Bug 164531

Summary:	Spread(ArrayUse:@child) should participate in CSE
Product:	WebKit	Reporter:	Saam Barati <saam>
Component:	JavaScriptCore	Assignee:	Saam Barati <saam>
Status:	NEW ---
Severity:	Normal	CC:	benjamin, ddkilzer, fpizlo, ggaren, gskachkov, jfbastien, keith_miller, mark.lam, msaboff, oliver, ticaiolima, ysuzuki
Priority:	P2
Version:	WebKit Nightly Build
Hardware:	Unspecified
OS:	Unspecified
Bug Depends on:	164258
Bug Blocks:

Description Saam Barati 2016-11-08 15:48:08 PST

After I land https://bugs.webkit.org/show_bug.cgi?id=164258, we will emit Spread(ArrayUse:@child) only when we know that the iterator protocol is proven to be non-observable.
This means we can CSE two Spread(ArrayUse:@child) together when we've proven that nobody changes @child in between the two Spreads.

Comment 1 Saam Barati 2016-11-08 15:48:53 PST

It's important to also note that Spread will allocate a JSFixedArray, but JSFixedArray is a VM internal cell, it's not observable to user code. Therefore, it's ok to simply drop the second allocation and use the first one.

Comment 2 David Kilzer (:ddkilzer) 2018-01-19 13:37:35 PST

Since the comment mentioning this bug was removed in r227229, can this be changed to RESOLVED/WONTFIX?

Comment 3 Saam Barati 2018-01-19 13:41:35 PST

(In reply to David Kilzer (:ddkilzer) from comment #2)
> Since the comment mentioning this bug was removed in r227229, can this be
> changed to RESOLVED/WONTFIX?

I think it's worth keeping this. We could theoretically do this if we speculated in some particular way. Or if we proved the input to Spread doesn't have a symbol iterator.