Add reactive-inspired stream transformers:

lib/src/transformers/group_by.dart

+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+import "dart:async";
+import "dart:collection" show HashMap;
+
+/// A group created by [GroupBy].
+///
+/// The stream created by [GroupBy] emits a [Group] for each distinct key
+/// it sees.
+/// This group contains the key itself, along with a stream of the values
+/// associated with that key.
+class Group<K, V> {

[nweiz, 2016/03/01 02:10:04]

What do you think of calling this "GroupByGroup"? "Group" on its own is a pretty
broad term, and this is a pretty widely-used library, so it's likely to cause
conflicts. I think it may actually cause some internal libraries in test to
break, in fact.

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

It's a long name for something fairly inconsequential.
What this basically is is a Pair<K,Stream<V>>.

We generally don't use unnamed pairs because it's better (for some value of
"better") to have properly named accessors instead of "first" and "second", but
it has the disadvantage that we have to name a pair-class that itself doesn't
really correspond to any concept, it is just a pair. So, no obvious good name.

Using "GroupBy" as prefix does show the connection, but it's also redundant when
you use it as part of the GroupBy signature.

How about GroupValues or KeyGroup or ValuesByKey?

[nweiz, 2016/03/08 23:57:34]

I like "ValuesByKey".

+  /// The key that identifiers the values emitted by [values].
+  final K key;
+
+  /// The [values] that [GroupBy] have grouped by the common [key].
+  final Stream<V> values;
+
+  Group(this.key, this.values);
+}
+
+/// Groups events by a computed key.
+///
+/// A key is extracted from incoming events, and a stream is created for each
+/// unique key. A value is computed from the event as well, and emitted on the
+/// corresponding stream.
+///
+/// The returned stream emits a [Group] object for each distinct key seen
+/// by the transformation, and the values associated with the key are output
+/// on the [Group.values] stream.
+///
+/// An error on the source stream, or when calling the `key` functions,
+/// will emit the error on the returned stream.
+///
+/// An error when calling the `value` function, but with a successful call
+/// of the `key` function, is reported on the stream for the corresponding key.
+///
+/// Canceling the subscription on the returned stream will stop processing
+/// and close the streams for all groups.

[nweiz, 2016/03/01 02:10:04]

The cancel and pause behavior feels wrong. Semantically, the stream of groups is
distinct from any group's stream, and as a user I wouldn't expect any
modification of it to affect the other streams—it would feel like
action-at-a-distance.

Here's a more concrete example of where this could be an issue:

var groups = await stream.transform(new GroupBy(key: (value) => value.isEven))
    .take(2).toList();
var evens = groups[groups[0].key == true ? 0 : 1];
var odds = groups[groups[0].key == false ? 0 : 1];

On the surface, this looks like a reasonable way to split a stream into even and
odd components, but in practice it will stop emitting events as soon as at least
one even and one odd value are encountered in the original stream.

I'd make this behave like StreamSplitter, and only pause the underlying stream
if *all* derived streams are paused, and same for canceling. I think that best
represents what the user is likely to mean when they cancel individual streams.

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

Agree. 
Maybe a better behavior is to keep trucking on the source stream, but not
creating any new groups - instead values that doesn't correspond to an existing
group are just discarded (which is effectively the same as creating a new group
and sending it on the cancelled stream, just without storing the events that you
don't need).
True. And it's code you *do* want to write because the toList will not work
without the take(2) - the transformed stream will only output two events, but it
won't close until it has seen the last event.
I considered this for pause, but that doesn't work. You may have paused the six
groups you have seen so far, but the next event may create a new group, so
pausing the source is overdoing it. If the group stream has been cancelled, then
you can start pausing when all value streams are paused - because you actually
*know* all the streams then.

[nweiz, 2016/03/08 23:57:34]

When I say "all derived streams", I'm including the Stream<Group> as well as all
the Stream<V>s. If *all* of them are paused or canceled, it's safe to pause the
source because no new events or groups will be processed.

+///
+/// Pausing the subscription on the returned stream will pause processing
+/// and no further events are added to streams for the individual groups.
+///
+/// Pausing or canceling the a group stream has no effect other than

[nweiz, 2016/03/01 02:10:04]

"the a" -> "a"

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

Acknowledged.

+/// on the individual stream. Events will be queued while the group stream
+/// is paused and until it is first listened to.
+class GroupBy<S, K, V> implements StreamTransformer<S, Group<K, V>> {
+  final Function _key;
+  final Function _value;
+
+  /// Groups values returned by [value] by the key returned by [key].
+  ///
+  /// If [key] or [value] (or both) are omitted, they default to the identity
+  /// function. In that case, [K] and/or [V] should be types that [S] is
+  /// assignable to.
+  GroupBy({K key(S source), V value(S source)})

[nweiz, 2016/03/01 02:10:04]

It seems very unlikely that anyone would want to omit the key and get a bunch of
streams of identical events.

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

If you omit both functions, you will get streams of *equal* values, not
identical. So, you can group equal events, keyed by the first one from each
equivalence class.

[nweiz, 2016/03/08 23:57:34]

That's true, but it's very rare in Dart to have objects that are == without
being functionally identical. I think it's fine to require users to explicitly
write "(key) => key" in that case to avoid making them write "key:" in all the
others.

+      : _key = key ?? _identity,
+        _value = value ?? _identity;
+
+  // Helper function.
+  static _identity(x) => x;
+
+  Stream<Group<K, V>> bind(Stream<S> stream) {
+    var controller;
+    controller = new StreamController<Group<K, V>>(sync: true, onListen: () {
+      var groupControllers = new HashMap();
+      void closeAll() {
+        for (var groupController in groupControllers.values) {
+          groupController.close();
+        }
+      }
+      var subscription = stream.listen((data) {
+        K key;
+        try {
+          key = _key(data);
+        } catch (e, s) {
+          controller.addError(e, s);
+          return;
+        }
+        var groupController = groupControllers[key];

[nweiz, 2016/03/01 02:10:04]

putIfAbsent?

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

I also need to do controller.add for a new value, but not for an existing one.
That would mean putting the controler.add inside the "ifAbsent" function, which
I think is too complicated - I prefer to have "ifAbsent" *only* create the new
object, and not also do more with it.

+        if (groupController == null) {
+          groupController = new StreamController<V>(sync: true);
+          groupControllers[key] = groupController;
+          controller.add(new Group<K, V>(key, groupController.stream));
+        }
+        V value;
+        try {
+          value = _value(data);
+        } catch (e, s) {
+          groupController.addError(e, s);
+          return;
+        }
+        groupController.add(value);
+      }, onError: controller.addError, onDone: () {
+        controller.close();
+        closeAll();
+      });
+      controller.onPause = subscription.pause;
+      controller.onResume = subscription.resume;
+      controller.onCancel = () {
+        subscription.cancel();
+        closeAll();
+      };
+    });
+    return controller.stream;
+  }

[nweiz, 2016/03/01 02:10:04]

Nit: putting some newlines in here would help me read it.

[Lasse Reichstein Nielsen, 2016/03/01 16:51:17]

Acknowledged.

+}