Module Eio.Fiber

both f g runs f () and g () concurrently.

They run in a new cancellation sub-context, and if either raises an exception, the other is cancelled. both waits for both functions to finish even if one raises (it will then re-raise the original exception).

f runs immediately, without switching to any other thread. g is inserted at the head of the run-queue, so it runs next even if other threads are already enqueued. You can get other scheduling orders by adding calls to yield in various places. e.g. to append both fibers to the end of the run-queue, yield immediately before calling both.

If both fibers fail, Exn.combine is used to combine the exceptions.

pair f g is like both, but returns the two results.

all fs is like both, but for any number of fibers. all [] returns immediately.

first f g runs f () and g () concurrently.

They run in a new cancellation sub-context, and when one finishes the other is cancelled. If one raises, the other is cancelled and the exception is reported.

As with both, f runs immediately and g is scheduled next, ahead of any other queued work.

If both fibers fail, Exn.combine is used to combine the exceptions.

Warning: it is always possible that both operations will succeed. This is because there is a period of time after the first operation succeeds when it is waiting in the run-queue to resume during which the other operation may also succeed.

If both fibers succeed, combine a b is used to combine the results (where a is the result of the first fiber to return and b is the second result). The default is fun a _ -> a, which discards the later result.

any fs is like first, but for any number of fibers.

any [] just waits forever (or until cancelled).

n_any fs is like any, expect that if multiple fibers return values then they are all returned, in the order in which the fibers finished.

await_cancel () waits until cancelled.

• raises Cancel.Cancelled

fork ~sw fn runs fn () in a new fiber, but does not wait for it to complete.

The new fiber is attached to sw (which can't finish until the fiber ends).

The new fiber inherits sw's cancellation context. If the fiber raises an exception, Switch.fail sw is called. If sw is already off then fn fails immediately, but the calling thread continues.

fn runs immediately, without switching to any other fiber first. The calling fiber is placed at the head of the run queue, ahead of any previous items.

fork_promise ~sw fn schedules fn () to run in a new fiber and returns a promise for its result.

This is just a convenience wrapper around fork. If fn raises an exception then the promise is resolved to the error, but sw is not failed.

fork_seq ~sw fn creates (but does not start) a new fiber to run fn yield.

Requesting the next item from the returned sequence resumes the fiber until it calls yield x, using x value as the next item in the sequence. If fn returns without producing a value then the result is Seq.Nil (end-of-sequence).

The returned sequence can be consumed safely from another domain. fn itself always runs in the domain that called fork_seq.

Example:

Switch.run @@ fun sw ->
let seq = Fiber.fork_seq ~sw (fun yield ->
    for i = 1 to 3 do
      traceln "Yielding %d" i;
      yield i
    done
  ) in
Seq.iter (traceln "Got: %d") seq

If fn raises an exception then the consumer receives it. If the consumer cancels while awaiting a value, the producer is cancelled when it next calls yield. It is an error to request two items at once, or to request items out of sequence.

• parameter sw

  When the switch finishes, the fiber is cancelled (if still running). Attempting to read from the sequence after this raises an exception.

fork_daemon is like fork except that instead of waiting for the fiber to finish, the switch will cancel it once all non-daemon fibers are done.

The switch will still wait for the daemon fiber to finish cancelling.

The return type of [`Stop_daemon] instead of unit is just to catch mistakes, as daemons normally aren't expected to return.

check () checks that the fiber's context hasn't been cancelled. Many operations automatically check this before starting.

• raises Cancel.Cancelled

  if the fiber's context has been cancelled.

is_cancelled () is true iff check would raise an exception.

yield () asks the scheduler to switch to the next runnable task. The current task remains runnable, but goes to the back of the queue. Automatically calls check just before resuming.

Concurrent list operations.

'a key is a fiber-local variable of type 'a.

Since the key is required to get or set a variable, a library can keep its key private to control how the variable can be accessed.

create_key () creates a new fiber-local variable.

get key reads key from the map of fiber local variables, returning its value or None if it has not been bound.

with_binding key value fn runs fn with key bound to the provided value.

Whilst this binding only exists for the duration of this function on this fiber, it will be propagated to any forked fibers. If fn creates fibers using an external switch, the bound value may be continue to be used after this function returns.

with_binding key value fn runs fn with any binding for key removed.