Skip to content

Latest commit

 

History

History
959 lines (573 loc) · 54.5 KB

S17-concurrency.pod

File metadata and controls

959 lines (573 loc) · 54.5 KB
Raw

TITLE

Synopsis 17: Concurrency

VERSION

Created: 3 Nov 2013

Last Modified: 27 December 2014
Version: 25

This synopsis is based around the concurrency primitives and tools currently being implemented in Rakudo on MoarVM and the JVM. It covers both things that are already implemented today, in addition to things expected to be implemented in the near future (where "near" means O(months)).

Design Philosophy

Focus on composability

Perl 6 generally prefers constructs that compose well, enabling large problems to be solved by putting together solutions for lots of smaller problems. This also helps make it easier to extend and refactor code.

Many common language features related to parallel and asynchronous programming lack composability. For example:

  • Locks do not compose, since two independently correct operations using locks may deadlock when performed together.

  • Callback-centric approaches tend to compose badly, with chains of asynchronous operations typically leading to deeply nested callbacks. This essentially is just leaving the programmer to do a CPS transform of their own logical view of the program by hand.

  • Directly spawning threads on a per-component basis tends to compose badly, as when a dozen such components are used together the result is a high number of threads with no ability to centrally schedule or handle errors.

In Perl 6, concurrency features aimed at typical language users should have good composability properties, both with themselves and also with other language features.

Boundaries between synchronous and asynchronous should be explicit

Asynchrony happens when we initiate an operation, then continue running our own idea of "next thing" without waiting for the operation to complete. This differs from synchronous programming, where calling a sub or method causes the caller to wait for a result before continuing.

The vast majority of programmers are much more comfortable with synchrony, as in many senses it's the "normal thing". As soon as we have things taking place asynchronously, there is a need to coordinate the work, and doing so tends to be domain specific. Therefore, placing the programmer in an asynchronous situation when they didn't ask for it is likely to lead to confusion and bugs. We should try to make places where asynchrony happens clear.

It's also worthwhile trying to make it easy to keep asynchronous things flowing asynchronously. While synchronous code is pull-y (for example, eating its way through iterable things, blocking for results), asynchronous code is push-y (results get pushed to things that know what to do next).

Places where we go from synchronous to asynchronous, or from asynchronous to synchronous, are higher risk areas for bugs and potential bottlenecks. Thus, Perl 6 should try to provide features that help minimize the need to make such transitions.

Implicit parallelism is OK

Parallelism is primarily about taking something we could do serially and using multiple CPU cores in order to get to a result more quickly. This leads to a very nice property: a parallel solution to a problem should give the same answer as a serial solution.

While under the hood there is asynchrony and the inherent coordination it requires, on the outside a problem solved using parallel programming is still, when taken as a whole, a single, synchronous operation.

Elsewhere in the specification, Perl 6 provides several features that allow the programmer to indicate that parallelizing an operation will produce the same result as evaluating it serially:

  • Hyper operators ("Hyper operators" in S03) express that parallel operator application is safe.

  • Junctions ("Junctions" in S09) may auto-thread in parallel.

  • Feeds ("Feed operators" in S06) form pipelines and express that the stages may be executed in parallel in a producer-consumer style relationship (though each stage is in itself not parallelized).

  • hyper and race list operators ("The hyper operator" in S02) express that iteration may be done in parallel; this is a generalization of hyper operators.

Make the hard things possible

The easy things should be easy, and able to be built out of primitives that compose nicely. However, such things have to be built out of what VMs and operating systems provide: threads, atomic instructions (such as CAS), and concurrency control constructs such as mutexes and semaphores. Perl 6 is meant to last for decades, and the coming decades will doubtless bring new ways do do parallel and asynchronous programming that we do not have today. They will still, however, almost certainly need to be built out of what is available.

Thus, the primitive things should be provided for those who need to work on such hard things. Perl 6 should not hide the existence of OS-level threads, or fail to provide access to lower level concurrency control constructs. However, they should be clearly documented as not the way to solve the majority of problems.

Schedulers

Schedulers lie at the heart of all concurrency in Perl 6. While most users are unlikely to immediately encounter schedulers when starting to use Perl 6's concurrency features, many of them are implemented in terms of it. Thus, they will be described first here to avoid lots of forward references.

A scheduler is something that does the Scheduler role. Its responsibility is taking code objects representing tasks that need to be performed and making sure they get run, as well as handling any time-related operations (such as, "run this code every second").

The current default scheduler is available as $*SCHEDULER. If no such dynamic variable has been declared, then $PROCESS::SCHEDULER is used. This defaults to an instance of ThreadPoolScheduler, which maintains a pool of threads and distributes scheduled work amongst them. Since the scheduler is dynamically scoped, this means that test scheduler modules can be developed that poke a $*SCHEDULER into EXPORT, and then provide the test writer with control over time.

The cue method takes a Callable object and schedules it.

$*SCHEDULER.cue: { say "Golly, I got scheduled!" }

Various options may be supplied as named arguments. (All references to time are taken to be in seconds, which may be fractional.) You may schedule an event to fire off after some number of seconds:

$*SCHEDULER.cue: in=>10, { say "10s later" }

or at a given absolute time, specified as an Instant:

$*SCHEDULER.cue: at=>$instant, { say "at $instant" }

If a scheduled item dies, the scheduler will catch this exception and pass it to a handle_uncaught method, a default implementation of which is provided by the Scheduler role. This by default will report the exception and cause the entire application to terminate. However, it is possible to replace this:

$*SCHEDULER.uncaught_handler = sub ($exception) {
    $logger.log_error($exception);
}

For more fine-grained handling, it is possible to schedule code along with a code object to be invoked with the thrown exception if it dies:

$*SCHEDULER.cue:
    { upload_progress($stuff) },
    catch => -> $ex { warn "Could not upload latest progress" }

Use :every to schedule a task to run at a fixed interval, possibly with a delay before the first scheduling.

# Every second, from now
$*SCHEDULER.cue: :every(1), { say "Oh wow, a kangaroo!" };

# Every 0.5s, but don't start for 2s.
$*SCHEDULER.cue: { say "Kenya believe it?" }, :every(0.5), :in(2);

Since this will cause the given task to be executed at the given interval ad infinitum, there are two ways to make sure the scheduling of the task is halted at a future time. The first is provided by specifying the :times parameter in the .cue:

# Every second, from now, but only 42 times
$*SCHEDULER.cue: :every(1), :times(42), { say "Oh wow, a kangaroo!" };

The second is by specifying code that will be checked at the end of each interval. The task will be stopped as soon as it returns a True value. You can do this with the :stop parameter.

# Every second, from now, until stopped
my $stop;
$*SCHEDULER.cue: :every(1), :stop({$stop}), { say "Oh wow, a kangaroo!" };
sleep 10;
$stop = True;  # task stopped after 10 seconds

The .cue method returns a Cancellation object, which can also be used to stop a repeating cue:

my $c = $*SCHEDULER.cue: :every(1), { say "Oh wow, a kangaroo!" };
sleep 10;
$c.cancel;  # task stopped after 10 seconds

Schedulers also provide counts of the number of operations in various states:

say $*SCHEDULER.loads;

This returns, in order, the number of cues that are not yet runnable due to delays, the number of cues that are runnable but not yet assigned to a thread, and the number of cues that are now assigned to a thread (and presumably running). [Conjecture: perhaps these should be separate methods.]

Schedulers may optionally provide further introspection in order to support tools such as debuggers.

There is also a CurrentThreadScheduler, which always schedules things on the current thread. It provides the same methods, just no concurrency, and any exceptions are thrown immediately. This is mostly useful for forcing synchrony in places that default to asynchrony. (Note that .loads can never return anything but 0 for the currently running cues, since they're waiting on the current thread to stop scheduling first!)

Promises

A Promise is a synchronization primitive for an asynchronous piece of work that will produce a single result (thus keeping the promise) or fail in some way (thus breaking the promise).

The simplest way to use a Promise is to create one:

my $promise = Promise.new;

And then later keep it:

$promise.keep;      # True
$promise.keep(42);  # a specific return value for kept Promise

Or break it:

$promise.break;                             # False
$promise.break(X::Some::Problem.new);       # With exception
$promise.break("I just couldn't do it");    # With message

The current status of a Promise is available through the status method, which returns an element from the PromiseStatus enumeration.

enum PromiseStatus (:Planned(0), :Kept(1), :Broken(2));

The result itself can be obtained by calling result. If the Promise was already kept, the result is immediately returned. If the Promise was broken then the exception that it was broken with is thrown. If the Promise is not yet kept or broken, then the caller will block until this happens.

A Promise will boolify to whether the Promise is already kept or broken. There is also an cause method for extracting the exception from a Broken Promise rather than having it thrown.

if $promise {
    if $promise.status == Kept {
        say "Kept, result = " ~ $promise.result;
    }
    else {
        say "Broken because " ~ $promise.cause;
    }
}
else {
    say "Still working!";
}

You can also simply use a switch:

given $promise.status {
    when Planned { say "Still working!" }
    when Kept    { say "Kept, result = ", $promise.result }
    when Broken  { say "Broken because ", $promise.cause }
}

There are various convenient "factory" methods on Promise. The most common is start.

my $p = Promise.start(&do_hard_calculation);

This creates a Promise that runs the supplied code, and calls keep with its result. If the code throws an exception, then break is called with the Exception. Most of the time, however, the above is simply written as:

my $p = start {
    # code here
}

Which is implemented by calling Promise.start.

There is also a method to create a Promise that is kept after a number of seconds, or at a specific time:

my $kept_in_10s      = Promise.in(10);
my $kept_in_duration = Promise.in($duration);
my $kept_at_instant  = Promise.at($instant);

The result is always True and such a Promise can never be broken. It is mostly useful for combining with other promises.

There are also a couple of Promise combinators. The anyof combinator creates a Promise that is kept whenever any of the specified Promises are kept. If the first promise to produce a result is instead broken, then the resulting Promise is also broken. The cause is passed along. When the Promise is kept, it has a True result.

my $calc     = start { ... }
my $timeout  = Promise.in(10);
my $timecalc = Promise.anyof($calc, $timeout);

There is also an allof combinator, which creates a Promise that will be kept when all of the specified Promises are kept, or broken if any of them are broken.

[Conjecture: there should be infix operators for these resembling the junctionals.]

The then method on a Promise is used to request that a certain piece of code should be run, receiving the Promise as an argument, when the Promise is kept or broken. If the Promise is already kept or broken, the code is scheduled immediately. It is possible to call then more than once, and each time it returns a Promise representing the completion of both the original Promise as well as the code specified in then.

my $feedback_promise = $download_promise.then(-> $res {
    given $res.status {
        when Kept   { say "File $res.result().name() download" }
        when Broken { say "FAIL: $res.cause()"                 }
    }
});

[Conjecture: this needs better syntax to separate the "then" policies from the "else" policies (and from "catch" policies?), and to avoid a bunch of switch boilerplate. We already know the givens here...]

One risk when working with Promises is that another piece of code will sneak in and keep or break a Promise it should not. The notion of a promise is user-facing. To instead represent the promise from the viewpoint of the promiser, the various built-in Promise factory methods and combinators use Promise::Vow objects to represent that internal resolve to fulfill the promise. ("I have vowed to keep my promise to you.") The vow method on a Promise returns an object with keep and break methods. It can only be called once during a Promise object's lifetime. Since keep and break on the Promise itself just delegate to self.vow.keep(...) or self.vow.break(...), obtaining the vow before letting the Promise escape to the outside world is a way to take ownership of the right to keep or break it. For example, here is how the Promise.in factory is implemented:

method in(Promise:U: $seconds, :$scheduler = $*SCHEDULER) {
    my $p = Promise.new(:$scheduler);
    my $v = $p.vow;
    $scheduler.cue: { $v.keep(True) }, :in($seconds);
    $p;
}

The await function is used to wait for one or more Promises to produce a result.

my ($a, $b) = await $p1, $p2;

This simply calls result on each of the Promises, so any exception will be thrown.

Channels

A Channel is essentially a concurrent queue. One or more threads can put values into the Channel using send:

my $c = Channel.new;
$c.send($msg);

The call to .send does not block.

Meanwhile, other threads can .receive the values:

my $msg = $c.receive;

The .receive call does block. Alternatively, the .poll method takes a message from the channel if one is there or immediately returns Nil if nothing is there. Only one call to .receive or .poll returns per call of .send, and which listening thread receives each value is left up to the implementation's scheduler.

Channels are ideal for producer/consumer scenarios, and since there can be many senders and many receivers, they adapt well to scaling certain pipeline stages out over multiple workers also. [Conjectural: The two feed operators ==> and <== are implemented using Channel to connect each of the stages.]

A Channel may be "forever", but it is possible to close it to further sends by telling it to close:

$c.close();

Trying to send any further messages on a closed channel will throw the X::Channel::SendOnDone exception. Closing a channel has no effect on the receiving end until all sent values have been received. At that point, any further calls to receive will throw X::Channel::ReceiveOnDone. The closed method returns a Promise that is kept when a sender has called close and all sent messages have been received. Note that multiple calls to the same channel's closed method return the same promise, not a new one.

Like .close, calling .fail will close the channel. However this represents an abnormal termination, and so, the method must be provided an exception which should be thrown instead of X::Channel::ReceiveOnDone.

A Promise may be obtained from a Channel via the .closed method. This Promise will be kept when the Channel is closed or broken if it is failed.

A whenever clause (described below) on a Channel will fire for each received value, and may also mark the whenever as "done" using the same criteria as .closed, so it can also be used in order to write a reactor to receive from a channel until it is closed:

react {
    whenever $channel {
        "Got a $_".say;
    }
}
"Done".say;

A whenever clause on a Channel competes for values inside the scheduler alongside any .receive and .poll calls on that Channel from other threads. This also is true when the clause is part of a supply block (described below).

A Channel in list context will iterate all received values lazily, and stop iterating when the channel is closed:

for @$channel -> $val { ... }
for $channel.list -> $val { ... }

Note that this is not a combinator, but a means for transfering data from the reactive realm to the lazy realm. Some reasonable amount of buffering is assumed between the two. Just like all the above constructs, it competes for values in the scheduler against any other readers.

Supplies

Channels are good for producer/consumer scenarios, but because each worker blocks on receive, it is not such an ideal construct for doing fine-grained processing of asynchronously produced streams of values. Additionally, there can only be one receiver for each value. Supplies exist to address both of these issues.

A Supply pushes or pumps values to one or more receivers who have registered their interest, by (often indirecty) calling the .tap method on the Supply. This returns a Tap object unique to the receiver, which one may then .close to tell the Supply that the receiver is no longer interested. Note that this does not happen automatically if the Tap is thrown away, as the Supply also keeps a reference to the Tap. The .tap method takes up to three callables as arguments, one block and two optional named arguments:

$supply.tap: -> $value { say "Got a $value" },
    done => { say "Reached the end" },
    quit => {
        when X::FooBar { die "Major oopsie" };
        default        { warn "Supply shut down early: $_" }
    }

The first unnamed closure is known as the emit closure. It is invoked whenever a value is emitted into the thing that has been tapped. The optional named parameter done specifies the code to be invoked when all expected values have been produced and no more values will be emitted. The optional named parameter quit specifies the code to be invoked if there is an error. This also means there will be no further values.

For the same reasons that we have Vows for Promises, supplies are split into a Supply role and a Supplier role. The Supplier role has a .Supply method used to create a live Supply, and emit, done, and quit methods which send the corresponding events to any Supply objects so created. A Supplier is not tapped directly. The Supply role does not have emit, done, or quit methods, and may be tapped. This allows the provider of a Supply, if they so desire, to be sure that one user cannot inject data to other users of a Supply. Multiple supplies may be created from a Supplier, and all taps of all such supplies receive events sent using said Supplier.

my $e = Supplier.new;
my $y = $e.Supply;

my $t1 = $y.tap({ say $_ });
$e.emit(1);                              # 1\n
$e.emit(2);                              # 2\n

my $t2 = $y.tap({ say 2 * $_ },
          :done({ say "End"  }));
$e.emit(3);                              # 3\n6\n

$t1.close;
$e.emit(4);                              # 8\n
$e.done;                                 # End\n

The above example demonstrates a simple live supply -- note that $t2 missed events 1 and 2 because it was not listening at the time they happened. From an outside perspective, a live supply will behave as if it were alive and happily producing values even when it never had taps or all its taps have been closed:

my $r = Supplier.new;
my $z = $r.Supply;
$r.emit("A tree fell in the woods");     #

A more common type of supply is an on-demand supply, which creates multiple independent streams of events, starting a new stream each time it is tapped. Each Tap receives only the events emitted on its private stream. For example, the factory method Supply.interval produces a fresh timer with the appropriate interval each time it is tapped, each of which will stop emitting values when its corresponding Tap is closed:

my $i = Supply.interval(2);
my $tick = $i.tap({ say " tick $_" });
sleep 3;                                 # tick 0\n tick 1\n
my $tock = $i.tap({ say " tock $_" });
sleep 3.5;                               # tock 0\n tick 2\n tock 1\n tick 3\n
$tock.close;
sleep 3;                                 # tick 4\n
$tick.close;

By default most Supply objects are serial, meaning they will not emit simultaneous events (be they emit, done, or quit events.) Other supplies are entirely asyncronous: it is possible for values to be pumped from an any number of asynchronous workers running on different threads. To tell the difference between the two, the .serial method may be used. Note that calls to a Supplier's .emit, .done, or .quit methods always block (and may deadlock in the presence of feedback) until all taps have been run -- the difference between a serial and a non-serial Supply is when multiple threads call .emit simultaneously: on a non-serial Supply, multiple taps may run in parallel, one on each thread, each processing values emitted by that thread. On a serial Supply, only one tap may run at the same time, and the threads calling .emit may block on the taps processing each other's values, not just their own. Because of the semantic of "not emit simultaneous events" it is also guaranteed, on a serial Supply, that all taps corresponding to one thread emission will complete before any of the taps from another thread's emission are run.

The Supply class has various methods that produce more interesting kinds of Supply. These mostly produce serial supplies. Even the Supplier.Supply method produces a serial supply -- to get a fully asyncronous live supply one must use the Supplier.unsanitized-supply method. Most factory methods handle creation/retention/destruction of a Supplier internally, or combine prexisting Supply objects, so their users need only to concern themselves with Supply objects.

The following methods are class methods which create a new Supply without requiring an existing Supply. The methods merge, zip and zip-latest also have class method forms, documented further below with their instance method counterparts.

from-list
my $fl = Supply.from-list(^10);

Takes a (potentially lazy) list of values, and returns an on-demand Supply that, when tapped, will iterate over the values and invoke the Tap's emit callable for each of them, and any done callable at the end. If the iteration at some point produces an exception, then the Tap's quit callable will be invoked to pass along the exception.

interval
my $e1    = Supply.interval(1);     # Once a second, starting now
my $e5i10 = Supply.interval(5, 10); # Each 5 seconds, starting in 10 seconds

Produces an on-demand Supply that, when tapped, will produce successive integer values at a regular time interval, with the integer values counting up from 0.

Take the returned tap object and close it to stop the ticks:

my $e1 = Supply.interval(1).tap(&say);
# ...later...
$e1.close();

Supplies are mathematically dual to iterators, and so it is possible to define the same set of operations on them as are available on lazy lists. The key difference is that, while grep on a lazy list pulls a value to process, working synchronously, grep on a Supply has values pushed through it, and pushes those that match the filter onwards to anything that taps it.

The following methods are available on an instantiated Supply ($s in these examples). Note that calling most of these methods on an on-demand Supply does not constitute demanding a fresh set of values from the Supply -- a tap is not immediately performed. Instead, tapping the resulting Supply will do so, per Tap. In other words, connected networks of Supplies do not sit around in the background pointlessly feeding each other values internally, they wait until something is listening. (This behavior is generally accomplished via whenever clauses, described below.)

list
my @l := $s.list;

Produces a lazy List with the values of the Supply.

wait
$s.wait;

Waits until the specified Supply is done or quit. In the latter case, throws the exception with which the Supply was quit.

Channel
my $c = $s.Channel;

Produces a Channel of the values of the given Supply.

Promise
my $p = $s.Promise;

Produces a Promise that will be kept when the Supply is done or broken if it is quit.

last
my $l = $s.last(42);  # default: 1

Produces a Supply that will only emit the last N values of the given Supply (after that Supply is done). Default is the final value.

grab
my $g = $s.grab( { .sort } ); # sort the values of a Supply

Produces a Supply which will cache all values emitted by the given Supply until it is done. It will then call the given closure and then .emit each of the return values of the closure, and then call .done on itself.

flat
my $f = $s.flat;

Produces a Supply in which all values of the original supply are flattened then individually emitted.

do
my $seen;
my $d = $s.do( {$seen++} );

Produces a Supply that is identical to the original supply, but will execute the given code for its side-effects, once for each emitted value, before running any taps. Only one thread will ever be executing the side-effect code at a time; others will block behind it. In addition to serializing the side-effect, the resulting supply is also serial, even if the one it was created from is not. The side-effect code is only run for emitted values, not when the original is quit or done.

act
my $seen;
$s.act( {$seen++} );

This is a special case of Supply.do, that will also create a tap on the given Supply, so that you only need to worry about writing the side-effect code. Returns the Tap.

grep
my $g = $s.grep( * > 5 );
my $g = $s.grep(Int);

Produces a Supply that only provides values that you want. Takes either a Callable (which is supposed to return a True value to pass on emitted values) or a value to be smartmatched against.

map

Analogous to List's .map method, but produces a Supply.

my $m = $s.map( * * 5 );

Produces a Supply that provides its original's Supply values multiplied by 5.

my $m2 = $s.map( { $_ xx 2 } );

Produces a Supply that provides its original's Supply values twice.

unique
my $u = $s.unique( :as( {$_} ), :with( &[===] ), :expires(1) );

Produces a Supply that only provides unique values, as defined by the optional as and with named parameters (same as List.unique). The optional expires parameter specifies how long to wait (in seconds) before "resetting" and not considering a value to have been seen, even if it's the same as an old value.

squish
my $q = $s.squish( :as( {$_} ), :with( &[===] ), :expires(1) );

Produces a Supply that only provides sequentially different values, as defined by the optional as and with named parameters (same as List.squish). The optional expires parameter specifies how long to wait (in seconds) before "resetting" and not squishing a new value with an old one, even if they are the same.

max
my $a = $s.max(&by); # default &infix:<cmp>

Produces a Supply that produces the maximum values of the specified Supply. In other words, from a continuously ascending Supply it will produce all the values. From a continuously descending Supply it will only produce the first value. The optional parameter specifies the comparator, just as with Any.max.

min
my $i = $s.min(&by); # default &infix:<cmp>

Produces a Supply that produces the minimum values of the specified Supply. In other words, from a continuously descending Supply it will produce all the values. From a continuously ascending Supply it will only produce the first value. The optional parameter specifies the comparator, just as with Any.min.

minmax
my $m = $s.minmax(&by); # default &infix:<cmp>

Produces a Supply that, for each value emitted, produces Ranges with the minimum and maximum values seen thus far on the specified Supply. The optional parameter specifies the comparator, just as with Any.minmax.

batch
my $b = $s.batch( :elems(100), :seconds(1) );

Produces a Supply that batches the values of the given Supply by either the number of elements (using the elems named parameter) or the maximum number of seconds (using the seconds named parameter) or both. Values are grouped in a single array element when flushed.

throttle
my $t = $s.throttle( $elems, $seconds );

Produces a Supply that throttles emitting the values of the given Supply on the created Supply by the number of elements (specified by the first parameter) per number of seconds (specified by the second parameter).

elems
my $e = $s.elems($seconds?); # default: see all

Produces a Supply that, for each value emitted, produces the number of elements seen thus far in the given Supply. You can also specify an interval to only see the number of elements seen once per that interval.

rotor
my $b = $s.rotor(@cycle);

Produces a "rotoring" Supply with the same semantics as List.rotor.

delayed
my $d = $s.delayed( 3.5 );  # delay supply 3.5 seconds

Produces a Supply that passes on the values of the given Supply with the given delay (in seconds).

stable
my $u = $s.stable( $seconds, :$scheduler );

Produces a Supply that only passes on a value if it wasn't superseded by another value in the given time (in seconds). Optionally uses another scheduler than the default scheduler, using the scheduler named parameter.

start
my $t = $s.start( {...} );

Takes a closure and, for each supplied value, schedules the closure to run on another thread. It then emits a Supply (resulting in us having a supply of supplies) that will either have a single value emitted and then be done if the async work completes successfully, or quit if the work fails. Useful for kicking off work on the thread pool if you do not want to block up the thread pushing values at you (maybe 'cus you are reacting to UI events, but have some long-running work to kick off). Usually used in combination with migrate.

migrate
my $m = $t.migrate;

Produces a continuous Supply from a Supply, in which each value is a Supply which emits a value from the original Supply unchanged. As soon as a new Supply appears, it will close the previously emitted Supply and send the next value from the original Supply to the most recently emitted Supply instead. Can be used in combination with schedule-on.

schedule-on
my $o = $m.schedule-on( $scheduler );

This allows a Supply's emit/done/quit to be scheduled on another scheduler. Useful in GUI situations, for example, where the final stage of some work needs to be done on some UI scheduler in order to have UI updates run on the UI thread.

reduce
my $r = $s.reduce( {...} );

Produces a Supply that will emit each reduction from the given Supply, much like the triangular reduction metaoperator ([\...]) on Lists.

lines
my $l = $s.lines;              # chomp lines
my $l = $s.lines( :!chomp );   # do *not* chomp lines

Produces a Supply that will emit the characters coming in line by line from a Supply that's usually created by some asynchronous I/O operation. The optional :chomp named parameter indicates whether to remove line separators: the default is True.

words
my $w = $s.words;

Produces a Supply that will emit the characters coming in word by word from a Supply that's usually created by some asynchronous I/O operation.

classify
my $c = $s.classify( {.WHAT} );  # one Supply per type of value
my $h = $s.classify( %mapper );
my $a = $s.classify( @mapper );

Produces a Supply which emits Pairs consisting of a classification value as .key, and a Supply as the .value. That supply will then emit any values from the original Supply which match the classifier. Parameters behave similar to List.classify, but does not support multi-level classification.

categorize
my $c = $s.categorize( {@categories} );
my $h = $s.categorize( %mapper );
my $a = $s.categorize( @mapper );

Produces a Supply in which emits Pairs consisting of a classification value as .key, and a Supply as the .value. That supply will then emit any values from the original Supply which match the classification value. Unlike .classify, more than one Pair may be emitted per value emitted from the original Supply. See List.categorize for a description of the behavior of the parameters.

reverse
my $r = $s.reverse;

Produces a Supply that emits the values of the given Supply in reverse order. Please note that this Supply will only start delivering values when the given Supply is .done, much like .grab.

sort
my $o = $s.sort(&by);  # default &infix:<cmp>

Produces a Supply that emits the values of the given Supply in sorted order. Please note that this Supply will only start delivering values when the given Supply is .done. Optionally accepts a comparator Block.

There are some combinators that deal with bringing multiple supplies together:

merge
my $m = $s1.merge($s2);

my $m = Supply.merge(@s);  # also as class method

Produces a Supply containing the values produced by given and the specified supply or supplies, and triggering done once all of the supplies have done so. The resulting supply is serial, even if any or all of the merged supplies are not.

zip
my $z = $s1.zip($s2);                   # defaults to :with( &[,] )

my $z = Supply.zip(@s, :with( &[,] ));  # also as class method

Produces a Supply that pairs together items from the given and the specified supply or supplies, using infix:<,> by default or any other user-supplied function with the with named parameter. The resulting supply is serial, even if any or all of the zipped supplies are not.

zip-latest
my $z = $s1.zip-latest($s2);                  # like zip, defaults to :with( &[,] )

my $z = Supply.zip-latest(@s, :with( &[,] )); # also a method on Supply.

my $z = Supply.zip-latest( @s, :initial(42,63) ); # initial state

Produces a Supply that will emit tuples of values as soon as any combined Supply produces a value. Before any tuples are emitted, all supplies have to have produced at least one value. By default, it uses infix:<,> to produce the tuples, but the named parameter with can override that. The resulting supply is serial, even if any or all of the merged supplies are not.

The named parameter initial can optionally be used to indicate the initial state of the values to be emitted.

[TODO: plenty more of these: while, until...]

The above combinators which involve multiple source supplies need care in their implementation, since values may arrive at any point from each source, and possibly at the same time. To help write such combinators, the supply block and whenever clause are used. These, along with the react block, are available for general use as well.

A supply block is a convenient way to create an on-demand Supply. It is just a declaration and does not run until the Supply is tapped (which may happen multiple times resulting in multiple runs/clones of the block.) Within a supply block, emit can be used to emit values to the tapper, and done can be used to convey that there will be no more values.

my $s = supply {
    emit 'a value!';
    emit 'another!';
    done;
}

The emit and done can be in nested scopes, and follow the same rules as gather and take, except they look for an enclosing supply or react rather than a gather. There is no corresponding quit statement: instead, any unhandled exception thrown inside of a supply block will be passed onwards using .quit.

Likewise, a whenever clause also looks around itself lexotically for an enclosing supply or react. A whenever clause takes a predicate which may be a Supply, Channel, Promise or Iterable, followed by a consequent block. These clauses guarantee that only one thread will enter their consequent or that of any other whenever clause inside the same enclosing supply or react block at the same time -- blocking any .emit call on a Supply or leaving any additional values in a Channel's .send queue in the scheduler.

my $s = Supply.interval(1);
my $c = Channel.new;
my $l := (while $++ < 4 { NEXT { sleep 1 }; Int(now - BEGIN now) });
my $p = Promise.in(3);
my %contended_hash;
start {
    for 0..4 { sleep 1; $c.send($_); LAST { $c.close } }
}
react {
    # This is safe, because the whenevers protect %contended_hash
    whenever $s { %contended_hash{$_}++; done if $_ > 4 };
    whenever $c { %contended_hash{$_}++ };
    whenever $l { %contended_hash{$_}++ };
    whenever $p { %contended_hash{$_}++ };
}
%contended_hash.say; # 0 => 3, 1 => 3, 2 => 3, 3 => 3, 4 => 2, 5 => 1, True => 1

A whenever with a Supply as a predicate will tap that predicate when the enclosing block is run (assuming control reaches the whenever clause.) It will then offer its consequent block as the emit closure for the resulting Tap. However, it is sometimes best just to think of a whenever as a loop which runs once for each value produced by the predicate Supply, using the produced values as the topic. Following this model, a QUIT or LAST phaser inside the consequent block may be used to provide a quit or done closure (respectively) to the Tap. The resulting Tap itself is ensconced internally, but may also be accessed as the return value of the entire whenever clause. This is not necessary just in order for a whenever to close its own tap: last may be called from within the consequent to do so. Labels may be used as normal to close the taps of outer clauses. Closing the last active tap closes the generated Supply.

A whenever block with a Channel as a predicate behaves analogously, running the closure once for every value sent to it by the scheduler, as if it were constantly calling .receive on the Channel. A Promise predicate runs the block when the Promise is kept or broken as if it were calling await. An Iterable predicate will start a new iterator when the whenever clause is run, entering the consequent closure for each value pulled from the Iterator. In all such cases, a QUIT block inside the whenever will catch exceptions generated from within the predicate. Handling these exceptions will prevent .quit from being called on the Supply created by the block.

Note that it is possible for control never to reach a whenever clause when a supply or react block is run. In this case, the whenever clause has no effect. If control reaches a whenever clause more than once (for example, if it is inside a loop) multiple taps on the predicate (which may be different each time) are created. A whenever clause may also be executed later, after a supply block has produced a Supply which has then been tapped, or while waiting for a react block to fall off the bottom. This could happen from inside another whenever clause in response to an event, or from an externally added Tap. This is legal, and dynamically adds a new tap (of the predicate) to the Supply which caused this to happen. That is to say, once active, a Supply may indeed use whenever clauses to self-modify, as long as it was originally syntactically anchored to a supply or react block:

my $s1 = Supply.interval(1).share;
my $s2 = supply {
    whenever $s1 {
        emit($_);
        done if $_ > 4;
    }
};
$s2.tap({
    if 2 < $_ < 5 {
        "MORE $_".say;
         whenever $s1 {
             "OHAI $_".say;
         }
    }
});
sleep 10;
# MORE 3
# MORE 4
# OHAI 4
# OHAI 5
# OHAI 5

Destructuring subsignatures may be used with whenever clause topics via pointy block syntax.

A supply block containing whenever clauses will call done on itself when all whenever clause predicates (which have executed) have had .done/quit, .close/.fail or .keep/.break called on them, appropriately, or in the case of an Iterable predicate, when the end has been reached. These conditions are equivalent to manually closing the clause with last.

A react block will not wait to be tapped -- it will immediately run and then wait for a done or quit condition, after which point, control will resume below the react block.

System events exposed as Supplies

System events, such as signals, or mouse events, can be exposed as Supplies. Because of lack of portability, these will most likely be implemented as third-party modules.

Basic signal support is offered by the signal function, which takes one or more Signal enums, and an optional scheduler named parameter. It produces a Supply which, when tapped, will emit any signal coming in. For example:

signal(SIGINT).tap( { say "Thank you for your attention"; exit 0 } );

would catch Control-C, thank you, and then exit. Of course, you don't need to exit immediately. Here's an example of how you would make sure that an iteration in a loop is completed before exiting:

for @todo {
    state $quitting;
    state $tap = signal(SIGINT).tap( { $quitting = True } );
    LAST  $tap.close;
    LEAVE exit(0) if $quitting;
    ... # code to protect
}

This probably could use some syntactic sugar.

The list of supported Signals can be found by checking Signal::.keys, as you would any enum.

I/O features exposed as Supplies

Various I/O-related things are also exposed as supplies. For example, it is possible to get notifications on changes to files or files (directly) in a directory, using:

IO::Notification.watch-path(".").tap(-> $file {
    say "$file changed";
});

This is quite a mouthful, so there is a shortcut available with the IO coercer and the watch method:

".".IO.watch.tap: -> $file { say "$file changed" };

Note that since I/O callbacks are, by default, scheduled on the thread pool, then it's possible that your callback will be executing twice on the same thread. One way to cope is with act:

".".IO.watch.act(-> $file {
    state %changes;
    say "$file changed (change {++%changes{$file}})";
});

It can also take done and quit named parameters; these go to the tap, while the emit closure is put in a do. A Tap is returned, which may be closed in the usual way. (Note that the name act is also a subtle reference to actor semantics.)

Inter-Process Communication exposed as Promises and Supplies

Starting external processes is rather easy: shell(), run() and qx//. Having external processes run asynchronously, is slightly more involved. But not much. The workhorse of asynchronous IPC in Perl 6 is Proc::Async:

my $proc = Proc::Async.new( $path, @args );

If you like to send data to the process, you need to open it with the :w named parameter.

my $proc = Proc::Async.new( $path, @args, :w );

By default, the current environment (as available in %*ENV) will be set for the external process. You can override this with the :ENV named parameter:

my $proc = Proc::Async.new( $path, @args, :ENV(%hash) );

The returned object can then be called whenever needed to start the external process. However, before you do that, one needs to be clear what to do about the output of the external process. Getting information back from the external process's STDOUT or STDERR, is done by a Supply that either gets characters or bytes.

$proc.stdout.act(&say);   # simply pass it on to our $*OUT as chars
$proc.stderr.act(&note);  # and $*ERR as chars, but could be any code

or:

$proc.stdout(:bin).act: { # process STDOUT bytes };
$proc.stderr(:bin).act: { # process STDERR bytes };

So, to make sure no information will be lost, you need to create and tap the supplies before the process is started.

To start the external process, you need to call the .start method. It returns a Promise that becomes Kept (and True) if the process concludes successfully, or Broken (and False) if the process failed for some reason.

my $done = $proc.start( :$scheduler = $*SCHEDULER );

To send data to the running process, you can use the .print, .say and .write methods on the Proc::Async object:

my $printed = $proc.print( "Hello world\n", :$scheduler = $*SCHEDULER );
my $said    = $proc.say(   "Hello world",   :$scheduler = $*SCHEDULER );
my $written = $proc.write( $buffer,         :$scheduler = $*SCHEDULER );

They all also return a Promise that is Kept when communication with the process was successful.

Some programs expect their STDIN to be closed to signify the end of their processing. This can be achieved with the .close-stdin method:

$proc.close-stdin;

Finally, if your process as going awry, you can stop it with the .kill method:

$proc.kill;            # sends HUP signal to process
$proc.kill("SIGINT");  # send INT signal
$proc.kill(1);         # if you just know the signal number on your system

The parameter should be something that is acceptable to the Kernel.signal method.

The Event Loop

There is no event loop. Previous versions of this synopsis mentioned an event loop that would be underlying all concurrency. In this version, this is not the case.

Threads

VM-level threads, which typically correspond to OS-level threads, are exposed through the Thread class. Whatever underlies it, a Thread should always be backed by something that is capable of being scheduled on a CPU core (that is, it may not be a "green thread" or similar). Most users will not need to work with Threads directly. However, those building their own schedulers may well need to do so, and there may be other exceptional circumstances that demand such low-level control.

The easiest way to start a thread is with the start method, which takes a Callable and runs it on a new thread:

my $thread = Thread.start({
    say "Gosh, I'm in a thread!";
});

It is also possible to create a thread object, and set it running later:

my $thread = Thread.new(code => {
    say "A thread, you say?";
});
# later...
$thread.run();

Both approaches result in $thread containing a Thread object. At some point, finish should be called on the thread, from the thread that started it. This blocks until the thread has completed.

say "Certainly before the thread is started";
my $thread = Thread.start({ say "In the thread" });
say "This could come before or after the thread's output";
$thread.finish();
say "Certainly after all the above output";

As an alternative to finish, it is possible to create a thread whose lifetime is bounded by that of the overall application. Such threads are automatically terminated when the application exits. In a scenario where the initial thread creates an application lifetime thread and no others, then the exit of the initial thread will cause termination of the overall program. Such a thread is created by either:

my $thread = Thread.new(:code({ ... }), :app_lifetime);

Or just, by using the start method:

my $thread = Thread.start({ ... }, :app_lifetime);

The property can be introspected:

say $thread.app_lifetime; # True/False

Each thread also has a unique ID, which can be obtained by the id property.

say $thread.id;

This should be treated as an opaque number. It can not be assumed to map to any particular operating system's idea of thread ID, for example. For that, use something that lets you get at OS-level identifiers (such as calling an OS API using NativeCall).

A thread may also be given a name.

my $thread = Thread.start({ ... }, :name<Background CPU Eater>);

This can be useful for understanding its usage. Uniqueness is not enforced; indeed, the default is "<anon>".

A thread stringifies to something of the form:

Thread<id>(name)

For example:

Thread<1234>(<anon>)

The currently executing thread is available through $*THREAD. This is even available in the initial thread of the program, in this case by falling back to $PROCESS::THREAD, which is the initial thread of the process.

Finally, the yield method can be called on Thread (not on any particular thread) to hint to the OS that the thread has nothing useful to do for the moment, and so another thread should run instead.

Atomic Compare and Swap

The Atomic Compare and Swap (CAS) primitive is directly supported by most modern hardware. It has been shown that it can be used to build a whole range of concurrency control mechanisms (such as mutexes and semaphores). It can also be used to implement lock-free data structures. It is decidedly a primitive, and not truly composable due to risk of livelock. However, since so much can be built out of it, Perl 6 provides it directly.

A Perl 6 implementation of CAS would look something like this:

sub cas($ref is rw, $expected, $new) {
    my $seen = $ref;
    if $ref === $expected {
        $ref = $new;
    }
    return $seen;
}

Except that it happens atomically. For example, a crappy non-reentrant mutex could be implemented as:

class CrappyMutex {
    has $!locked = 0;

    method lock() {
        loop {
            return if cas($!locked, 0, 1) == 0;
        }
    }

    method unlock() {
        $!locked = 0;
    }
}

Another common use of CAS is in providing lock-free data structures. Any data structure can be made lock-free as long as you're willing to never mutate it, but build a fresh one each time. To support this, there is another &cas candidate that takes a scalar and a block. It calls the block with the seen initial value. The block returns the new, updated value. If nothing else updated the value in the meantime, the reference will be updated. If the CAS fails because another update got in first, the block will be run again, passing in the latest value.

So, atomically incrementing a variable is done thusly:

cas $a, { $_.succ };    # $a++

or more generally for all assignment meta-operators:

cas $a, { $_ * 5 };     # $a *= 5

Another example, implementing a top-5 news headlines list to be accessed and updated without ever locking, as:

class TopHeadlines {
    has $!headlines = [];   # Scalar holding array, as CAS needs

    method headlines() {
        $!headlines
    }

    method add_headline($headline) {
        cas($!headlines, -> @current {
            my @new = $headline, @current;
            @new.pop while @new.elems > 5;
            @new
        });
    }
}

It's the programmer's duty to ensure that the original data structure is never mutated and that the block has no side-effects (since it may be run any number of times).

Low-level primitives

Perl 6 offers high-level concurrency methods, but in extreme cases, like if you need to implement a fundamentally different mechanism, these primitives are available.

Locks

Locks are unpleasant to work with, and users are pushed towards higher level synchronization primitives. However, those need to be implemented via lower level constructs for efficiency. As such, a simple lock mechanism - as close to what the execution environment offers as possible - is provided by the Lock class. Note that it is erroneous to rely on the exact representation of an instance of this type (for example, don't assume it can be mixed into). Put another way, treat Lock like a native type.

A Lock is instantiated with new:

$!lock = Lock.new;

The best way to use it is:

$!lock.protect: {
    # code to run with the lock held
}

This acquires the lock, runs the code passed, and then releases the lock. It ensures the lock will be released even if an exception is thrown. It is also possible to do:

{
    $!lock.lock();
    # do stuff
    LEAVE $!lock.unlock()
}

When using the lock and unlock methods, the programmer must ensure that the lock is unlocked. Lock is reentrant. Naturally, it's easy to introduce deadlocks. Again, this is a last resort, intended for those who are building first resorts.

Semaphore

The Semaphore class implements traditional semaphores that can be initiated with a fixed number of permits and offers the operations acquire to block on a positive number of permits to become available and then reduce that number by one, tryacquire to try to acquire a permit, but return False instead of blocking if there are no permits available yet. The last operation is release, which will increase the number of permits by one.

The initial number of permits may be negative, positive or 0.

Some implementations allow for race-free acquisition and release of multiple permits at once, but this primitive does not offer that capability.

AUTHORS

Jonathan Worthington <jnthn@jnthn.net>
Elizabeth Mattijsen <liz@wenzperl.nl>