Futures can have a registered Callback of type
std::function<void(const uint32_t&)> which gets called in a new
Fiber with the result of the IO Request.
Note the first completion will cause the execution of a callback and
partial completion support must be implemented manually in the callback.

Callbacks are stored in a heap allocated std::function on registration
and are freed by the new Fiber after the callback returned;

The Future with a registered Callback is not referenced in any way in
the IO subsystem and therefore can be dropped after being submitted.
This also means that a Future with a registered callback will not be
signaled by the IO subsystem on completion.
If signaling is desired one must implement it manually in the registered
callback.

The gtest wrap can be updated with the meson wrap command

Use getaddrinfo instead to get the socket information instead of
using always ipv4 sockets.
Pass host and port as strings for use with getaddrinfo.

Florian Fischer authored 4 years ago and Florian Schmaus committed 4 years ago

Modified-by: Florian Schmaus <flow@cs.fau.de>

Florian Fischer authored 4 years ago and Florian Schmaus committed 4 years ago

On our bigboxes the build failed because both liburing.h and
linux/time_types.h define struct __kernel_time_spec.
The linux headers were added because of iwyu and are not unnecessary.

Modified-by: Florian Schmaus <flow@cs.fau.de>

To skip test if emper::IO is false the new io test runner in
tests/test-runner is introduced which checks emper::IO and skips the
test otherwise executes our normal testMain function.

Fixes #10.

Concurrently connecting and creating the server socket is racy.
Now the server socket is created before any Fiber is started
removing this race.
Note: The same race is theoretically in SimpleNetworkTest and
ConcurrenctEchoTest too but I haven't observed the failure yet

Empers IO design is based on a proactor pattern where each worker
can issue IO requests through its exclusive IoContext object which wraps an
io_uring instance.

IO completions are reaped at 4 places:
1. After a submit to collect inline completions
2. Before dispatching a new Fiber
3. When no new IO can be submitted because the completion queue is full
4. And by a global completer thread which gets notified about completions
   on worker IoContexts through registered eventfds

All IO requests are modeled as Future objects which can be either
instantiated and submitted manually, retrieved by POSIX-like non-blocking
or implicitly used by posix-like blocking functions.

User facing API is exported in the following headers:
* emper/io.hpp (POSIX-like)
* emper.h (POSIX-like)
* emper/io/Future.hpp

Catching short write/reads/sends and resubmitting the request without
unblocking the Fiber is supported.

Using AlarmFuture objects Fibers have a emper-native way to sleep for
a given time.

IO request timeouts with TimeoutWrapper class.
Request Cancellation is supported with Future::cancel() or the
CancelWrapper() Future class.

A proactor design demands that buffers are committed to the kernel
as long as the request is active. To guaranty memory safety Futures
get canceled in their Destructor which will only return after the committed
memory is free to use.

Linking Futures to chains is supported using the Future::SetDependency()
method. Future are submitted when their last Future gets submitted.
A linked Request will start if the previous has finished.
Error or partial completions will cancel the not started tail of a chain.

TODO: Handle possible situations where the CQ of the global completer is full
and no more sqe can be submitted to the SQ.

This introduces AbstractWorkStealingScheduler which holds the common
work-stealing scheduling strategy.

We see SimpleActorTest timing out a lot in our CI when LAWS is used,
and from the logs it appears likely that this is a valid
timeout (versus the test just being stuck). Let us increase the test
timeout and see if it helps. Or if it is a "real" timeout.

We see SimpleActorTest timing out a lot in our CI, and from the logs
it appears likely that this is a valid timeout (versus the test just
being stuck). Let us increase the test timeout and see if it helps.

A spurious wake-up can be produced by the new UnblockOnMainActorTest which
triggers the assert(!mpscQueue.empty()) in UnboundedBlockingMpscQueue::get.

Those spurious wake-ups are possible because the push and wake-up pair in
UnboundedBlockingMpscQueue::put are not atomic.
The following sequence diagram demonstrates a spurious wake-up:

   T1          T2            Q
   .           .            { }
  put(e)       .            { }
 push 54-57    .            {e}
   .         get()          {e}
   .        consume e       { }
   .           .            { }
   .         get()          { }
   .         block          { }
 unblock       .            { }
   .           .            { }
   .         wakeup         { }
   .           .            { }
               X
       assert(!queue.Empty())

To deal with spurious wake-ups we recheck the wake-up condition (a non empty queue)
and block again if we find it empty.
We assume spurious wake-ups are rare because it was difficult to reproduce them
even with a dedicated Test (the new UnblockOnMainActorTest) therefore we declare
the empty queue branch as unlikely.

Fixes #4.

We are unable to switch to C++20 until
https://github.com/mesonbuild/meson/issues/8084 is fixed.

In debug builds, the schedule() method is now guarded by an assert(),
as due to this change, the schedule() method will no longer work from
everywhere.

This also improves the worker sleep method.

This reverts commit b33ac93c.

Now that logging is done independently from NDEBUG, we can re-enable
"undef NDEBUG for test executables". The rationale that we want to
undefine NDEBUG for test executable is that we want, even in release
builds, asserts in tests enabled. Note that this also applies to
asserts in the emper core library, if those are in *headers*.

Disable the userpace-rcu support by default.
Our used userspace-rcu flavor memb is rather new and not available in
liburcu version 0.10 available in debian buster.
This change switches from using DCE and "if constexpr" to the C
preprocessor so the library is only needed when the userspace-rcu support
is actually enabled.

And fix the #include issues reported by the newer IWYU version.

The explicit -UNDEBUG was introduced in the CMake to meson migration (62900cd1)
but as far as I can tell it was not present in the CMake tests definition.

It makes tests less reliable because it enables only some debug messages but not
all.
For example if emper is build with NDEBUG a test using a Semaphore
will output debug messages from the Semaphore but not from the Runtime.

Introduce a new Actor test using BinaryPrivateSemaphores and an Actor.
Multiple fibers are created which create a BPS on the stack, submit it to the actor
and wait on the semaphore.
The Actor simply signals each semaphore it receives.

The SimpleURCUTest creates a lock-less hash table and inserts values from
separate fibers and verifies the correct insertion.

This allows to run indiviual tests using the meson runner, e.g.

meson --repeat 100 SimpleActorTest

This should not change much, but ensure that this method is also
invoked in the test (even though it is not explicitly tested).

Otherwise meson will use a random value.

This test fails for yet unknown reasons.