Bug 185411 – [Simple line layout] Cache run and line resolvers.

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[zalan]

Whenever a (run/line)resolver is constructed (for painting/hit testing) we iterate through and collect all the flow's content in FlowContents's c'tor.
In case of long text content with large amount of elements (foobar1<br>foobar2<br>.....foobar9999<br>) this repeated work could hang the WebProcess.

[zalan]

Created attachment 339794 [details]
Patch

[Radar WebKit Bug Importer]

<rdar://problem/40047659>

[Antti Koivisto]

Comment on attachment 339794 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=339794&action=review

> Source/WebCore/rendering/SimpleLineLayout.cpp:979
> +    , m_runResolver(std::make_unique<RunResolver>(blockFlow, *this))
> +    , m_lineResolver(std::make_unique<LineResolver>(*m_runResolver))

Doesn't this massively increase memory use of simple line layout? FlowContents objects have way more per-line memory use than Layout, that's why they have been temporaries so far. Can we have a cache that is created on-demand and goes away when not needed (with a timer for example).

[zalan]

Comment on attachment 339794 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=339794&action=review

>> Source/WebCore/rendering/SimpleLineLayout.cpp:979
>> +    , m_lineResolver(std::make_unique<LineResolver>(*m_runResolver))
> 
> Doesn't this massively increase memory use of simple line layout? FlowContents objects have way more per-line memory use than Layout, that's why they have been temporaries so far. Can we have a cache that is created on-demand and goes away when not needed (with a timer for example).

Absolutely! I was going to create them on demand (hence the std::unique<>). -though destroying them on a timer didn't occur to me. That's a good idea.

[Antti Koivisto]

Actually since FlowContents::Segments are per-renderer, the resolvers are not that big in any reasonable case (and unreasonable ones benefit from caching). I think this approach is actually fine as-is. You probably don't need to cache RunResolver and LineResolver separately though.

[zalan]

Created attachment 339834 [details]
Patch

[Antti Koivisto]

Comment on attachment 339834 [details]
Patch

r=me

[zalan]

Created attachment 339885 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 339885 [details]
Patch

Clearing flags on attachment: 339885

Committed r231529: <https://trac.webkit.org/changeset/231529>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.