On Mon, Apr 23, 2018 at 5:18 PM David Lloyd <david.ll...@redhat.com> wrote:
> FWIW I strongly doubt this will improve performance; probably the > opposite in fact, as IIRC an enum switch generates an extra class > (though perhaps this has changed). The original code was quite > compact and utilized identity comparisons, and given there are only > three alternatives it probably was able to exploit branch prediction > as well (if such a thing even matters in this context). Well, there are enum switches on the same enum elsewhere in the class, so should those instead be replaced by if checks? A larger change could remove this branch entirely, with different classes for each of the types, which are known during compile.
