[18:35:44] He, I'm looking at the code path around the recordOldUpload() call and noticing that the $timestamp it passes to the DB is not the same as it uses for the archive filename. The latter is determined as now(). [18:35:54] via publishTo() [23:07:24] TimStarling: if you have some time, could you have a look at https://gerrit.wikimedia.org/r/c/mediawiki/core/+/702219 ? I'm not sure I'm right there, I can't understand why WikiPage::getContentModel and Title::getContentModel both exist and have very different implementations, but I think they both should always provide the same result [23:40:02] for performance reasons, in the original ContentHandler project at least, we didn't want to populate or use content models in the database for english wikipedia pages [23:40:31] so the solution was to deduce content models from titles [23:41:46] I wanted content models for existing pages to always be null in the database -- I'm not sure that's true anymore after MCR [23:42:37] how exactly you present that in terms of Title and WikiPage methods is open to criticism -- I'm just explaining the reason for the conceptual split [23:44:13] I think originally the Title methods only gave you the configured content model, based on title only, and the WikiPage methods gave you the composite content model [23:48:31] hm. ok, thank you. I'll investigate some more. like, right now page_content_model is written to the page table in WikiPage::updateRevisionOn unconditionally [23:49:55] I'm looking at f2b218381bfa0d8fc18df5578fa29e269e8b6ede [23:50:11] change from 2012 in which database access was added to Title::getContentModel() [23:52:04] Daniel is the authority on this code, I was just a reviewer on some of these changes [23:53:17] yeah, talked to Daniel today, he doesn't really know either :) [23:54:19] heh, I have a better memory for discussions we had 10 years ago [23:55:13] heh... you did give me more to think about [23:56:09] I'll look more into it, but do I understand correctly that you're saying that deducing from main slot and then storing in wan cache is preferrable? [23:58:09] no, loading from page_content_model should be more efficient [23:59:41] I think you're discovering some architectural cruft left over from a change made in 2012