From: René Scharfe Date: Fri, 29 May 2026 15:32:58 GMT Subject: Re: [PATCH] commit-reach: stop sorting in paint_down_to_common() Message-ID: <107314f5-0057-4ed3-9bee-9dca4f424bd1@web.de> In-Reply-To: <20260529084325.GF1106035@coredump.intra.peff.net> On 5/29/26 10:43 AM, Jeff King wrote: > On Wed, May 27, 2026 at 05:52:17PM +0200, René Scharfe wrote: > >> None of the three callers of paint_down_to_common() care about the order >> of its result list: merge_bases_many() sorts it again after removing >> stale items, remove_redundant_no_gen() and repo_in_merge_bases_many() >> throw the list away without even looking at it. So drop the unnecessary >> commit_list_sort_by_date() call. > > Seems like an easy win. If some of the callers do not even look at the > result, could we avoid building it at all in those cases (e.g., by > passing in a NULL result pointer)? Yes, at the cost of adding NULL checks to paint_down_to_common(). Which is probably worth it. > I guess there is not much to be gained, though. The result is a list of > merge bases, so it should usually be rather small. The benefit in your > patch is probably not performance, but just reducing the size of the > code. True. The list can be arbitrarily long, but should only contain a handful commits in normal repos. René