[08:33:33] errand [10:09:16] Trey314159 (for when around): I see in your notes: "Not sure if I should deprioritize Kuromoji analysis in favor of finishing up unpacking". What do you think? I'm missing context to have an opinion. [10:50:29] lunch [11:05:13] lunch [11:55:23] lunch [14:19:45] o/ [15:28:28] ryankemper did we ever restart eqiad to apply the new LTR plugin? [15:39:16] inflatador: I think so yes, I remember Erik saying that the jobs requiring the new version worked properly now [15:47:41] gehel, basically, we can narrow the implicit scope to "unpacking" rather than "improving analyzers" for the current epic. [15:47:55] We didn't get around to discussing it in the Wednesday meeting, but unpacking is a lot less work per language than working on Kuromoji, and unpacking will yield benefit in the future, though Kuromoji may not. And Kuromoji is distinct from unpacking (I added it into the list when I was young and naive and hopeful, because it is somewhat related). [15:47:57] We could focus on current unpacking and then work on getting additional benefits from unpacking (e.g., word_break_helper and acronym processing, and other global improvements), and then come back to Kuromoji in the future if we have time and interest. [15:49:31] So if I understand correctly, the initial plan was to focus on unpacking all analyzers. Kuromoji was added scope (that might have made things easier / better). But current understanding is that Kuromoji is more difficult than expected. [15:49:53] If I'm understanding correctly, we should probably focus on unpacking and then see if we want to invest more. [15:55:46] The backstory is almost right.. I included Kuromoji for variety and because it was related. It's not more difficult than expected, but it does take more time than unpacking, and I was kind of thinking, "why fix cjk when you can replace it?", but I've decided I should fix it anyway (and I have now). But your conclusion is correct: we could just focus on unpacking and then see if we want to invest more. [15:56:14] Sounds like a plan! [15:56:27] ok! [15:57:32] In other news, a reminder for everyone: Monday is a holiday in the US, so many of us will be out then. [16:00:53] \o [16:01:17] o/ [16:05:50] dcausse thanks , looks like ltr plugin 1.5.4-wmf1-es7.10.2 in both eqiad. and codfw [16:20:18] at coffee shop, will not make gaming mtg today [16:33:17] I'll skip the unmeeting as well. ERC working group... [16:42:24] has anyone ever tried reading kafka streams over SSH tunnel? I'm getting failures for DNS resolution [16:43:52] inflatador: it might work if using SSH proxies. Unlikely to work over a tunnel since the kafka broker will redirect to the various kafka nodes (I might have the wrong terminology, Kafka isn't my area) [16:44:39] gehel makes sense, I'll try the proxy approach [17:04:55] going offline, have a nice week-end [18:21:33] lunch, back in ~90m [20:38:44] ebernhardson: as discussed earlier, if you are looking for a diversion on a Friday afternoon, Cindy is still not behaving. The exact same seemingly irrelevant tests have now failed five times. OTOH, it's Friday and you certainly don't have to look at anything right this minute! [20:57:59] back [21:02:40] :S [21:49:49] Trey314159: i realized after looking at it...there is a real problem there. Typing `África` and pushing enter used to take you to the page, now it takes us to the search results page instead [21:50:07] not with your patch specifically though, that would be some sort of change in core most likely [21:50:44] Ahh.. right. Ugh. Thanks for checking [21:57:07] ebernhardson, messing around with it, the behavior is very weird. On the main page, searching in the Go box gives no suggestions and then takes you to the results page (where there is one more result than earlier today). If you search in the Go box on the Search results page, you do get suggestions and go directly to the page. Sounds like a UI change of some sort, maybe redefining the box on the main page? [22:27:07] i'm not quite sure yet...i'll try and find an old version of core that it still worked on and do a git bisect from there, should turn up something (if the change was in core) [23:13:51] I can't say i understand why, but after https://gerrit.wikimedia.org/r/c/mediawiki/skins/Vector/+/876040 i no longer get autocomplete on some pages, and it looks like when autocomplete doesn't work vector sends users to the fulltext search page