Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:125659
Date: Mon, 23 Sep 2024 21:40:16 +0000
To: "Christoph M. Becker" <cmbecker69@gmx.de>
Cc: Jordan LeDoux <jordan.ledoux@gmail.com>, PHP internals <internals@lists.php.net>
Subject: Re: [PHP-DEV] [Pre-RFC Discussion] User Defined Operator Overloads (again)
Message-ID: <PGn9OsCEjGVpaLetiCVdOVYaDxWe5RJmyYe6KDS7dtjrj_9JQmWuRvA1Zbhd3Zja61Sjnj61XcYeUsN3yay2EQjWCvlilPE5Gqv8CjjZazI=@gpb.moe>
In-Reply-To: <ac5dca88-4336-4a9a-a6d0-32011b00e6be@gmx.de>
References: <CAMrTa2E2m00gOoGy5FbvkrLwM1AUP3ZmJXHkxh_R2qEszQ=NWA@mail.gmail.com> <pgOG2R8lXA-Z6fc0wXKWUiZh1-WLqTp-pJoA6sAVn2UgFeRsCK3hNgeQIq7mrgjF4S2f5GPxZBk2d-O1V9KwbMTGC5o95Eo_fO5E-UvpcYs=@gpb.moe> <ac5dca88-4336-4a9a-a6d0-32011b00e6be@gmx.de>
Feedback-ID: 96993444:user:proton
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From: internals@gpb.moe ("Gina P. Banyard")

On Friday, 20 September 2024 at 15:09, Christoph M. Becker <cmbecker69@gmx.=
de> wrote:

> On 20.09.2024 at 15:00, Gina P. Banyard wrote:
>=20
> > This leads to a rather insane situation where the engine does not assum=
e + is always commutative but does so for *, something which makes no sense=
 as multiplication or product operations are (in the grand scheme of mathem=
atics) rarely commutative.
>=20
>=20
> Gosh, I had completely forgotten about ZEND_TRY_BINARY_OBJECT_OPERATION.
> So I withdraw my statement about the only reasonable way to implement
> operator overloading in PHP[1], and state that there is no reasonable
> way to implement operator overloading in PHP at all (excluding
> comparison and equality operations), since apparently, we cannot even
> get the only two internal bundled classes to properly handle this.
> Cf. https://3v4l.org/gksqI/rfc#vgit.master and
>=20
> https://3v4l.org/o5Uhh/rfc#vgit.master.
>=20
>=20
> [1] https://externals.io/message/125550#125621

The issue here is that GMP does not return FAILURE when it does not support=
 an operand, but instead throws an exception, which blocks any polymorphic =
handling.
Being currently knees deep in the implementation of GMP it does some questi=
onable stuff and should be refactored.
However, the fact that if only one operand is an object and returns FAILURE=
 and the binary operation can still succeed is a bit of an issue which need=
s to be address regardless IMHO.

Best regards,

Gina P. Banyard