Measure cache size more accurately

[Antti Koivisto]

Currently we just estimate from the item count.

[Antti Koivisto]

Created attachment 246253 [details]
patch

[Antti Koivisto]

rdar://problem/19760224

[Chris Dumez]

Comment on attachment 246253 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=246253&action=review

r=me with comments

> Source/WebKit2/NetworkProcess/cache/NetworkCacheStorageCocoa.mm:458
> +            m_approximateSize += bodyOffset + bodySize;

Why are we adding an offset to the size here? Maybe it is just a naming issue but it looks wrong.

> Source/WebKit2/NetworkProcess/cache/NetworkCacheStorageCocoa.mm:554
> +    static const double deletionProbability { 0.25 };

Maybe a ASSERT(RunLoop::isMain()); before this?

> Source/WebKit2/NetworkProcess/cache/NetworkCacheStorageCocoa.mm:572
> +            bool shouldDelete = randomNumber() < deletionProbability;

Is it safe to call randomNumber() from another thread? The implementation seems to call sharedRandomNumberGenerator() which is a singleton.

[Antti Koivisto]

> Is it safe to call randomNumber() from another thread? The implementation
> seems to call sharedRandomNumberGenerator() which is a singleton.

sharedRandomNumberGenerator() is ARC4RandomNumberGenerator which is thread safe (internal mutex).

[Antti Koivisto]

http://trac.webkit.org/changeset/179823