Rietveld Code Review Tool
Help | Bug tracker | Discussion group | Source code

Unified Diff: lib/task.js

Issue 9084006: task.js: cooperative multitasking for long-running tasks (Closed)
Patch Set: Created Dec. 26, 2012, 5:06 p.m.
Use n/p to move between diff chunks; N/P to move between comments.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « lib/counter_task.js ('k') | no next file » | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: lib/task.js
===================================================================
new file mode 100644
--- /dev/null
+++ b/lib/task.js
@@ -0,0 +1,227 @@
+/**
+ * The default value for runaway_limit, used only in the constructor.
+ * @type {Number}
+ */
+const default_runaway_limit = 1000;
+
+/**
+ * Class for long-running tasks. Such tasks divide work into discrete units of effort, which allows them to be
+ * interrupted and to post progress.
+ * <p/>
+ * Note that are no callback or events in this class. Any state feedback should be implemented as needed within the
+ * task generator. task generator can distinguish cancellation by examining the rvalue of its yield statements. The
+ * generator cannot detect being paused, but the control object that started the task can.
+ * <p/>
+ * PLANNED: There's no 'pause()' in this version. This is very similar to 'cancel()' except that it's restartable. The
+ * code isn't particularly difficult, but it would require reworking the main 'run()' loop.
+ * <p/>
+ * PLANNED: The current version will handle AJAX calls by polling the task generator every millisecond to see if there
+ * have been pending calls that have completed since the last poll. It would make for better performance for the task
+ * to signal that it should pause execution and then continue only after a pending call completes. This would eliminate
+ * run-time polling overhead.
+ * <p/>
+ * There would also need to be an interface presented to the task for pause and resume. Resuming is the harder one,
+ * because the task needs an object to call when 'Long_Task.prototype.run' is not in the call stack. Currently, the
+ * generator is instantiated before the constructor for this class, which means either passing such an object with
+ * 'send()' if that constructor is to be called. The better way is likely for this class to instantiate the generator
+ * with a pause/resume object as argument, but this only allows parametric generators (pretty much a requirement) if
+ * if 'Function.protoype.bind()' works on generator-functions (untried) or if an equivalent can be hacked up.
+ *
+ * @param {Generator} task_generator
+ * The task generator is the task to be run, implemented as a generator. Each call to the generator performs an
+ * increment of computation, whose size is determined by the task. This is part of a cooperative multitasking
+ * system.
+ * <p/>
+ * Note that this argument is a generator, not the function that returns a generator when called. Instantiating
+ * the generator with a function call is the responsibility of the code that instantiates this class.
+ * <p/>
+ * Note 'task_generator.next()' is always called at least once, because 'run()' calls that method before it detects
+ * cancellation. This is required by the interface to a generator, since it's valid to call 'send()', which is how
+ * the runner signals cancellation to the task, only after the first call to 'next'. If, for whatever reason, it's
+ * necessary to detect cancellation before the work begins, the generator should have an extra do-nothing 'yield'
+ * statement at the beginning.
+ *
+ * @param {Number} runaway_limit
+ * Optional. The maximum number of iterations before 'run()' will throw an exception. Disable runaway detection by
+ * setting this value to zero.
+ *
+ */
+var Long_Task = exports.Long_Task = function ( task_generator, runaway_limit )
+{
+ /**
+ * The runner is a generator that performs a single increment of computation with each call to next(). It returns
+ * 'false' when it is not yet completed and 'true' once it has. Calling the generator with 'send( true )' notifies
+ * the generator that it has been cancelled; thereafter it must return 'true' always.
+ * @type {Function}
+ */
+ this.task_generator = task_generator;
+
+ /**
+ * Cancellation flag. Set in the cancel() method. Tested each iteration in run().
+ * @type {Boolean}
+ */
+ this.cancelled = false;
+
+ /**
+ * Runnable flag. This is essentially the state variable of a two-state machine, which starts at "runnable" and
+ * goes to "completed".
+ * @type {Boolean}
+ */
+ this.runnable = true;
+
+ /**
+ * The maximum number of iterations before 'run()' will throw an exception. Disable runaway detection by setting
+ * this value to zero.
+ * @type {Number}
+ */
+ this.runaway_limit = (arguments.length < 2) ? default_runaway_limit : runaway_limit;
+
+ /**
+ * Iteration counter. This is only incremented when a runaway limit is in effect.
+ * @type {Number}
+ */
+ this.count = 0;
+
+ /**
+ * Internal timer. This is an nsITimer rather than the window time to allow background operation.
+ * @type {nsITimer}
+ */
+ this.timer = Cc["@mozilla.org/timer;1"].createInstance( Ci.nsITimer );
+};
+
+/**
+ * Cancel command. Calling this function cancels the pending task as soon as possible, which is nowhere near
+ * immediate with JavaScript.
+ * <p/>
+ * WARNING: The current way that cancellation is implemented, there will be one additional call to the task
+ * generator before cancellation. If that's a problem, it's time to fix the algorithm, which means making
+ * an initial call to 'next()' before setting up the standing loop, and swapping the order of iterating and checking
+ * for cancellation.
+ */
+Long_Task.prototype.cancel = function ()
+{
+ this.cancelled = true;
+};
+
+/**
+ * Run command
+ */
+Long_Task.prototype.run = function ()
+{
+ Long_Task.log( "run begin. runaway_limit = " + this.runaway_limit );
+
+ if ( !this.runnable )
+ {
+ throw "Long_Task no longer runnable";
+ }
+
+ this._run_once();
+};
+
+/**
+ * The main body of the runner.
+ */
+Long_Task.prototype._run_once = function ()
+{
+ /*
+ * The bulk of this function is structure as 'do ... while( false )' in order to use 'break' for flow control,
+ * instead of splitting off a second function and using 'return'.
+ */
+ do
+ {
+ /*
+ * Main iteration call. The call to run() goes into a try-block to ensure we stop gracefully if the generator
+ * throws, since that doesn't always signal an error
+ */
+ try
+ {
+ if ( this.task_generator.next() )
+ {
+ /*
+ * The task generator returned true, which means that it's finished.
+ */
+ break;
+ }
+ }
+ catch ( ex )
+ {
+ if ( ex === StopIteration )
+ {
+ Long_Task.log( "run iteration stopped" );
+ /*
+ * StopIteration is not an error but just signals the end of data for an ordinary iterator. Since the
+ * generator has signalled us, we don't send any signal by calling 'send()'.
+ */
+ break;
+ }
+ else
+ {
+ Long_Task.log( "run iteration exception: " + ex.toString() );
+ throw ex;
+ }
+ }
+
+ /*
+ * Runaway detection.
+ */
+ if ( this.runaway_limit > 0 )
+ {
+ ++this.count;
+ Long_Task.log( "run iteration " + this.count, true );
+ if ( this.count >= this.runaway_limit )
+ {
+ this.cancelled = true;
+ /*
+ * FUTURE: This should really throw an exception after cancelling the generator.
+ */
+ }
+ }
+
+ /*
+ * Cancellation detection.
+ */
+ if ( this.cancelled )
+ {
+ Long_Task.log( "run cancellation begin" );
+ try
+ {
+ /*
+ * We've received a command to cancel from elsewhere. Notify the generator that we're shutting down and
+ * exit the loop. We're doing this within a try-block because the generator will typically throw
+ * StopIteration at this point, which isn't an error.
+ */
+ this.task_generator.send( true );
+ }
+ catch ( ex )
+ {
+ /*
+ * StopIteration is not an error as a result of cancellation, but any other exception is.
+ */
+ if ( ex !== StopIteration )
+ {
+ Long_Task.log( "run cancellation exception: " + ex.toString() );
+ throw ex;
+ }
+ }
+ Long_Task.log( "run cancellation end" );
+ break;
+ }
+
+ /*
+ * Infinite loop behavior happens here, where we schedule ourselves for another run as soon as possible
+ * after we complete. In practice, this means "at the next millisecond timer tick, or after all event handlers
+ * have run, whichever comes last". If the unit of computation is less than 50 ms, the duration overhead will be
+ * over 1%, but it's just fine for scheduling asynchronous events where this internal timer is not the limiting
+ * factor in total duration, such as AJAX calls.
+ */
+ this.timer.initWithCallback( this._run_once.bind( this ), 0, Ci.nsITimer.TYPE_ONE_SHOT );
+ return;
+ } while ( false )
+ this.runnable = false;
+};
+
+Long_Task.log = function ( msg, allow )
+{
+ if ( arguments.length < 2 || allow )
+ Cu.reportError( "Long_Task: " + msg );
+};
« no previous file with comments | « lib/counter_task.js ('k') | no next file » | no next file with comments »

Powered by Google App Engine
This is Rietveld