[01:31:06] legoktm: Not that I know of. But at some point I may submit a patch to remove BC stuff. [01:36:26] anomie: Hmm. I actually just finished prepping a Scribunto Debian package, which needs to be compatible with the LTS (1.27 right now). But as long as that branch would continue to recieve security fixes, it would be fine [02:57:41] anomie: Regarding https://phabricator.wikimedia.org/T50133 - what would you think about a radical approach to consistently expand // as https:// in wfExpandUrl? [02:57:52] Drop the whole relative url thing. [03:34:56] Sounds like https://phabricator.wikimedia.org/T54253 a bit. [13:24:57] Krinkle: That'd break on third-party non-HTTPS wikis. [19:12:54] Krinkle: Can't you accomplish that by simply setting $wgServer to a non-protocol-relative value? [19:13:46] Relative urls are yucky. [19:14:09] They were a useful migration tool for us to avoid cache-busting, but their use has proliferated. [19:14:21] I don't remember all the PROTO_* stuff as well as I should given that I designed it, but it's pretty sensitive to which protocol you use in $wgServe [19:14:29] If a site uses HTTPS, link to HTTPS. If a site supports both, still link to HTTPS. [19:14:52] Why would you *want* to link to HTTP if you know they support both HTTPS? [19:22:56] T54253, T33327, T118413 [19:22:56] T118413: Wikimedia wikis should use https:// in $wgServer - https://phabricator.wikimedia.org/T118413 [19:22:56] T33327: Make default interwiki map links protocol-relative - https://phabricator.wikimedia.org/T33327 [19:22:56] T54253: Protocol-relative URLs are poorly supported or unsupported by a number of HTTP clients - https://phabricator.wikimedia.org/T54253 [19:23:08] I declined the 2nd one yesterday [19:26:38] anomie: Certainly not? Embedding HTTPS on a HTTP page works fine. [19:27:10] "The industry" at large has already made that decision regardless. If you're using anything from Wikimedia, Twitter, Facebook, GitHub, etc. it's HTTPS-enforced regardless. [19:28:10] Krinkle: You asked about making wfExpandUrl do it always, not just on Commons or whatever. [19:28:11] RainbowSprinkles: Yeah, my point exactly. Secure-first, if it works both ways, why not. I don't think scenarios about performance or bad certificates are still valid concerns today. Not anymore. [19:29:23] And they become less relevant with each month. If not us, the rest of the industry is forcing the matter. [19:29:42] anomie: I wasn't suggesting making wfExpandUrl change HTTP to HTTPS. [19:29:47] It's quickly becoming impossible to access large parts of the internet without semi-modern TLS support. [19:29:49] Merely to consistently expand relative to HTTPS [19:29:58] in other words, drop _CANONICAL, _HTTP, _HTTPS, _RELATIVE options [19:30:03] _INTERNAL, .. [19:30:19] (Well, internal perhaps not, but presumably that wold require a different inpout in the first place) [19:31:48] We'll need to figure how we want to use wgInternalServer for Varnish/Squid kind of settings, if that still applies. [20:45:34] I need help people. https://phabricator.wikimedia.org/tag/mw-1.29-release/ [20:46:15] The first two seem pretty serious, but are things I don't really get. [20:46:38] Last one is possibly just a "release anyway, then do followup later" [20:50:15] I'll look at the DeferredUpdates one [20:59:12] found the cause, working on a patch [20:59:26] \o/ [21:04:31] RainbowSprinkles: [14:04:20] (PS1) Legoktm: Always log exceptions in rollbackMasterChangesAndLog() [core] - https://gerrit.wikimedia.org/r/361780 (https://phabricator.wikimedia.org/T168347) [21:08:03] Hmm [21:08:34] Looks good [21:09:25] I stick https://paste.fedoraproject.org/paste/6dPkA-8NZtOb1JtGaZ~spg/raw in my LocalSettings to test it [21:09:27] stuck* [21:10:27] I wonder if we could turn that into a phpunit test [21:10:30] gonna look at the category one now but I need to leave soon [21:10:39] With a @throwsException [21:11:10] the exception never bubbles up, it gets logged [21:11:19] Ah yes, right [21:11:22] So not easy [21:11:27] Ok, we'll merge as is [21:12:06] if there's a way to create a fake logger that just makes the log buffer accessible it would be doable, but you still need to mock all the DB/LB functions [21:12:20] Yeah, I don't think we have such a mocked logger either [21:12:24] (would be generally useful) [21:23:10] RainbowSprinkles: I have to go afk now, but I left a short note on the ticket [21:23:15] (the categorization one) [21:23:16] tyvm [22:07:31] I *think* we have mocked the logger for tests before [22:09:53] RainbowSprinkles: MediaWikiTestCase has a setLogger() method for overloading the logger in tests. [22:09:57] you can see it used in phpunit/includes/session/CookieSessionProviderTest.php