Bug 99639 – V8StringResource's low-number cache should either be pushed into WTF or removed if it's no longer needed

• Status: RESOLVED
• Resolution: INVALID
• Component: New Bugs

[Eric Seidel (no email)]

V8StringResource's low-number cache should either be pushed into WTF or removed if it's no longer needed

http://trac.webkit.org/browser/trunk/Source/WebCore/bindings/v8/V8StringResource.cpp#L109

// Fast but non thread-safe version.
String int32ToWebCoreStringFast(int value)
{
    // Caching of small strings below is not thread safe: newly constructed AtomicString
    // are not safely published.
    ASSERT(isMainThread());

    // Most numbers used are <= 100. Even if they aren't used there's very little cost in using the space.
    const int kLowNumbers = 100;
    DEFINE_STATIC_LOCAL(Vector<AtomicString>, lowNumbers, (kLowNumbers + 1));
    String webCoreString;
    if (0 <= value && value <= kLowNumbers) {
        webCoreString = lowNumbers[value];
        if (!webCoreString) {
            AtomicString valueString = AtomicString(String::number(value));
            lowNumbers[value] = valueString;
            webCoreString = valueString;
        }
    } else
        webCoreString = String::number(value);
    return webCoreString;
}

it seems silly to implement this at the v8 binding layer.  If this is a big win for a small memory hit, then all of WebCore should benefit from this.  But I'm not convinced it's a win (or at least I'd like to know how JSC avoids needing it).