DFG should inline binary string concatenations (i.e. ValueAdd with string children)

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Filip Pizlo]

And it should do so even if the children are StringObjects, or possibly StringObjects.  This should also lay the groundwork for making this fast:

"foo" + x

where x is an int.

[Filip Pizlo]

Created attachment 193612 [details]
the patch

[Early Warning System Bot]

Comment on attachment 193612 [details]
the patch

Attachment 193612 [details] did not pass qt-ews (qt):
Output: http://webkit-commit-queue.appspot.com/results/17187464

[Early Warning System Bot]

Comment on attachment 193612 [details]
the patch

Attachment 193612 [details] did not pass qt-wk2-ews (qt):
Output: http://webkit-commit-queue.appspot.com/results/17160560

[Oliver Hunt]

Comment on attachment 193612 [details]
the patch

r=me.

I've found myself pondering making us plant calls to the valueOf functions in the DFG, that way the function calls are cacheable and we could add intrinsics for the more common implementations, and custom calls would "just work"

That said i'm not sure how much of a win that would be for concat, etc now that the common case of String objects is handled.  But I do recall toString being sufficiently common on generic objects to be something we might consider investigating in future.

Other than that, make the qt bots non-red (!PLATFORM(QT) ?) and r=me

[EFL EWS Bot]

Comment on attachment 193612 [details]
the patch

Attachment 193612 [details] did not pass efl-ews (efl):
Output: http://webkit-commit-queue.appspot.com/results/17235080

[Build Bot]

Comment on attachment 193612 [details]
the patch

Attachment 193612 [details] did not pass mac-ews (mac):
Output: http://webkit-commit-queue.appspot.com/results/17080771

[Filip Pizlo]

(In reply to comment #4)
> (From update of attachment 193612 [details])
> r=me.
> 
> I've found myself pondering making us plant calls to the valueOf functions in the DFG, that way the function calls are cacheable and we could add intrinsics for the more common implementations, and custom calls would "just work"

Can't do that.  Because you have to do the weirdo thing where you either look up valueOf, or toString if the former finds nothing.  And then convert the result to a string.  Or not a string.  And then either do a numeric addition or a string concatenation.

So, I mean, you could plant all of that.  But you'll be planting a national forest.

> 
> That said i'm not sure how much of a win that would be for concat, etc now that the common case of String objects is handled.  But I do recall toString being sufficiently common on generic objects to be something we might consider investigating in future.
> 
> Other than that, make the qt bots non-red (!PLATFORM(QT) ?) and r=me

[Filip Pizlo]

Landed with 32-bit fixes in http://trac.webkit.org/changeset/146164

[Csaba Osztrogonác]

(In reply to comment #8)
> Landed with 32-bit fixes in http://trac.webkit.org/changeset/146164

It made all inspector tests timeout on ARM traditional platform:
https://bugs.webkit.org/show_bug.cgi?id=112697