) rand(0, 1) ? 'foo' : new MyStringableObject('foo');
> ```

There's a long list of reasons we want that functionality even before we get to variable types. :-)  IIRC, it falls into the popular category of "this would be really useful but is harder than it sounds for complex implementation reasons, and a few people feel it's too much ceremony for a loose language."  There's a number of features stuck in that bucket.

> Note that this behavior would require making some decisions whether or not
> in the future this opt-in behavior should change when a default value is
> given, such as with C# and type inference when declaring a variable, based
> on its assigned value.

I could see arguments either way, but BC would probably dictate that types are not set implicitly.  Otherwise, this currently valid code would break:

$things = 'foo';
...
$things = ['foo', 'bar'];

if (is_array($things)) { ... }
else { ... }

(Whether or not that's good practice is beside the point; there's ample code like that out there and breaking it when we don't have to is a no-no.)

--Larry Garfield