[07:30:05] hello [07:30:46] hello [08:09:40] New patchset: Hashar; "WLMMobile nightly builds generation" [integration/jenkins] (master) - https://gerrit.wikimedia.org/r/16949 [08:10:02] Change merged: Hashar; [integration/jenkins] (master) - https://gerrit.wikimedia.org/r/16949 [15:11:17] chrismcmahon: are you around? [15:19:38] hi matanya [15:19:50] hi,I'm ready when you are [19:21:43] jdlrobson: https://gerrit.wikimedia.org/r/gitweb?p=mediawiki/core.git;a=blob;f=resources/jquery/jquery.localize.js;h=27cae292bf69a88b92e1d444d7f7784a789c090f;hb=HEAD [19:24:08] thanks TrevorParscal [20:11:30] Reedy, hi. You cleaned up whitespace in E3Experiments. a) did you use a tool or vim's retab? b) does a .vimrc line like 'autocmd Filetype php setlocal ts=4 sts=4 sw=4' look right? [20:11:56] I used my IDE for most of it [20:12:02] Some of it had to be manually realigned etc [20:13:30] spagewmf: Filetype php leaves out JavaScript [20:13:41] Reedy yeah, IDE= vim ;-) , I'm curious which IDE do you use? That setting seems to work in vim, I'll add it to Manual:Coding_conventions and see if it sticks. [20:13:44] And CSS, and HTML if you hit them [20:14:35] PhpStorm [20:14:41] I've got a license for wmf/mw usage [20:14:46] ^sharable [20:15:54] marktraceur Sure, I'll add an "etc." Surely people aren't using 4-space tabs for HTML, you'll end up 32 characters over by your first

tag 8-) [20:16:26] spagewmf: ts should be 8 [20:16:41] spagewmf: And there should only be 4 characters, since they're tab characters, not spaces [20:17:43] spagewmf: sts shouldn't be set, because that inputs spaces [20:18:07] Same with sw, methinks [20:18:22] Ah, no, that should be 8 [20:18:36] So ts=8 sw=8 oughta do it [20:18:45] Oh, I'm sorry [20:18:49] You're right to have 4 [20:18:58] I forgot the code conventions :( [20:19:16] ts=4 sw=4 [20:19:35] * marktraceur final answer [20:21:18] marktraceur yup thanks. https://www.mediawiki.org/wiki/Manual:Coding_conventions#Tab_size [20:25:06] marktraceur, I *think* if no expandtab and softtabstop is the same as tabstop it's the same as not setting sts and you'll get appropriate tabs; but yeah leave it out [20:26:36] spagewmf: I don't know; if you were the source of the whitespace troubles then clearly something was up, else maybe it was harmless? Have we git blame'd to a certainty? [20:29:12] marktraceur wasn't me, I'm just trying to get it right [20:29:32] :) mmmkay [21:07:18] kaldari: so it *sort of* works [21:08:01] I think the problem is the absolute positioning [21:09:00] yeah, theeere we go [21:11:04] yeah, it has to be in an element with relative positioning [21:13:40] Still fairly ugly in Chrome [21:13:43] but could be worse [21:13:45] and we can fix it later [21:17:05] kaldari: here we go… [21:17:15] https://gerrit.wikimedia.org/r/17008 [21:21:03] RoanKattouw: Is something still pending here? https://gerrit.wikimedia.org/r/#/c/16718/ [21:23:57] werdna: testing it... [21:26:59] Krinkle: As far as I'm concerned there's not [21:27:07] Wanna merge it? [21:27:15] Well I haven't actually reviewed it [21:27:19] I've just read Nike's comments [21:27:20] Oh :P [21:27:31] hahaha [21:27:45] I'll look at it tomorrow in my 20% time [21:27:52] k [21:27:56] Trying to implement annotation matching with 8 priority ranks right now :) [21:34:33] kaldari: done [21:34:43] re-reviewing... [21:40:49] werdna: found one more problem [21:41:06] yeah, so the parsing benchmark was about 15% slower with LLVM than without it [21:41:52] that's with everything precompiled to machine code [21:42:08] kaldari: fixed [21:42:14] no caches or anything, just everything loaded into the process [21:42:52] werdna: merged [21:43:00] kaldari: if you have copious spare time (hahahaha), feel free to start having a look at Echo in general :) [21:43:08] kaldari: sweeeeet thanks [21:43:18] heh, maybe next month [21:43:25] oh wait, that's like tomorrow [21:43:29] well wednesday [21:43:34] no thursday [21:44:48] preilly: ^^ [21:47:50] werdna: don't forget to create documentation for the new jQuery extension somewhere [21:48:12] TimStarling: interesting [21:48:36] I don't know why, but I have a few theories [21:48:53] one possibility is hardware instruction cache exhaustion [21:49:05] since the code size is probably doubled [21:49:16] werdna: https://www.mediawiki.org/w/index.php?title=JQuery.badge&action=edit&redlink=1 maybe [21:49:55] there's a small amount of extra overhead for function calls, but a tight loop of function calls wasn't slow enough in LLVM to explain the whole performance difference [21:49:58] werdna: or maybe in https://www.mediawiki.org/wiki/ResourceLoader/Default_modules#jQuery_.26_plugins [21:52:18] and stick {{MW 1.20}} on it [21:52:41] TimStarling: yeah I'm not sure why that's happening [21:53:27] there's a slight performance reduction for any C code moved from gcc to clang, due to less effective optimisation [21:53:34] but again, not enough to explain it [21:54:34] but it became clear that I'm optimising a part of the code that's already fast, rather than optimising a slow part [21:56:05] the main VM loop in PHP's execute() is only 5 instructions in the most common case [21:56:29] with one branch prediction error [21:57:18] kaldari: done :) [21:57:23] thanks! [21:57:41] branch prediction freaks me out. [21:58:21] compare that to the ZEND_JMPZ_* handler, which stood out in profiling [21:58:54] which has a switch(Z_TYPE_P(op)) which is converted by both gcc and clang to a binary tree of conditional jumps [21:59:34] so every time you change the variable type you feed into if(), you get a few branch prediction errors [22:02:14] zend_do_fcall_common_helper_SPEC() and zend_parse_parameters() also showed up as hotspots [22:08:00] bbl [22:08:27] Tim-away: okay let's talk later about it [23:04:45] pretty KCacheGrind graph: http://tstarling.com/stuff/PPFrame_expand_callgrind.png [23:05:12] coloured by source file [23:05:43] using phpllvm gives you an interesting view of MediaWiki performance, since you get both PHP and C functions in the call stack [23:18:31] TimStarling: nice graph [23:26:27] preilly: do you know what happened to zend_fast_cache.h? it looked like a useful thing [23:34:54] TimStarling: D php/php-src/trunk/Zend/zend_fast_cache.h was deleted by http://svn.php.net/viewvc?view=revision&revision=303402 [23:35:04] yeah, but it was disabled long before then [23:35:08] http://svn.php.net/viewvc?view=revision&revision=86644 [23:35:11] TimStarling: pajoye deleted it [23:36:10] TimStarling: I think it was just disabled for performance reasons [23:39:41] ^demon: so I am officially fed up with ant [23:39:51] ^demon: I will most probably rewrite the Jenkins script using rake :-D [23:41:37] hashar, I wanted to annoy you with something... [23:41:42] I don't remember which :/ [23:41:50] well 2am there [23:41:52] New patchset: Hashar; "ant: let phpunit run on extensions" [integration/jenkins] (master) - https://gerrit.wikimedia.org/r/17025 [23:41:54] here too [23:41:57] so I am just heading to bed sorry :-( [23:41:58] perhaps reviewing the several changes on gerrit [23:42:10] my little daughter wake me up at 7am every morning :-D [23:42:18] how is she going? [23:42:20] well this week I am fully staffed [23:42:25] and will be on vacations friday [23:42:32] so unlikely to review anything :/ [23:43:01] Change merged: Hashar; [integration/jenkins] (master) - https://gerrit.wikimedia.org/r/17025 [23:43:20] Platonides: she is fine apart from teeth pain :-) [23:43:43] in 5 hours, she will be 1 year old! [23:49:49] oh, I thought it was a bit layer [23:49:53] *later [23:50:19] I had in mind to keep the date, though [23:50:22] congratulations! [23:56:04] New patchset: Hashar; "ant: update the shared git repositories" [integration/jenkins] (master) - https://gerrit.wikimedia.org/r/17027