On Sep 28, 2015, at 3:29 AM, S.A.N ua.san.alex@gmail.com wrote:> > Are there any future plans for - async/await?> This need to know now, not to use these words to constants, and class names...> > -- > > >
Would this be awesome? Yes. The issue is that all of PHP's built-in IO
operations are currently synchronous. It would/will require a massive
effort to (1) install the infrastructure to offload blocking operations to
worker threads and/or (2) convert many synchronous IO operations to a
non-blocking paradigm. Someone has to do that work ... and it's a lot. I
don't personally see this happening before 8.0 though I'd be happy to be
surprised.
FWIW I'm a big proponent of libuv here and think we could significantly
improve PHP as a language by incorporating it to solve this (and other)
outstanding issues in a cross-OS manner.
In the nearer term, this is exactly the sort of thing the Generator Return
and Generator Delegation RFCs implemented in 7.0 are designed to handle.
With a minimal amount of userland code you can resolve coroutines in
userland as lightweight threads in which IO is accomplished with
non-blocking libraries.
The obvious drawbacks to using yield in userland as a surrogate for
async/await is the proliferation of non-standard userland libraries (of
which several good options already exist) to manage IO, timers and event
loops.
IMHO userland shouldn't know anything about how concurrency is achieved
(whether using threads or nbio). With something like libuv internal C code
would have access to a built-in worker thread pool designed specifically
for a non-blocking paradigm. This would allow us to expose the best of both
worlds.
Just my $0.02 on the topic.
On Sep 28, 2015, at 3:29 AM, S.A.N ua.san.alex@gmail.com wrote:> > Are there any future plans for - async/await?> This need to know now, not to use these words to constants, and class names...> > -- > > >
Would this be awesome? Yes. The issue is that all of PHP's built-in IO
operations are currently synchronous. It would/will require a massive
effort to (1) install the infrastructure to offload blocking operations to
worker threads and/or (2) convert many synchronous IO operations to a
non-blocking paradigm. Someone has to do that work ... and it's a lot. I
don't personally see this happening before 8.0 though I'd be happy to be
surprised.FWIW I'm a big proponent of libuv here and think we could significantly
improve PHP as a language by incorporating it to solve this (and other)
outstanding issues in a cross-OS manner.In the nearer term, this is exactly the sort of thing the Generator Return
and Generator Delegation RFCs implemented in 7.0 are designed to handle.
With a minimal amount of userland code you can resolve coroutines in
userland as lightweight threads in which IO is accomplished with
non-blocking libraries.The obvious drawbacks to using yield in userland as a surrogate for
async/await is the proliferation of non-standard userland libraries (of
which several good options already exist) to manage IO, timers and event
loops.IMHO userland shouldn't know anything about how concurrency is achieved
(whether using threads or nbio). With something like libuv internal C code
would have access to a built-in worker thread pool designed specifically
for a non-blocking paradigm. This would allow us to expose the best of both
worlds.Just my $0.02 on the topic.
I think we can spoonfeed the massive undertaking a bit by spreading it
across several iterations just to get things going. An idea I had was to
just begin by providing coroutines as part of the language. That would
include a \Coroutine class, as well as await / async keywords. This
would at least enable third party async libraries to be more compatible
since they would have the same await types and semantics.
The next step (or at the same time as the first) is to provide some sort
of \EventLoopInterface along with
\EventLoop::getLoop(): \EventLoopInterface
\EventLoop::setLoop(\EventLoopInterface)
\EventLoop::run()
\EventLoop::stop()
\EventLoop::tick()
To start out, just provide the internal coroutine scheduler as a default
implementation. Coroutines would use the "event loop" to schedule resumes.
Doing so would allow even more interoperability if a library chose to
implement \EventLoopInterface in its event loop, where it handles
coroutine scheduling (or lets it fall back to the internal one), and
could also provide I/O "extras".
The final step would be to bring libuv into PHP, itself as a better
default implementation of \EventLoopInterface.
Just my idea.
--
Stephen
I think we can spoonfeed the massive undertaking a bit by spreading it
across several iterations just to get things going. An idea I had was to
just begin by providing coroutines as part of the language. That would
include a\Coroutineclass, as well as await / async keywords. This would
at least enable third party async libraries to be more compatible since they
would have the same await types and semantics.The next step (or at the same time as the first) is to provide some sort of
\EventLoopInterfacealong with\EventLoop::getLoop(): \EventLoopInterface \EventLoop::setLoop(\EventLoopInterface) \EventLoop::run() \EventLoop::stop() \EventLoop::tick()To start out, just provide the internal coroutine scheduler as a default
implementation. Coroutines would use the "event loop" to schedule resumes.Doing so would allow even more interoperability if a library chose to
implement\EventLoopInterfacein its event loop, where it handles
coroutine scheduling (or lets it fall back to the internal one), and could
also provide I/O "extras".The final step would be to bring libuv into PHP, itself as a better default
implementation of\EventLoopInterface.
EventLoop interface, on development stage:
https://github.com/async-interop/event-loop
PHP wrappers for libev and libeio supported PHP 7.
https://pecl.php.net/package/ev
https://pecl.php.net/package/eio
libuv - certainly better because it has everything you need and a huge
community.
Very need async/await in the PHP core (based on generators).
Perhaps there are plans in core developers, for implement async/await?
EventLoop interface, on development stage:
https://github.com/async-interop/event-loop
That's a userland design for event loops; async/await and coroutines
don't necessarily depend on an event loop. They could be added to the
language without an event loop, and simply require the user to provide
the event loop.
My idea was to make a transition into async as simple as possible.
Also, I'm partially responsible for the event loop interface above. :P
PHP wrappers for libev and libeio supported PHP 7.
https://pecl.php.net/package/ev
https://pecl.php.net/package/eiolibuv - certainly better because it has everything you need and a huge
community.
I agree; libuv is probably the best of the bunch, if we include an event
loop library in future versions of PHP. It is not necessary to have a
PHP wrapper for libuv, since if it was included in the interpreter
itself we would be just using the C interface.
Very need async/await in the PHP core (based on generators).
Perhaps there are plans in core developers, for implement async/await?
/Legend speaks of such plans, but they come and go in whispers. Like a
shadow, or a mist from the east. The prophecy spake of such features
targeting PHP8; lo, most believe it to be myth./
--
Stephen
2016-05-04 16:40 GMT+02:00 Stephen Coakley me@stephencoakley.com:
EventLoop interface, on development stage:
https://github.com/async-interop/event-loopThat's a userland design for event loops; async/await and coroutines don't
necessarily depend on an event loop. They could be added to the language
without an event loop, and simply require the user to provide the event
loop.
Would be strange to have async / await in the language without an event
loop, but it's possible, yes.
My idea was to make a transition into async as simple as possible.
Also, I'm partially responsible for the event loop interface above. :P
PHP wrappers for libev and libeio supported PHP 7.
https://pecl.php.net/package/ev
https://pecl.php.net/package/eiolibuv - certainly better because it has everything you need and a huge
community.I agree; libuv is probably the best of the bunch, if we include an event
loop library in future versions of PHP. It is not necessary to have a PHP
wrapper for libuv, since if it was included in the interpreter itself we
would be just using the C interface.Very need async/await in the PHP core (based on generators).
Perhaps there are plans in core developers, for implement async/await?
/Legend speaks of such plans, but they come and go in whispers. Like a
shadow, or a mist from the east. The prophecy spake of such features
targeting PHP8; lo, most believe it to be myth./
Why do we have to wait until PHP 8? Should be mostly backwards compatible
and be fine in 7.2 or so. Issue is probably more deciding on an API.
--
Stephen
/Legend speaks of such plans, but they come and go in whispers. Like a
shadow, or a mist from the east. The prophecy spake of such features
targeting PHP8; lo, most believe it to be myth./Why do we have to wait until PHP 8? Should be mostly backwards compatible
and be fine in 7.2 or so. Issue is probably more deciding on an API.
Yes, I think so, too, do not need to wait for version 8, can start
7.1-2 possible even as another SAPI.
P.S.
These works will accelerate 5x-10x
https://gnugat.github.io/2016/04/13/super-speed-sf-react-php.html
2016-05-04 16:40 GMT+02:00 Stephen Coakley me@stephencoakley.com:
Why do we have to wait until PHP 8? Should be mostly backwards compatible
and be fine in 7.2 or so. Issue is probably more deciding on an API.
I don't think we should wait, I was just thinking that it might not be
ready until that time.
Also, if we plan on rewriting streams and I/O to all be async and use
libuv underneath, that would probably be a BC break unless the existing
functions just become blocking interfaces for a separate async API. If
it was a large BC break, it probably would need to wait for PHP 8.
Now that I think about it, that would be our chance to replace stream
resources with classes...
$file = \php\stream::open("file.txt");
$bytes = await $file->read(1024);
$file->close();
Then rewrite fread() and friends to be aliases that block instead of
await results.
--
Stephen
I don't think we should wait, I was just thinking that it might not be ready
until that time.Also, if we plan on rewriting streams and I/O to all be async and use libuv
underneath, that would probably be a BC break unless the existing functions
just become blocking interfaces for a separate async API. If it was a large
BC break, it probably would need to wait for PHP 8.Now that I think about it, that would be our chance to replace stream
resources with classes...$file = \php\stream::open("file.txt");
$bytes = await $file->read(1024);
$file->close();Then rewrite
fread()and friends to be aliases that block instead of await
results.
This is a great opportunity to normalize API standard libraries PHP :)
$bytes = await $file->read(1024);
It is rarely necessary, more often needs
$result = await $db->query($sql);
and
$result = await $noSql->get($key);