Set the shouldLazyAttach flag to true in the tree builder's adoption agency algorithm
Created attachment 63696 [details] Patch
I don't fully understand why the render tree output is different. Both the before and after seem sane to me.
Comment on attachment 63696 [details] Patch I'd like to see the dom diffs on these tests. They're easy to generate with a dump-as-markup script tag.
WebCore/benchmarks/parser/html-parser.html says: before: avg 3172.45 stdev 23.66743543352342 after: avg 3172.5 stdev 26.693632199459106 This makes sense - I would not expect the HTML5 spec to have many misnested tags requiring the adoption agency.
Running the tests as dump-as-markup shows no changes with or without the lazy attach flag set. It's render tree only diffs.
Comment on attachment 63696 [details] Patch Awesome. LGTM.
Thanks for the review. Will land this tomorrow so I can take care of any platform-specific baselines that need updating.
We desperately need to move to the world of full lazy attach.
I think we had skipped the hang test on gtk before because it was too slow. Now it might be fast enough!
Committed r64954: <http://trac.webkit.org/changeset/64954>
Landed, watchin' the bots now to see if anything needs rebaselines that I couldn't generate myself. I (perhaps optimistically) unskipped residual-style-hang on Gtk to see if it runs OK now - if it's too slow I'll just skip it again.
http://trac.webkit.org/changeset/64954 might have broken Qt Windows 32-bit Debug
Comment on attachment 63696 [details] Patch Clearing flags (was abarth r+)
(In reply to comment #12) > http://trac.webkit.org/changeset/64954 might have broken Qt Windows 32-bit Debug Sorry, it was false positive alarm. I fixed the bot.