31
183
41
Events
trowski push amphp/http-server
Add ForwardedForMiddleware
trowski push amphp/http-server
Add create prefix to static constructors
trowski push amphp/http-server
More docs
RFC: Introduce an interface for `Session` (v3 only)
We never use Interface in any interface names. We can still have BC breaks in the beta, especially for something as easy to fix as a name. So definitely Session as the interface name, and… I dunno for the implementation, I'm terrible at naming things.
trowski push amphp/http-server
On second thought, static constructors
Single log message for request and response
trowski push amphp/http-server
Move header folding test to expected failures
Split SocketHttpServer
trowski push amphp/http-client
Fix test issues revealed after updating amphp/http-server
trowski push amphp/http-server
Update for "raw" to "pairs" rename
trowski push amphp/http-client
Update for amphp/http beta 2
Fix Psalm issues
unexpected segmentation fault
More specifically, 8.1.15-16 and 8.2.2-3 have a GC bug that will segfault in a fiber, which was fixed in 8.1.17 and 8.2.4.
I was considering throwing an exception or triggering an error in the Revolt event loop if the affected versions are used since the application will segfault eventually.
trowski push amphp/http
Change "raw" to "pairs"
trowski push amphp/parallel
Allow calling Context::join() immediately
trowski push amphp/parallel
Include context backtrace in message (#168)
Include context backtrace in message
PHP does not call __toString() on a previous exception, so our decision to wrap ContextPanicError in a ContextException had the unfortunate side-effect of removing the context trace from the output of an uncaught exception.
There's a couple ways we can approach this. This PR puts the trace as a string into the message. This has the side-effect of needing to create a stack-trace string for uncaught exceptions from Tasks, but is mostly negligible I'd think.
Another approach would be to use reflection to set the $trace property of the parent exception, forcing the child trace to be the trace of the ContextPanicError object.
(new \ReflectionProperty(parent::class, 'trace'))->setValue($this, $originalTrace);
Can't get CancelledException ends the task properly without killing the child process
A closure provided to a Cancellation should not throw. These functions are executed within the event loop, not within the context where they are defined. The Cancellation documentation doesn't make this clear, so I apologize for the confusion.
The cancellation subscriber should instead trigger some action that would then stop the operation and potentially throw from the original context, such as failing a deferred.
$deferred = new DeferredFuture();
$cancellation->subscribe(fn ($e) => $deferred->error($e));
return $deferred->await();
Many APIs can accept an optional Cancellation, such as HttpClient::request() in amphp/http-client, eliminating the need to set up a subscriber yourself. delay() is another such function. Your example could be simplified to:
class Job implements Task {
public function run(Channel $channel, Cancellation $cancellation): bool {
delay(2, cancellation: $cancellation); // fake work, longer than the timeout
return true;
}
}
