GetNameSpaceID() can get used a fair amount of time in DHTML applications. For example on Bugzilla query page clicking on a product and moving the mouse a tiny bit it can get called about 10 times. Just starting up it is called only 3 times, and it won't be called at all during jrgm's page load tests. It would need to be tested if it gets called a lot with non-interactive DHTML, like animations. Anyway, this is not a huge win, GetNameSpaceID() takes about 1ms. From brendan in bug 41983: can we deCOMtaminate GetNameSpaceID and its descendents, and also wean them off of nsHashtable and onto PLDHashTable instead? That should save a fair amount of time. http://lxr.mozilla.org/mozilla/source/content/base/src/nsNameSpaceManager.cpp#476 http://lxr.mozilla.org/mozilla/source/content/base/src/nsNameSpaceManager.cpp#139 I don't see any reason for infallible methods to return NS_OK directly, and to return their real result via a reference out param. DeCOMtaminate! I don't see any reason to subroutine FindNameSpaceID if you can look directly into a global PLDHashTable with PL_DHashTableOperate(&gURIToIDTable, aURI.get(), PL_DHASH_LOOKUP), cast the returned entry pointer down to an nsURIToIDEntry*, and return that entry's mID (or else return kNameSpaceID_Unknown if PL_DHASH_ENTRY_IS_FREE(entry)). You'll save on malloc overhead elsewhere, too (when adding to the table). If I can help with pldhash.h usage, please let me know. /be

It doesn't seem like this would be a big win in any situation, futuring.

Comment on attachment 194229 [details] [diff] [review] v1 I've changed NS_INAMESPACEMANAGER_IID in my tree.

Comment on attachment 194229 [details] [diff] [review] v1 r+sr=jst