[11:19:13] I feel like something might be weird about the most recent PHP releases or the SURY repo specifically: I get 'Can not upgrade from versions older than 1.39, please upgrade to that version or later first.' when running update script (1.46.2 or 1.46.3, does not matter) and the table that triggers this definitely exists… [11:19:23] anyone else encountering this? [11:22:32] and I think the patch is missing parsoid 0.23.3 update while referring to it in composer.lock [11:22:43] (I mean the .patch file) [13:41:21] Remilia: Unfortunately it's basically impossible to stage composer/packagist updates without making the security patches public [13:41:39] So it's a weird state, but the changes are there, just not the lock file etc [13:41:58] By the time most people ran composer update in MW core the tag was there, so.. [13:46:33] Reedy: makes sense, though I really need to find a way to get around the issue of running composer update leading to further patches not applying cleanly haha [13:47:04] Yeah.. It's a bit of a messy thing [13:47:18] Running composer in the extensions/skins can work around that, while also opening other problems [13:49:32] I guess I'll just do the archive extraction thing every time from now on, since I rearranged things to only need one file inside the MW directory [16:17:10] Reedy: oh, I get it, it's like even if you do the full upgrade or clean install you cannot get around having to run composer because of this huh [16:18:23] having the new version of parsoid in composer.json meant that if someone did run composer it didn't just blow the security patches away (after the parsoid tags had been pushed) [16:18:35] o> [16:19:21] I can see some users getting lost here because not every host has composer [16:19:49] yeah... I'll probably do a followup release to tidy it up [16:19:59] But it took long enough to get ~200 commits through gerrit [16:20:13] haha [16:20:26] I went to bed just after 0500 local time [16:20:28] I wonder why most such tools are so unfriendly [16:22:30] * Remilia is about to see if she'll get the same 'your 1.46.2 install is too old, you need at least 1.39 to upgrade to 1.46.3' error on the postgresql host as she got on MariaDB [16:24:01] I can't shake off the feeling that something is broken in php [16:42:19] hmm this is very intriguing: I get the error for one wiki in the farm but not the other, and while this check is based on a shared table, the table in question exists in both databases [16:43:11] which table is it complaining about? [16:43:22] it is not complaining, it is the compatibility error [16:44:01] Lucas_WMDE: if ( !$updater->tableExists( 'user_autocreate_serial' ) ) { $this->fatalError("Can not upgrade from versions older than 1.39, please upgrade to that version or later first.");} [16:44:18] I get this error message [16:44:37] it clearly can connect to the DB server or it would complain earlier... [16:45:09] Mine is a fresh 1.46.2 install and I have that table. (sqlite3) [16:45:34] greycat: I also have that table in both databases, or I would not be so confused [16:45:50] maybe I should try placing the old .php files back (I am using YAML right now) and see if that helps [16:47:26] the wiki that has the SharedDB as its own has no issues, the one that uses SharedTables from that SharedDB does have them... and I have a vague suspicion now, thinking out loud does help [16:51:13] I wonder if the updater isn’t using the shared DB [16:52:09] I’m guessing when that line of code was written, it just needed some easy way to distinguish between “old” and “new” databases, and user_autocreate_serial happened to have been added at the right time; the possibility that it might be a shared table might not have been taken into account [16:52:13] Lucas_WMDE: no, this is my fault, I did not grant rights to user_autocreate_serial (this isn't my setup, I was asked to upgrade 1.35.3 a month or so ago, so I am not very well versed in the install's complications but they use different DB users for different wikis) [16:52:22] ah :D then that makes sense ^^ [16:53:03] it was hell to be honest because this thing comes with some massive technical debt in the form of a million enabled but unused extensions [16:53:20] hello. how can i reset my mediawiki? [16:53:54] Paddster: if you just want to start from complete zero in a different world, you can wipe the directory and drop the database? [16:54:30] no, i wrote reset cause i meant reset instead of a restart [16:54:39] reset what exactly? [16:54:45] the created pages [16:55:19] Maybe a better question would be, what do you want to keep? [16:55:53] the entire configuration is fine [16:56:58] what do you include under that term? LocalSettings.php? users? groups? [16:57:20] i have the feeling i am too stupid for this [16:57:25] LocalSettings.php and the installed extensions, perhaps [17:00:17] Paddster: it is more that you need to define what you consider configuration in your case; if you want to keep your LocalSettings but entirely wipe users, page and upload history, you'd still want to start from scratch, drop all DB tables, and just use the same parameters as before, as that will give you a clean slate [17:00:38] then you can put LocalSettings.php back with your extra customisations [17:00:51] i wish you all happy times [17:00:52] MW is not designed to be reset/wiped [17:01:13] huh [17:29:42] o_O [17:36:00] (I think that if someone is using weechat they are probably not too stupid) [18:14:40] Trying to do "php8.4 run.php update" as the UPGRADE file says, I get: Error: your composer.lock file is not up to date. Run "composer update --no-dev" to install newer dependencies [18:14:54] How do I do that? [18:26:02] greycat: do you have composer installed? if not, just get the phar from their website [18:26:18] then `php composer.phar update --no-dev` [18:26:38] or if it is installed system-wide, just `composer ...` [18:27:12] https://getcomposer.org/download/ [18:27:31] I just learned that it's a separate package, and I've installed it (maybe the upgrade instructions could say more about this). Now it's warning me not to run it as root. [18:27:52] well yes you shouldn't run update.php as root either [18:28:14] in fact, just don't run anything PHP as root ever [18:29:36] I feel like I'm missing a *lot* of steps, then. Who's supposed to own the files in the mediawiki directory, then? [18:31:29] www-data owns LocalSettings.php and the images directory, but it doesn't own the rest [18:40:00] Well then. Guess I'm on my own. I created an "mw" user account, and made it the owner of all the mediawiki files other than the ones owned by www-data. Then I ran "composer update --no-dev" as mw. Got an error because it wanted to create a new file, so I also made mw the owner of the directory. Then did it again, and it seems to have worked. [18:40:25] greycat: sorry I was afk [18:41:12] for persimmons with mediawiki, you generally do not want the www user to write anywhere except cache and uploads [18:42:18] there are some exceptions (some extensions have their own cache directories) [18:43:35] OK, seems I'm on the right track then. Finally, I ran the maintenance/update.php script as www-data (who owns the sqlite DB files). [18:43:43] so you could chgrp cache and images to www-data (so it's mw:www-data) and make those g+w [18:44:26] yep, something like that [18:44:40] * Mahiro has never used SQLite with MW, so she has no clue about that one [18:44:50] PostgreSQL here [18:45:22] The cache subdirectory was empty, but I made it 775 mw:www-data as you suggested. [18:45:23] (I'm that annoying person who used to file Pg bugs on the regular) [18:46:10] yes it's localisation cache; I think the file storage mode is not enabled by default but is recommended for performance [18:46:39] greycat: https://www.mediawiki.org/wiki/Manual:Performance_tuning#Localisation_cache [18:46:42] there's a mywiki_l10n_cache.sqlite file as well [18:46:50] yes, it's slower [18:47:34] but does not matter if you run SQLite tbh [18:48:15] what I mean is your deployment is likely small-scale so you do not need to bother with all that optimisation stuff [18:57:25] greycat: by the way where do you keep the .sqlite files? I hope they aren't anywhere under the document root haha [18:57:52] saw this a few times with customers' deployments back when I worked part-time for a data-centre [18:58:28] They're not. I have the sqlite files in /var/www/data/ and the document root is /var/www/domainname/w/ [18:59:28] that is probably fine then [19:00:44] * Mahiro tends to follow the filesystem hierarchies so would never put anything that shouldn't be served under /var/www or similar haha [19:09:12] Debian's default document root is /var/www/html/ so /var/www/data/ and other siblings are safe.