Last modified: 2014-11-18 21:20:59 UTC
Take a look at: http://parsoid.wmflabs.org:8001/perfstats?orderby=time:parse Take a look at the first entry there: https://ja.wikipedia.org/w/index.php?title=%E6%9D%91%E6%AD%A3%E3%81%AE%E5%A6%96%E5%88%80&action=edit On this page, check one of the transclusions and look at the expanded source on it as follows: "node parse --prefix ja --dump tplsrc < /tmp/wt_snippet" and you find this: .... |[[ファイル:Go u.svg|20x20px]] |[[ファイル:Go u.svg|20x20px]] |[[ファイル:Go u.svg|20x20px]] |[[ファイル:Go u.svg|20x20px]] |[[ファイル:Go u.svg|20x20px]] .... Now, Parsoid shoots off an API request for each image whereas this is unnecessary after the first response. This can significantly speed up the parse of this page. The same is true for https://pl.wikipedia.org/wiki/Podsumowanie_start%C3%B3w_zespo%C5%82u_Lotus_w_Formule_1 which has a lot of identical image flags. Caching could help here.
Arlo, IIRC you worked in this area. Is this case already handled?
It should be. In 6e3c87504514201fc628c51977b3fd14a37667b0 we cached ImageInfoRequests.