Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:129370
Feedback-ID: ifab94697:Fastmail
Precedence: list
MIME-Version: 1.0
Date: Fri, 21 Nov 2025 15:33:43 +0100
To: internals@lists.php.net
Message-ID: <f388fc8b-6237-4c7c-9e4c-31a02f7f2daa@app.fastmail.com>
In-Reply-To: 
 <CAMW7n8Dogjj2JFJBNArk-BfvOHRqHKMAFReXJAhDm2UPV1TF7A@mail.gmail.com>
References: 
 <CAMW7n8D0QshP8VhGOUqVnC+Uwf88TC_yy_mQdeNZW_go5YLdcw@mail.gmail.com>
 <C58ADF49-6E21-4E25-8859-2A0769829E38@php.net>
 <e8625214-1c6f-40b8-b24f-e592e6f92012@beccati.com>
 <CAMW7n8Dy3=QyCrG4BMwVXWAipgOr2Vyhy_jbLVTYeq2Jc5v15g@mail.gmail.com>
 <dbc31109-a43d-4afc-9f4f-62c6439328e5@rwec.co.uk>
 <ea59722c-2c84-4a59-a442-a8b217ff577d@gmail.com>
 <b151687a-f4bd-48cf-90e6-440f763ccd6b@rwec.co.uk>
 <CAMW7n8CLZ7yFhUhh89wx7L+YwQ25LA7HZoGWUFB3R76KgXko2w@mail.gmail.com>
 <CAEKnhAHRPbiU4auDuTtFL2ueNnACjmYm_ff9c3VciYE9HfeGZg@mail.gmail.com>
 <CAMW7n8D5B=XVX_Jes+LV=2fPwjQr_RBVphPi_Cxm8-_Rwpdo_Q@mail.gmail.com>
 <CADK1yX+jGtqQVYFTo5YmkenPrNcXz7Dd7RCUKWGrq+50pnKOnw@mail.gmail.com>
 <CAMW7n8Dogjj2JFJBNArk-BfvOHRqHKMAFReXJAhDm2UPV1TF7A@mail.gmail.com>
Subject: Re: [PHP-DEV] [VOTE] True Async RFC 1.6
Content-Type: multipart/alternative;
 boundary=579c8032fb9b442e84494e06ebf558a6
From: rob@bottled.codes ("Rob Landers")

--579c8032fb9b442e84494e06ebf558a6
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable

I'm not sure why the To/CC list got so long. Please reply to the list, n=
ot to each person...

On Fri, Nov 21, 2025, at 14:45, Edmond Dantes wrote:
> Hello.
>=20
> To use AMPHP you don=E2=80=99t need to run its server. But let=E2=80=99=
s simplify the situation.
>=20
> You have Fiber.
> With Fiber you can build an EventLoop + Scheduler, create two
> coroutines, and then run code inside them that, for example, renders a
> template.
>=20
> It will require a fair amount of code, but it=E2=80=99s possible.
> At the same time, ob_start/ob_end will immediately break, because they
> don=E2=80=99t support Fiber switching. This is a known bug.
> And if you pass global state into closures, and that state changes
> unpredictably, the application logic will essentially break.

I don=E2=80=99t know why you=E2=80=99d build all of that from scratch. h=
ttps://revolt.run/ will allow you to run amphp/reactphp in any request. =
It "just works"

You do need to consciously kick off the event loop, though, as well as i=
nstall some specific extensions, or you=E2=80=99re stuck with the stream=
_select implementation, which isn=E2=80=99t the best.

That being said, the only problem you=E2=80=99d run into trying to insta=
ll this in a WordPress plugin is if *any other plugin* is using Fibers o=
r another Fiber library. From experience, they don=E2=80=99t always play=
 nicely together, and there should only be one event loop running at a t=
ime.

> **What=E2=80=99s different with TrueAsync**
>=20
> 1. You don=E2=80=99t need to write a lot of code. Yes, just spawn();
> 2. `ob_start` and `ob_end` will work correctly.
> 3. If closures share a variable that multiple coroutines write to
> unpredictably, the application will still break.

Without well-defined suspension points, literally all code is unpredicta=
ble. This is what people have been trying to tell you for the last sever=
al weeks. How can I know when the code will yield to the scheduler? How =
do I reason about it? The foundation is there in the RFC and the code, b=
ut it needs more definition and guarantees.

> In other words, no matter how asynchrony is implemented in PHP (and
> other langs), it will always require correct handling of shared
> memory.
> And I want to clarify a bit. This is not about a global variable. It=E2=
=80=99s
> about shared memory. These are not the same things. A local variable
> can be passed into a closure by reference.
> But if you don=E2=80=99t call WP() inside a coroutine, nothing prevent=
s you
> from using asynchrony.

PHP is full of shared memory, even applications. Take, for example:

$val =3D $this->cache->get('key') ?? $this->cache->set('key', 'value');

Currently in PHP, this is unsafe if the cache is a proper cache (like me=
mcached/redis), but for an in-memory implementation, it=E2=80=99s effect=
ively a Check-And-Set. If the cache is implemented as async, then it bec=
omes unsafe even for in-memory implementations. This may-or-may-not be d=
esired.

However, this gets weird, very quickly, very fast. $_SERVER/$_REQUEST/$_=
ENV/etc are all global memory that usually doesn=E2=80=99t change. Howev=
er, frameworks change it all the time, to inject environment variables f=
rom .env files, to sanitise inputs, etc. Swoole avoids this by having a =
Request object. This is simply not present in PHP.

Global state is a core part of PHP, like it or not.

=E2=80=94 Rob
--579c8032fb9b442e84494e06ebf558a6
Content-Type: text/html; charset=utf-8
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE html><html><head><title></title></head><body><div>I'm not sure=
 why the To/CC list got so long. Please reply to the list, not to each p=
erson...</div><div><br></div><div>On Fri, Nov 21, 2025, at 14:45, Edmond=
 Dantes wrote:</div><blockquote type=3D"cite" id=3D"qt" style=3D""><div>=
Hello.</div><div><br></div><div>To use AMPHP you don=E2=80=99t need to r=
un its server. But let=E2=80=99s simplify the situation.</div><div><br><=
/div><div>You have Fiber.</div><div>With Fiber you can build an EventLoo=
p + Scheduler, create two</div><div>coroutines, and then run code inside=
 them that, for example, renders a</div><div>template.</div><div><br></d=
iv><div>It will require a fair amount of code, but it=E2=80=99s possible=
.</div><div>At the same time, ob_start/ob_end will immediately break, be=
cause they</div><div>don=E2=80=99t support Fiber switching. This is a kn=
own bug.</div><div>And if you pass global state into closures, and that =
state changes</div><div>unpredictably, the application logic will essent=
ially break.</div></blockquote><div><br></div><div>I don=E2=80=99t know =
why you=E2=80=99d build all of that from scratch.&nbsp;<a href=3D"https:=
//revolt.run/">https://revolt.run/</a>&nbsp;will allow you to run amphp/=
reactphp in any request. It "just works"</div><div><br></div><div>You do=
 need to consciously kick off the event loop, though, as well as install=
 some specific extensions, or you=E2=80=99re stuck with the stream_selec=
t implementation, which isn=E2=80=99t the best.</div><div><br></div><div=
>That being said, the only problem you=E2=80=99d run into trying to inst=
all this in a WordPress plugin is if&nbsp;<b>any other plugin</b> is usi=
ng Fibers or another Fiber library. From experience, they don=E2=80=99t =
always play nicely together, and there should only be one event loop run=
ning at a time.<br></div><div><br></div><blockquote type=3D"cite" id=3D"=
qt" style=3D""><div>**What=E2=80=99s different with TrueAsync**</div><di=
v><br></div><div>1. You don=E2=80=99t need to write a lot of code. Yes, =
just spawn();</div><div>2. `ob_start` and `ob_end` will work correctly.<=
/div><div>3. If closures share a variable that multiple coroutines write=
 to</div><div>unpredictably, the application will still break.</div></bl=
ockquote><div><br></div><div>Without well-defined suspension points, lit=
erally all code is unpredictable. This is what people have been trying t=
o tell you for the last several weeks. How can I know when the code will=
 yield to the scheduler? How do I reason about it? The foundation is the=
re in the RFC and the code, but it needs more definition and guarantees.=
</div><div><br></div><blockquote type=3D"cite" id=3D"qt" style=3D""><div=
>In other words, no matter how asynchrony is implemented in PHP (and</di=
v><div>other langs), it will always require correct handling of shared</=
div><div>memory.</div><div>And I want to clarify a bit. This is not abou=
t a global variable. It=E2=80=99s</div><div>about shared memory. These a=
re not the same things. A local variable</div><div>can be passed into a =
closure by reference.</div><div>But if you don=E2=80=99t call WP() insid=
e a coroutine, nothing prevents you</div><div>from using asynchrony.</di=
v></blockquote><div><br></div><div>PHP is full of shared memory, even ap=
plications. Take, for example:</div><div><br></div><div>$val =3D $this-&=
gt;cache-&gt;get('key') ?? $this-&gt;cache-&gt;set('key', 'value');</div=
><div><br></div><div>Currently in PHP, this is unsafe if the cache is a =
proper cache (like memcached/redis), but for an in-memory implementation=
, it=E2=80=99s effectively a Check-And-Set. If the cache is implemented =
as async, then it becomes unsafe even for in-memory implementations. Thi=
s may-or-may-not be desired.</div><div><br></div><div>However, this gets=
 weird, very quickly, very fast. $_SERVER/$_REQUEST/$_ENV/etc are all gl=
obal memory that usually doesn=E2=80=99t change. However, frameworks cha=
nge it all the time, to inject environment variables from .env files, to=
 sanitise inputs, etc. Swoole avoids this by having a Request object. Th=
is is simply not present in PHP.</div><div><br></div><div>Global state i=
s a core part of PHP, like it or not.<br></div><div><br></div><div id=3D=
"sig121229152">=E2=80=94 Rob</div></body></html>
--579c8032fb9b442e84494e06ebf558a6--