diff options
-rw-r--r-- | doc/todo/smarter_sorting.mdwn | 33 |
1 files changed, 33 insertions, 0 deletions
diff --git a/doc/todo/smarter_sorting.mdwn b/doc/todo/smarter_sorting.mdwn new file mode 100644 index 000000000..5a6d63ef1 --- /dev/null +++ b/doc/todo/smarter_sorting.mdwn @@ -0,0 +1,33 @@ +I benchmarked a build of a large wiki (my home wiki), and it was spending +quite a lot of time sorting; `CORE::sort` was called only 1138 times, but +still flagged as the #1 time sink. (I'm not sure I trust NYTProf fully +about that FWIW, since it also said 27238263 calls to `cmp_age` were +the #3 timesink, and I suspect it may not entirely accurately measure +the overhead of so many short function calls.) + +`pagespec_match_list` currently always sorts *all* pages first, and then +finds the top M that match the pagespec. That's innefficient when M is +small (as for example in a typical blog, where only 20 posts are shown, +out of maybe thousands). + +As [[smcv]] noted, It could be flipped, so the pagespec is applied first, +and then sort the smaller matching set. But, checking pagespecs is likely +more expensive than sorting. (Also, influence calculation complicates +doing that, since only influences for the M returned pages should be tracked.) + +Another option, when there is a limit on M pages to return, might be to +cull the M top pages without sorting the rest. This could be done using +a variant of Ye Olde Bubble Sort. Take the first M pages, and (quick)sort. +Then for each of the rest, check if it is higher than the Mth page. +If it is, bubble it up so it's sorted. +If not, throw it out (that's the fast bit and why this is not O(N^2)). + +This would be bad when M is very large, and particularly, of course, when +there is no limit and all pages are being matched on. (For example, an +archive page shows all pages that match a pagespec specifying a creation +date range.) Well, in this case, it *does* make sense to flip it, limit by +pagespe first, and do a (quick)sort second. (No influence complications, +either.) + +Adding these special cases will be more complicated, but I think the best +of both worlds. --[[Joey]] |