Class: Concurrent::Promise

Promises are inspired by the JavaScript Promises/A and Promises/A+ specifications.

A promise represents the eventual value returned from the single completion of an operation.

Promises are similar to futures and share many of the same behaviours. Promises are far more robust, however. Promises can be chained in a tree structure where each promise may have zero or more children. Promises are chained using the then method. The result of a call to then is always another promise. Promises are resolved asynchronously (with respect to the main thread) but in a strict order: parents are guaranteed to be resolved before their children, children before their younger siblings. The then method takes two parameters: an optional block to be executed upon parent resolution and an optional callable to be executed upon parent failure. The result of each promise is passed to each of its children upon resolution. When a promise is rejected all its children will be summarily rejected and will receive the reason.

Promises have several possible states: :unscheduled, :pending, :processing, :rejected, or :fulfilled. These are also aggregated as #incomplete? and #complete?. When a Promise is created it is set to :unscheduled. Once the #execute method is called the state becomes :pending. Once a job is pulled from the thread pool's queue and is given to a thread for processing (often immediately upon #post) the state becomes :processing. The future will remain in this state until processing is complete. A future that is in the :unscheduled, :pending, or :processing is considered #incomplete?. A #complete? Promise is either :rejected, indicating that an exception was thrown during processing, or :fulfilled, indicating success. If a Promise is :fulfilled its #value will be updated to reflect the result of the operation. If :rejected the reason will be updated with a reference to the thrown exception. The predicate methods #unscheduled?, #pending?, #rejected?, and #fulfilled? can be called at any time to obtain the state of the Promise, as can the #state method, which returns a symbol.

Retrieving the value of a promise is done through the value (alias: deref) method. Obtaining the value of a promise is a potentially blocking operation. When a promise is rejected a call to value will return nil immediately. When a promise is fulfilled a call to value will immediately return the current value. When a promise is pending a call to value will block until the promise is either rejected or fulfilled. A timeout value can be passed to value to limit how long the call will block. If nil the call will block indefinitely. If 0 the call will not block. Any other integer or float value will indicate the maximum number of seconds to block.

Promises run on the global thread pool.

Copy Options

Object references in Ruby are mutable. This can lead to serious problems when the Concern::Obligation#value of an object is a mutable reference. Which is always the case unless the value is a Fixnum, Symbol, or similar "primitive" data type. Each instance can be configured with a few options that can help protect the program from potentially dangerous operations. Each of these options can be optionally set when the object instance is created:

• :dup_on_deref When true the object will call the #dup method on the value object every time the #value method is called (default: false)
• :freeze_on_deref When true the object will call the #freeze method on the value object every time the #value method is called (default: false)
• :copy_on_deref When given a Proc object the Proc will be run every time the #value method is called. The Proc will be given the current value as its only argument and the result returned by the block will be the return value of the #value call. When nil this option will be ignored (default: nil)

When multiple deref options are set the order of operations is strictly defined. The order of deref operations is:

• :copy_on_deref
• :dup_on_deref
• :freeze_on_deref

Because of this ordering there is no need to #freeze an object created by a provided :copy_on_deref block. Simply set :freeze_on_deref to true. Setting both :dup_on_deref to true and :freeze_on_deref to true is as close to the behavior of a "pure" functional language (like Erlang, Clojure, or Haskell) as we are likely to get in Ruby.

Examples

Start by requiring promises

require 'concurrent/promise' 

Then create one

p = Concurrent::Promise.execute do # do something  42 end 

Promises can be chained using the then method. The then method accepts a block and an executor, to be executed on fulfillment, and a callable argument to be executed on rejection. The result of the each promise is passed as the block argument to chained promises.

p = Concurrent::Promise.new{10}.then{|x| x * 2}.then{|result| result - 10 }.execute 

And so on, and so on, and so on...

p = Concurrent::Promise.fulfill(20). then{|result| result - 10 }. then{|result| result * 3 }. then(executor: different_executor){|result| result % 5 }.execute 

The initial state of a newly created Promise depends on the state of its parent:

• if parent is unscheduled the child will be unscheduled
• if parent is pending the child will be pending
• if parent is fulfilled the child will be pending
• if parent is rejected the child will be pending (but will ultimately be rejected)

Promises are executed asynchronously from the main thread. By the time a child Promise finishes initialization it may be in a different state than its parent (by the time a child is created its parent may have completed execution and changed state). Despite being asynchronous, however, the order of execution of Promise objects in a chain (or tree) is strictly defined.

There are multiple ways to create and execute a new Promise. Both ways provide identical behavior:

# create, operate, then execute p1 = Concurrent::Promise.new{ "Hello World!" } p1.state #=> :unscheduled p1.execute # create and immediately execute p2 = Concurrent::Promise.new{ "Hello World!" }.execute # execute during creation p3 = Concurrent::Promise.execute{ "Hello World!" } 

Once the execute method is called a Promise becomes pending:

p = Concurrent::Promise.execute{ "Hello, world!" } p.state #=> :pending p.pending? #=> true 

Wait a little bit, and the promise will resolve and provide a value:

p = Concurrent::Promise.execute{ "Hello, world!" } sleep(0.1) p.state #=> :fulfilled p.fulfilled? #=> true p.value #=> "Hello, world!" 

If an exception occurs, the promise will be rejected and will provide a reason for the rejection:

p = Concurrent::Promise.execute{ raise StandardError.new("Here comes the Boom!") } sleep(0.1) p.state #=> :rejected p.rejected? #=> true p.reason #=> "#<StandardError: Here comes the Boom!>" 

Rejection

When a promise is rejected all its children will be rejected and will receive the rejection reason as the rejection callable parameter:

p = Concurrent::Promise.execute { Thread.pass; raise StandardError } c1 = p.then(-> reason { 42 }) c2 = p.then(-> reason { raise 'Boom!' }) c1.wait.state #=> :fulfilled c1.value #=> 42 c2.wait.state #=> :rejected c2.reason #=> #<RuntimeError: Boom!> 

Once a promise is rejected it will continue to accept children that will receive immediately rejection (they will be executed asynchronously).

Aliases

The then method is the most generic alias: it accepts a block to be executed upon parent fulfillment and a callable to be executed upon parent rejection. At least one of them should be passed. The default block is { |result| result } that fulfills the child with the parent value. The default callable is { |reason| raise reason } that rejects the child with the parent reason.

• on_success { |result| ... } is the same as then {|result| ... }
• rescue { |reason| ... } is the same as then(Proc.new { |reason| ... } )
• rescue is aliased by catch and on_error