Ran into this while profiling XHTML node creation. Why are we bothering to stringify the atom, hash the string, etc? Imho we should just have a hashtable mapping atoms to IDs here or something.

I'm working on this

This patch adds a hash directly from atom -> id. I also inlined some of the functions for additional speed goodness. This forced me to move some of the statics into the class which wasn't too pretty, but I think it's ok.

Comment on attachment 241219 [details] [diff] [review] Patch to fix >Index: parser/htmlparser/public/nsHTMLTags.h >+ friend void TestTagTable(); Just make that a static member function, perhaps? Otherwise you probably need to declare it separately before the |friend| thing; some of our compilers don't deal with effectively declaring stuff via the |friend| keyword. >Index: parser/htmlparser/src/nsHTMLTags.cpp >-static const PRUnichar* const kTagUnicodeTable[] = { >+const PRUnichar* const nsHTMLTags::sTagUnicodeTable[] = { This doesn't make me all that happy, but given the inlining I don't see a better way. :( >+HTMLTagsHashCodeAtom(const void *key) >+{ >+ return NS_PTR_TO_INT32(key); Toss in a ">> 2" there because otherwise all keys will end with 00? With that, looks good.

Checked in.

Comment on attachment 241759 [details] [diff] [review] Add testing for new code r+sr=jst