Bug 99990 – sCurrentPaintTimeStamp is not initialized when FrameView::paintContents returns in the middle

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[KyungTae Kim]

If it is TopLevelPainter, sCurrentPaintTimeStamp is set as currentTime() in the top of FrameView::paintContents,
and is reset as 0 in the bottom of the same function.
But, when the FrameView::paintContents returns in the middle, it is not reset as 0, and the value will be not updated anymore.

That value is used for MemoryCache and there was a bug that the MemoryCache is not pruned because that value is not reset.

[KyungTae Kim]

Created attachment 170014 [details]
Patch

[Darin Adler]

Comment on attachment 170014 [details]
Patch

(Not your fault, but why is sCurrentPaintTimeStamp a time instead of just a boolean?)

Seems to me that a better fix is to set up isTopLevelPainter after the early returns, rather than adding these two new exit sequences.

[KyungTae Kim]

For example, sCurrentPaintTimeStamp is referenced with currentPaintTimeStamp() in MemoryCache::::pruneLiveResourcesToSize to measure elapsedTime.

void MemoryCache::pruneLiveResourcesToSize(unsigned targetSize)
{
    double currentTime = FrameView::currentPaintTimeStamp();
...
    double elapsedTime = currentTime - current->m_lastDecodedAccessTime;
    if (elapsedTime < cMinDelayBeforeLiveDecodedPrune)
        return;
    current->destroyDecodedData();

So, this value need to be a time, 
and this value should not be initialized before the paintContents is exited.
(because it can be accessed by other functions during painting)

I considered using some constructor/destructor for initializing it automatically on exit, but It will become too complex.

[KyungTae Kim]

Created attachment 171793 [details]
Patch

[WebKit Review Bot]

Comment on attachment 171793 [details]
Patch

Clearing flags on attachment: 171793

Committed r133391: <http://trac.webkit.org/changeset/133391>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.