Minor optimization of theme() for lazy loaded theme functions

The unset could potentially outweight the cost of include_once. Remember those include_once calls operate on an absolute path so they are quick. Reopen if you ever make benchmarks for this.

Needs benchmarks means it needs review, not that it's won't fix.

We have a tag for that!

Subscribing. My hunch is it's a small net improvement, because the penalty of an unset would only be paid once per theme hook, so I can't imagine any page where it would be detectable, and the savings would be had on every invocation of a theme hook (after the 1st one), so at least in theory, that could add up to something measurable. The question is: is there a use-case of a theme hook that has a 'includes' that is called enough times per page request for the savings to be worth the extra line of code (or perhaps more to the point, worth someone's time to generate the benchmarks and drive the issue along).

theme.patch queued for re-testing.

Hmm tried to benchmark this but couldn't apply and after fuzzing it I got a bunch of errors. I'd consider a won't fix too.

Notice: Indirect modification of overloaded element of ThemeRegistry has no effect in theme() (line 1068 of includes/theme.inc).

Yep. 2010 is a long time ago.