Reapply "Make Map constructors return LinkedHashMap."

Reapply "Make Map constructors return LinkedHashMap."

Updated tests assuming a specific ordering of an unlinked hash map
to not be daft.

R=floitsch@google.com, jmesserly@google.com

Committed: https://code.google.com/p/dart/source/detail?r=26693

[Lasse Reichstein Nielsen, 2013-08-26 11:38:32]

PTAL

[floitsch, 2013-08-26 13:00:47]

LGTM.

https://codereview.chromium.org/22859069/diff/1/tests/corelib/map_values2_tes...
File tests/corelib/map_values2_test.dart (right):

https://codereview.chromium.org/22859069/diff/1/tests/corelib/map_values2_tes...
tests/corelib/map_values2_test.dart:45: Expect.isTrue(map5.values.first == 43 ||
map5.values.first == 500);
You can now make this deterministic.

[floitsch, 2013-08-26 13:01:40]

CCing John for polymer test changes and Konstantin for editor changes.

[scheglov, 2013-08-26 14:53:52]

LGTM

[Jennifer Messerly, 2013-08-26 17:15:39]

LGTM!

https://codereview.chromium.org/22859069/diff/3001/pkg/polymer/test/css_test....
File pkg/polymer/test/css_test.dart (right):

https://codereview.chromium.org/22859069/diff/3001/pkg/polymer/test/css_test....
pkg/polymer/test/css_test.dart:415: // Check for warning messages about var-
cycles in no expected order.
Oh wow, it was depending on hash order? This test is bad and it should feel bad!

Thanks for the fix :)

https://codereview.chromium.org/22859069/diff/3001/pkg/polymer/test/css_test....
pkg/polymer/test/css_test.dart:418: for (var errorMessage in messages.messages)
{
just brainstorming here, I think if I was the one writing this test I would do
something like:

    var msgs = messages.messages.map(
        (e) => '${e.span.formatLocation}|${e.message}|${e.span.text}');

    expect(msgs, unorderedEquals([
      'foo.css:12:23|var cycle detected var-def-1|@def-1: var(def-2)',
      'foo.css:9:23|var cycle detected var-five|@five: var(six',
      ...
    ]);

probably not worth the refactoring at this point tho :)

[Lasse Reichstein Nielsen, 2013-08-27 08:41:31]

https://codereview.chromium.org/22859069/diff/1/tests/corelib/map_values2_tes...
File tests/corelib/map_values2_test.dart (right):

https://codereview.chromium.org/22859069/diff/1/tests/corelib/map_values2_tes...
tests/corelib/map_values2_test.dart:45: Expect.isTrue(map5.values.first == 43 ||
map5.values.first == 500);
Good point.

[Lasse Reichstein Nielsen, 2013-08-27 08:44:04]

Committed patchset #3 manually as r26693 (presubmit successful).

[srdjan, 2013-08-28 14:48:51]

Havlak benchmark lost 14% with this CL. Could you investigate why?

[Lasse Reichstein Nielsen, 2013-08-29 12:19:13]

It's probably because it uses "new Map" a lot, and is hurt by the overhead of
using a linked hash map instead of just a hash map.
Luckily I have a change that makes linked hash maps faster too, so that should
outweigh the problem.