Use finer grained locking in FontDatabase

• Component: Text
• Status: RESOLVED FIXED

[Antti Koivisto]

Constructing a font can take a significant amount of time. We don't need to hold the lock during that.

[Antti Koivisto]

Created attachment 352038 [details]
patch

[Alex Christensen]

Comment on attachment 352038 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=352038&action=review

> Source/WebCore/platform/graphics/cocoa/FontCacheCoreText.cpp:-893
> -        std::lock_guard<Lock> locker(m_descriptorMapLock);
> -
>          auto folded = familyName.foldCase();
> -        return m_familyNameToFontDescriptors.ensure(folded, [&] {

Wouldn't moving the lock_guard to after the call to foldCase have the same result?  we could leave the call to ensure as it is.

[Antti Koivisto]

> Wouldn't moving the lock_guard to after the call to foldCase have the same
> result?  we could leave the call to ensure as it is.

No, the lock needs to be free during the CTFont stuff, that's the slow part (but inherently thread safe). 

We lookup the cache with the lock held. If we don't find anything we free the lock and do the CTFont dance. Then we retake the lock to add it to the cache (which might not do anything because other thread might have already added this font).

[WebKit Commit Bot]

Comment on attachment 352038 [details]
patch

Clearing flags on attachment: 352038

Committed r237040: <https://trac.webkit.org/changeset/237040>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/45199971>