1041586 - Implement Symbol.isConcatSpreadable

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Boris Zbarsky [:bzbarsky]]

We should have it for ArrayClass bits.

Comment 1

[Claude Pache]

Isn't it a dupe of Bug 1054758 ?

Comment 2

[Boris Zbarsky [:bzbarsky]]

Yes, except filed a month earlier.  So the dup is the other way around.  ;)

Not that I care much as long as the dependencies are propagated to whichever bug is used here.

Comment 4

[Tom S]

Changing the implementation of concat for this is really simple. However concat is also inlined in Ion, and just checking for the array class isn't enough anymore.

Comment 5

[Boris Zbarsky [:bzbarsky]]

As pointed out on public-script-coord, we should make @@isConcatSpreadable return undefined on cross-origin windows when we implement this.

Comment 6

[Tom S]

This should be relatively simple with the new selfhosted concat.

Comment 7

[Tom S]

Attachment: Autogenerate symbol names

I was a bit upset that we actually hard-coded @@iterator as the first symbol, so I decided to automatically generate all the symbol atoms. So now you can add new symbols at the frond and end of the enum!

Comment 8

[Tom S]

Attachment: Add Symbol.isConcatSpreadable

Comment 9

[Tom S]

Attachment: IsConcatSpreadable

Normally we would be done now, sadly this causes a regression between 10-25% on a micro benchmark using concat. It seems like the O[@@isConcatSpreadable] lookup is not optimized away and even uses a property cache. I will have to look into completely eliminating that.

Comment 10

[Tom S]

Attachment: IsConcatSpreadable

Comment 11

[Jason Orendorff [:jorendorff]]

Comment on attachment 8739665 [details] [diff] [review]
Autogenerate symbol names

Review of attachment 8739665 [details] [diff] [review]:
-----------------------------------------------------------------

Yes, of course. Thanks.

Comment 12

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/aceaeee7103f

Comment 13

[Tom S]

Attachment: Fold away property accesses to not-defined properties

This patch allows us to completely fold away O[Symbol.isContactSpreadable], this wins back all the performance on a micro benchmark. I tried running and didn't see any obvious, but the results are extremely noisy for me.

Comment 14

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/aceaeee7103f

Comment 15

[Jan de Mooij [:jandem]]

Comment on attachment 8741039 [details] [diff] [review]
Fold away property accesses to not-defined properties

Review of attachment 8741039 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks, that's a nice optimization to have for other code as well.

Comment 16

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/410d733097ac

Comment 17

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/410d733097ac

Comment 18

[Tooru Fujisawa [:arai]]

Comment on attachment 8739669 [details] [diff] [review]
IsConcatSpreadable

Review of attachment 8739669 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good, assuming the testcase is in another patch :)

Comment 19

[Tom S]

Attachment: Tests

Comment 20

[Tooru Fujisawa [:arai]]

Comment on attachment 8741906 [details] [diff] [review]
Tests

Review of attachment 8741906 [details] [diff] [review]:
-----------------------------------------------------------------

Nice testcases :D

Comment 21

[Tom S]

https://hg.mozilla.org/integration/mozilla-inbound/rev/db1b027cbeb1
https://hg.mozilla.org/integration/mozilla-inbound/rev/2021fd52a8c9
https://hg.mozilla.org/integration/mozilla-inbound/rev/35b1a010cb9f

Comment 22

[Carsten Book [:Tomcat]]

sorry had to back this out for assertion failures like https://treeherder.mozilla.org/logviewer.html#?job_id=25962392&repo=mozilla-inbound

Comment 23

[Tom S]

I think that test might be buggy sometimes? Try looked okay https://treeherder.mozilla.org/#/jobs?repo=try&revision=1c3b1879918a actually.

Comment 24

[Pulsebot]

Backout:
https://hg.mozilla.org/mozilla-central/rev/533daf1ab8c9
https://hg.mozilla.org/mozilla-central/rev/fe492b1cbc39
https://hg.mozilla.org/mozilla-central/rev/077bd7786468

Comment 25

[Tom S]

I am just going to assume this test is broken, inserting a print makes the failure go away.

Comment 26

[Tom S]

Attachment: Prevent jit compilation for one expression-autopsy test

It seems like this test case has been causing problems for a long time. I have to admit, I am not interested in fixing this just to get this bug finished.

Comment 27

[Jan de Mooij [:jandem]]

Comment on attachment 8742864 [details] [diff] [review]
Prevent jit compilation for one expression-autopsy test

Review of attachment 8742864 [details] [diff] [review]:
-----------------------------------------------------------------

I'm still a bit curious why your patch exposes this, but given that we have seen this failure before and investigated it last time, this makes sense.

Comment 28

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/bb2748e05cab
https://hg.mozilla.org/integration/mozilla-inbound/rev/1ddd1d9af872
https://hg.mozilla.org/integration/mozilla-inbound/rev/219bb598b6ad
https://hg.mozilla.org/integration/mozilla-inbound/rev/f569192eb5e8

Comment 29

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/bb2748e05cab
https://hg.mozilla.org/mozilla-central/rev/1ddd1d9af872
https://hg.mozilla.org/mozilla-central/rev/219bb598b6ad
https://hg.mozilla.org/mozilla-central/rev/f569192eb5e8

Comment 30

[Florian Scholz (Open Web Docs)]

Developer release notes
https://developer.mozilla.org/en-US/Firefox/Releases/48

Reference pages:
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Symbol
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Symbol/isConcatSpreadable