This caching complicated the logic but is used by a single subclass (HTMLFormControlsCollection) only. The subclass can do the caching itself.
Created attachment 216180 [details] patch
Comment on attachment 216180 [details] patch View in context: https://bugs.webkit.org/attachment.cgi?id=216180&action=review > Source/WebCore/html/HTMLCollection.cpp:134 > +HTMLCollection::HTMLCollection(ContainerNode& ownerNode, CollectionType type, ElementTraversalType traversalType) I think that the enum is great for the argument. > Source/WebCore/html/HTMLCollection.cpp:144 > + , m_elementTraversalType(traversalType) But for the local, I might suggest we use a boolean named m_requiresCustomForwardTraversal or something like that. In other words, I liked the readability of the old pattern where the data member was a boolean. > Source/WebCore/html/HTMLCollection.h:76 > + Element* traverseFirstElement(ContainerNode* root) const; Nothing new, but I am not a big fan of the name “traverse first element”. The meaning of traverse here is unclear. > Source/WebCore/html/HTMLCollection.h:143 > + virtual Element* virtualElementAfter(Element*) const { ASSERT_NOT_REACHED(); return nullptr; } So we can’t use pure virtual because we only need this for classes that are CustomForwardOnlyTraversal. Ugly. > Source/WebCore/html/HTMLFormControlsCollection.cpp:77 > + unsigned i = 0; > + for (; i < elementVector.size(); ++i) { > + auto& associatedElement = *elementVector[i]; > + if (associatedElement.isEnumeratable() && &associatedElement.asHTMLElement() == &element) > + break; > + } > + return i; I would put the return inside the loop, and use a normal for loop without an “outside the loop” i variable. I think that would also make it slightly clearer what the function returns when nothing is found. But maybe this is more elegant.
http://trac.webkit.org/changeset/158758