We allocate one new StringImpl every time we need a CSS property name.
Created attachment 158713 [details] Patch
Comment on attachment 158713 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=158713&action=review > Source/WebCore/css/makeprop.pl:141 > + static AtomicString* propertyStrings = new AtomicString[numCSSProperties]; // Leaked intentionally. It's not "leaked", it's just never destroyed. > Source/WebCore/css/makeprop.pl:153 > + return String(getPropertyNameAtomicString(id).impl()); Why not AtomicString::string()?
Created attachment 158950 [details] Patch
Addressed Alexey's comments + added perf numbers. My microbenchmark: http://jsperf.com/getpropertyname-memory-wasting
Comment on attachment 158950 [details] Patch ok
Comment on attachment 158950 [details] Patch Clearing flags on attachment: 158950 Committed r125934: <http://trac.webkit.org/changeset/125934>
All reviewed patches have been landed. Closing bug.