Split from bug 715181, see bug 715181 comment 127. Mrrrgn, can you take a look at this and see whether we should take it? 4esn0k, can you turn this into a proper patch so we can land it?

Comment on attachment 8733845 [details] Example JS of optimized sort This does indeed improve performance. Thanks for filing this !

Comment on attachment 8733927 [details] [diff] [review] MergeSort.patch Review of attachment 8733927 [details] [diff] [review]: ----------------------------------------------------------------- Thanks for the patch ! :) Moving to a buffer is alright, but we need to use List instead of std_Array_sort, see the failing test: Morgans-MacBook-Pro:_DBG.OBJ mrrrgn$ ../tests/jstests.py dist/bin/js ecma_6/Array/sort ## ecma_6/Array/sort_holes.js: rc = 3, run time = 0.50141 uncaught exception: Illegally touched the array's setter! (Unable to print stack trace) REGRESSION - ecma_6/Array/sort_holes.js [2|1|0|0] 100% ======================================================>| 1.6s REGRESSIONS ecma_6/Array/sort_holes.js FAIL

* to a single buffer

(In reply to Morgan Phillips [:mrrrgn] from comment #2) > Comment on attachment 8733845 [details] > Example JS of optimized sort > > This does indeed improve performance. Thanks for filing this ! We'll need to see how much of this performance boost came from using std_Array_sort, which we need to avoid.

`std_Array`, not `std_Array_sort`. var x = []; x[0] = 1; - is there some efficient way to do this without touching the prototype?

`std_Object_setPrototypeOf` makes it slow, `_DefineDataProperty` too... Nice... no way to use Arrays here.

without `std_Array`

(In reply to 4esn0k from comment #7) > `std_Object_setPrototypeOf` makes it slow, `_DefineDataProperty` too... > Nice... no way to use Arrays here. Yeah, I was going to suggest trying _DefineDataProperty but wasn't sure it would perform as well as List :( Thanks for the new patch. :)

Comment on attachment 8734261 [details] [diff] [review] MergeSort.patch Review of attachment 8734261 [details] [diff] [review]: ----------------------------------------------------------------- Nice. I'll land this after a try run. :)

Try run: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3476ff0134e5

I had to back this out because it broke a test: https://treeherder.mozilla.org/logviewer.html#?job_id=24640920&repo=mozilla-inbound https://hg.mozilla.org/integration/mozilla-inbound/rev/ac482faf3373 For the record, this showed up in the try run in comment 11.

The devtools test failure in that try run is also happening when this landed: https://treeherder.mozilla.org/logviewer.html#?job_id=24640238&repo=mozilla-inbound

Apologies, I took a quick look and believed it was an intermittent.

I do not know what is wrong and how to investigate this. Previous patch changed the behaviour of `sort`, when `comparefn` is incorrect. I have updated the patch to use `comparefn(...) <= 0` as it was before.

updated patch

Could you try this patch?