It does too many hashtable lookups and temporary string creations etc.

Also atomizations, free() calls... Lots of stuff in there.

yup, it all reduces to one hashtable lookup.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=05b3bdb6e24afbc24c8f208f9ceb21f9e42dc785 This reduces the temporary string creation and hashtable lookups.

Comment on attachment 8753390 [details] [diff] [review] patch I don't like adding bool argument because it's not unclear from caller side. But I have no better idea.

yeah, the method call is a bit ugly, but luckily this is used rarely. And since the method takes weird params anyhow, one needs to look at the documentation each time one uses it.

Hmm, I need to look at the SVG/SMIL events. I may have made a mistake there. https://treeherder.mozilla.org/#/jobs?repo=try&revision=5d47a2084eda

We pass the test current, and with this patch, but the previous patch didn't. And yes, SVG/SMIL events are crazy. (looks like try is closed atm)

Comment on attachment 8754545 [details] [diff] [review] v2 + tests for SMIL/SVG >+ if (sStringEventTable->Get(aName, &mapping)) { >+ if (mapping.mMaybeSpecialSVGorSMILEvent) { >+ // Try the atom version so that we should get the right message for >+ // SVG/SMIL. >+ atom = NS_Atomize(NS_LITERAL_STRING("on") + aName); >+ msg = GetEventMessage(atom); The events are not so many, so, it might be faster to use switch statement instead of hashtable. However, I guess that the hashtable's cost is not so better than switch statement and this is stronger for maintain. I think that we should make the SVG/SMIL events constructed as usual in another bug, though.