Forgot to add the compiled JS:

(using Dart Editor version 0.4.1_r19425)

https://github.com/johnmccutchan/tilebasedwordsearch/blob/f42057320f958e0932ed3c9403c8c6752aa24ce9/web/solver/solver.dart.js

I suspect that this is a (very useful) benchmark of the Map class. We have no special handling of maps with int or string keys but they should be implementable using JS objects in a lot of cases.

It would be great if you could try patching in the code in:

   https://codereview.chromium.org/12827018/

too see if the improved Map performance for non-linked hash maps matters.

Using Dart Editor version 0.1.2_r20838

I now see the following solve times:

V8 + Hand written JS: 0.5999999993946403

V8 + dart2js: 3.5500000085448846 ms

Dart VM: 1.6529999993508682 ms

V8 + dart2js is now approx 2.1X slower than the VM, and 5.9X slower than hand-written JS.

Changed the title to: "Boggle solver 2.1X slower on dart2js than Dart VM".

cc +iposva for the Dart VM / V8 timing differences

cc @iposva-google.

Added this to the Later milestone.

Added TriageForM5 label.

Removed TriageForM5 label.

Removed this from the Later milestone.
Added Oldschool-Milestone-Later label.

Removed Oldschool-Milestone-Later label.

This issue is really old... we'll have to measure again. There may not be a performance issue anymore.

There is some work happening on this, though. Traditionally Dart maps are implemetned by dart2js using multiple JS objects, so there is a possibility that the issue still exists. @hterkelsen is looking into making the map implementation faster by using ES6 maps whenever the keys of the map use the identity-hashcode, which should include int and string keys.

@hterkelsen Is this still an issue?

Yes, I have not enabled ES6 maps for Maps with identity-hashcode keys

I imagine this specific benchmark is not super important to us anymore