| 02:02.23 | *** join/#asterisk-dev babak (uid19622@gateway/web/irccloud.com/x-wfsqllopqhidnpak) |
| 02:17.49 | *** join/#asterisk-dev superscrat (~asanders@173-17-133-2.client.mchsi.com) |
| 02:31.57 | *** join/#asterisk-dev superscrat (~asanders@173-17-133-2.client.mchsi.com) |
| 05:12.46 | *** join/#asterisk-dev leedm777 (~leedm777@50-81-28-152.client.mchsi.com) |
| 05:27.43 | *** join/#asterisk-dev babak (uid19622@gateway/web/irccloud.com/x-onvwjmlmfzhlclqn) |
| 09:03.12 | *** join/#asterisk-dev doome_ (~doome@78-131-72-29.pool.digikabel.hu) |
| 11:35.42 | *** join/#asterisk-dev superscrat (~asanders@173-17-133-2.client.mchsi.com) |
| 14:44.26 | *** join/#asterisk-dev [TK]D-Fender (~joe@64.235.216.2) |
| 15:14.54 | *** join/#asterisk-dev gtjoseph (~gtj0@unaffiliated/gtj) |
| 16:20.55 | *** join/#asterisk-dev sruffell (~sruffell@asterisk/the-kernel-guy/sruffell) |
| 16:20.55 | *** mode/#asterisk-dev [+o sruffell] by ChanServ |
| 18:37.44 | *** join/#asterisk-dev infobot (ibot@rikers.org) |
| 18:37.44 | *** topic/#asterisk-dev is Asterisk Development Discussion -=- http://www.asterisk.org/developers -=- Tier 2 and 3.14159265 support is in #asterisk -=- Hell hath frozen over: Asterisk is moving to Git -=- Follow on Twitter at @AsteriskDev |
| 19:45.34 | coreyfarrell | gtjoseph: ping |
| 19:45.40 | gtjoseph | pong |
| 19:45.43 | gtjoseph | what's up? |
| 19:46.22 | coreyfarrell | so 1) thanks for helping me with that fix, my full testsuite run overnight got 100 less failures |
| 19:46.42 | gtjoseph | 100 less means there are more I assume. :) |
| 19:47.16 | coreyfarrell | 2) if you're busy no worry, but could I talk you into running a few tests with REF_DEBUG against some tests that are still leaking ast_sorcery_generic_alloc? |
| 19:47.30 | gtjoseph | easy enough. which tests? |
| 19:48.21 | gtjoseph | and if they're related to config_wizard, don't bother. i'm already working on them. |
| 19:49.59 | coreyfarrell | a couple are: tests/channels/pjsip/ami/show_endpoint tests/channels/pjsip/keep_alive tests/pjsip/subscriptions/presence/presence_pidf |
| 19:50.20 | gtjoseph | in progress... |
| 19:50.31 | coreyfarrell | actually the pjsip/keep_alive is something different, but those other two are sorcery objects |
| 19:52.28 | gtjoseph | yep all 3 failed for leaks |
| 19:53.44 | coreyfarrell | awesome, so I guess my system isn't as bad as I thought, I just picked the 3 last tests listed by './contrib/scripts/refleaks-summary -nr' |
| 19:54.47 | gtjoseph | new script? Handy! |
| 19:55.09 | coreyfarrell | it's been there since refdebug support was integrated with the testsuite |
| 19:55.30 | gtjoseph | ah. i'm not paying attention as usual. |
| 19:56.35 | coreyfarrell | or might have been added a week later.. I think I put it on so the bamboo refcheck job could create a file that listed all the tests with leaked object count, otherwise navigating the artifact dir over web sucks. |
| 19:57.18 | gtjoseph | cool. i've got to remember to run it |
| 19:58.32 | coreyfarrell | the test I ran overnight had: 104 tests with reference leaks, 28 failures, 9 core dumps |
| 20:08.18 | gtjoseph | 9 core dumps? hmmm. |
| 20:08.55 | coreyfarrell | uh not master |
| 20:10.03 | coreyfarrell | this is on a branch rewriting the module loader, so 9 core dumps on the full testsuite is a GIANT improvement. a week ago it was 50% of tests core dumped. |
| 20:10.15 | gtjoseph | ha! |
| 20:13.03 | nbjoerg | who beside mjordan is involved with the actual RAII implementation in asterisk? |
| 20:13.35 | coreyfarrell | nbjoerg: you should just ask your question, probably lots of people can help |
| 20:13.48 | nbjoerg | coreyfarrell: I'm not happy at all with the current mess it is. |
| 20:14.30 | nbjoerg | coreyfarrell: so the question is whether patches to get rid of the nested functions for gcc and the need for blocks with clang are acceptabled, even if they come with a small price in terms of redundancy |
| 20:15.27 | coreyfarrell | nbjoerg: it's hard for me to comment on a patch that I can't see :) |
| 20:18.02 | nbjoerg | btw, http://www.asterisk.org/community/developers should be updated for git |
| 20:18.42 | coreyfarrell | I'm not sure how you would get rid of the nested functions / blocks. if this means having a separate function declared for every single RAII variable then I doubt it. |
| 20:19.15 | nbjoerg | coreyfarrell: not per variable |
| 20:19.20 | nbjoerg | sevc |
| 20:19.21 | nbjoerg | sec |
| 20:20.48 | coreyfarrell | nbjoerg: not sure if anyone here today has access to that page, probably mention it to Rusty Newton (his IRC alias is either rnewton or newtonr I can't remember) |
| 20:21.42 | nbjoerg | older version of the changes is http://cvsweb.netbsd.org/bsdweb.cgi/~checkout~/pkgsrc/comms/asterisk/patches/patch-apps_app__skel.c?rev=1.1&only_with_tag=MAIN |
| 20:22.08 | nbjoerg | I would do it a bit cleaner now |
| 20:22.41 | nbjoerg | essentially, for every combination of type and destructor, a RAII_DECL defines the wrapping function in global scope |
| 20:22.56 | nbjoerg | the actual RAII_VAR would remain as is |
| 20:24.34 | nbjoerg | so for 11.17.1, it would mean adding ~40 lines for the declares |
| 20:24.41 | nbjoerg | if they are to remain per-functiono |
| 20:24.53 | gtjoseph | nbjoerg: you should submit a working patch that implements proposal |
| 20:25.37 | nbjoerg | I don't mind doing the manual work, if the reaction is not going to be "we are very attached to what we have right now" |
| 20:25.54 | nbjoerg | the blocks hack made me wonder a lot about that... |
| 20:26.15 | coreyfarrell | nbjoerg: is your goal to get rid of every single nested function / clang block? if so why does this need to happen? what is the perceived benefit(s)? |
| 20:26.29 | coreyfarrell | nbjoerg: not trying to discourage, just trying to understand.. |
| 20:26.30 | gtjoseph | i wouldn't say people are attached to the current implementstion |
| 20:26.35 | nbjoerg | coreyfarrell: we don't ship the blocks runtime for netbsd and I don't think freebsd does either |
| 20:26.45 | nbjoerg | coreyfarrell: nested functions in gcc are a security yissue |
| 20:28.07 | gtjoseph | If you can document the benefits to the community as a whole and not just to netbsd, then i don't see why it wouldn't be considered. |
| 20:30.07 | coreyfarrell | is it possible to have a generic ao2_raii_cleanup function? so instead of generating a new function for every type just have one that takes void* or void**? |
| 20:30.52 | nbjoerg | coreyfarrell: sadly, no |
| 20:31.08 | nbjoerg | coreyfarrell: not without violating strict aliasing rules |
| 20:31.46 | nbjoerg | hm. firefox doesn't like https://git.asterisk.org |
| 20:32.15 | gtjoseph | it's not httpd enabled |
| 20:32.18 | gtjoseph | https |
| 20:32.29 | coreyfarrell | looks like it's not actually ssl, try http |
| 20:32.49 | nbjoerg | right |
| 20:33.00 | nbjoerg | ah well, patch building against github works too |
| 20:33.44 | coreyfarrell | nbjoerg: for making a patch you probably want to do it against gerrit, that'll be where you need to submit for review. |
| 20:34.17 | nbjoerg | one step at a time |
| 20:38.01 | coreyfarrell | nbjoerg: strict aliasing rules are a royal pain in the ass when it comes to 'struct something **' vs 'void **' or 'void *' |
| 20:38.17 | nbjoerg | I know |
| 20:39.09 | nbjoerg | that said, I don't know if gcc or clang actually checks the typing for the cleanup handler. if they don't, I would consider it a separate bug :) |
| 20:39.23 | nbjoerg | so best not to depend on it |
| 20:40.38 | coreyfarrell | I'm still unsure why it would be a problem for 'void ao2_raii_cleanup(void *obj) { void **o = obj; ao2_cleanup(*o); *o = NULL; }' |
| 20:41.45 | nbjoerg | right, that assumes the compiler doesn't check that the argument matches the object type |
| 20:41.49 | coreyfarrell | for that to be a global function used for the cleanup attribute on all ao2 types |
| 20:42.25 | coreyfarrell | all ao2 functions take a void* function because they are generic for many types of structs |
| 20:43.36 | coreyfarrell | that's what I'm saying. take: void test(struct mystruct *obj) { ao2_cleanup(obj); } --- the compiler doesn't complain about passing 'struct mystruct *' to a parameter that takes 'void *' |
| 20:43.57 | nbjoerg | yes, because foo * to void * is a valid implicit cast |
| 20:44.08 | nbjoerg | but foo ** to void ** is not |
| 20:44.34 | coreyfarrell | but foo ** to void * is valid |
| 20:45.35 | nbjoerg | yes, but I also believe that the compiler should warn if "foo *f attribute((cleanup(foo_cleanup)))" does not take a foo ** argument |
| 20:45.56 | nbjoerg | (it seems neither clang nor gcc currently do) |
| 20:47.30 | coreyfarrell | that's my point though, I think the compiler should not warn in that situation, assuming you have a 'void foo_cleanup(void *obj);' |
| 20:48.52 | nbjoerg | it's one my biggest complain about C that it is not possible to correctly write a pointer to a generic pointer |
| 20:49.08 | nbjoerg | (after the memset with null pointer sillyness that glibc now enforces) |
| 20:49.32 | nbjoerg | anway, we'll see how much it matters |
| 21:05.17 | nbjoerg | why is AC_CONFIG_AUX_DIR using `pwd` and preventing out-of-tree build? is that intentional? |
| 21:07.52 | coreyfarrell | nbjoerg: not sure.. Asterisk use of autotools is not very good. if you can make out-of-tree build work I don't see why a patch wouldn't be accepted for master. |
| 21:10.20 | coreyfarrell | nbjoerg: if you are going to work on code you plan to submit, sign the contributor agreement if you haven't already.. that way when you're ready with a patch the system will allow you to submit it. |
| 21:11.35 | nbjoerg | ack |
| 21:13.00 | nbjoerg | coreyfarrell: have a link? it seems to be missing from /developers too :) |
| 21:13.24 | coreyfarrell | link is at the top of issues.asterisk.org |
| 21:13.41 | nbjoerg | ah! |
| 21:15.00 | coreyfarrell | anyhow I'm out for a couple hours, good luck :) |
| 23:07.14 | *** join/#asterisk-dev babak (uid19622@gateway/web/irccloud.com/x-wjbtaparwpvpwflm) |