| 00:25.03 | *** join/#bzflag porco (~PORCOCHKR@host184.190-31-149.telecom.net.ar) |
| 01:28.35 | *** join/#bzflag Spambot (~BZAnathem@50-90-149-18.res.bhn.net) |
| 01:39.06 | *** part/#bzflag afu (~afu@mranderson.cba.ua.edu) |
| 02:48.10 | blast007 | okay, here's what I found. |
| 02:49.04 | blast007 | Ubuntu 12.04 with g++ 4.6.3 and libcurl 7.22.0 = crash |
| 02:49.12 | blast007 | Ubuntu 12.04 with g++ 4.6.3 and our bundled libcurl = crash |
| 02:49.52 | blast007 | Debian Wheezy with g++ 4.7.1 and libcurl 7.26.0 = crash |
| 02:53.52 | blast007 | guess I'll try with older versions of g++ tomorrow night |
| 02:55.11 | blast007 | oh, and the test with our bundled libcurl also used our bundled ares |
| 03:14.10 | *** join/#bzflag L4m3r_Droid (~L4m3r@bzflag/developer/L4m3r) |
| 03:14.10 | *** mode/#bzflag [+v L4m3r_Droid] by ChanServ |
| 03:58.34 | *** join/#bzflag JeffM (~Toybox@unaffiliated/jeffm2501) |
| 03:58.35 | *** mode/#bzflag [+v JeffM] by ChanServ |
| 04:09.24 | *** join/#bzflag dngor_ (abuse@c-98-249-224-255.hsd1.fl.comcast.net) |
| 04:19.01 | *** join/#bzflag Djpenguin (~David@c-98-207-67-102.hsd1.ca.comcast.net) |
| 04:46.28 | *** join/#bzflag noyb (~noyb@pool-74-100-89-192.lsanca.dsl-w.verizon.net) |
| 05:07.51 | *** join/#bzflag mdskpr__ (~mdskpr@190.10.71.41) |
| 05:14.41 | I_Died_Once | is anyone on, able, and willing to approve an image just submitted on the image uploader? |
| 05:16.24 | I_Died_Once | i'd like to put it up and restart tonight if possible, thx |
| 05:25.33 | *** join/#bzflag L4m3r_Droid (~L4m3r@bzflag/developer/L4m3r) |
| 05:25.34 | *** mode/#bzflag [+v L4m3r_Droid] by ChanServ |
| 06:41.37 | *** join/#bzflag L4m3r_Droid (~L4m3r@bzflag/developer/L4m3r) |
| 06:41.37 | *** mode/#bzflag [+v L4m3r_Droid] by ChanServ |
| 06:51.43 | *** join/#bzflag nielsle (~nielsle@3239059-cl69.boa.fiberby.dk) |
| 07:11.22 | *** join/#bzflag allejo (~allejo@cpe-76-90-186-178.socal.res.rr.com) |
| 07:47.58 | *** join/#bzflag alezakos (~kongr45gp@ppp-2-85-40-107.home.otenet.gr) |
| 07:47.59 | *** join/#bzflag alezakos (~kongr45gp@unaffiliated/alezakos) |
| 10:49.33 | *** join/#bzflag bier|tp (~bier@p5085DC09.dip.t-dialin.net) |
| 10:50.17 | *** join/#bzflag bier (~bier@p5085DC09.dip.t-dialin.net) |
| 11:28.18 | *** join/#bzflag alezakos (~kongr45gp@ppp-2-85-19-123.home.otenet.gr) |
| 11:28.18 | *** join/#bzflag alezakos (~kongr45gp@unaffiliated/alezakos) |
| 13:11.56 | *** join/#bzflag I_Died_Once (~I_Died_On@unaffiliated/idiedonce/x-1828535) |
| 13:48.56 | *** join/#bzflag sirquine (~quine@67.176.80.111) |
| 13:53.03 | *** join/#bzflag Marzipan (~Markus@77.117.246.16.wireless.dyn.drei.com) |
| 14:08.22 | Marzipan | looking at the chat logs file size is demotivating :) |
| 14:09.25 | Marzipan | well, guess it just tracks the channel itself and not the private conversations so nvm |
| 15:09.58 | *** join/#bzflag Gnurdux (~gnurdux@99-105-56-72.lightspeed.sntcca.sbcglobal.net) |
| 15:37.59 | *** join/#bzflag Marzipan (~Markus@77.117.246.16.wireless.dyn.drei.com) |
| 15:51.26 | *** join/#bzflag Pimpinella (~frank@gondolin.pimpi.org) |
| 16:13.00 | *** join/#bzflag JeffM (~JeffM@76-218-80-21.lightspeed.irvnca.sbcglobal.net) |
| 16:13.01 | *** join/#bzflag JeffM (~JeffM@unaffiliated/jeffm2501) |
| 16:13.01 | *** mode/#bzflag [+v JeffM] by ChanServ |
| 16:18.38 | blast007 | JeffM: so far doesn't seem like libcurl is the actual problem |
| 16:19.41 | blast007 | crashes with every version of libcurl I've tried so far, including our bundled version |
| 16:20.03 | JeffM | fun |
| 16:20.13 | JeffM | I was able to build a mac version after some fighting last night |
| 16:20.17 | JeffM | I'll post it when I get home |
| 16:20.22 | blast007 | going to try an older g++ to see if that works around the problem |
| 16:20.36 | JeffM | makes me wonder if its' the API que system to curl then |
| 16:20.43 | BulletCatcher | Since it does crash with the bundled curl version you can compile that with debug enabled to get more complete stack traces. |
| 16:21.29 | blast007 | BulletCatcher: the backtrace stuff has mostly been in libc stuff. I haven't actually seen the trace specifically mention curl at all. |
| 16:21.34 | BulletCatcher | Also, turning off optimization (with -O0) should eliminate a lot of differences between compiler versions. |
| 16:21.40 | BulletCatcher | k |
| 16:22.07 | blast007 | it works fine with g++ 4.4.3 |
| 16:22.31 | blast007 | 4.6.3 and 4.7.1 result in a crash |
| 16:23.10 | blast007 | I'll install 4.5 and 4.4 in Ubuntu 12.04 and Debian Wheezy tonight and see how that goes |
| 16:23.36 | JeffM | just redo the server in php |
| 16:23.45 | JeffM | then people can host games on godady |
| 16:24.46 | blast007 | Or we could do it in NodeJS, and then use MongoDB to store all the game state and other information. Then BZFlag will be webscale and can shard. |
| 16:25.14 | BulletCatcher | (buzzword) bingo! |
| 16:25.16 | JeffM | sharding is lame |
| 16:25.46 | JeffM | clouding, much better |
| 16:26.06 | JeffM | with sharding you have to wash all those pants |
| 16:29.20 | blast007 | are clouds webscale? what about /dev/null? I heard that is really fast and is webscale. |
| 16:33.25 | JeffM | it still takes CPU |
| 17:08.03 | *** join/#bzflag Markus__ (~Markus@77.117.246.16.wireless.dyn.drei.com) |
| 18:05.56 | *** join/#bzflag L4m3r (~l4m3r@bzflag/developer/L4m3r) |
| 18:05.56 | *** mode/#bzflag [+v L4m3r] by ChanServ |
| 18:36.23 | *** join/#bzflag ahs3- (~ahs3-@adsl-065-005-193-158.sip.rdu.bellsouth.net) |
| 19:58.21 | *** join/#bzflag alezakos (~kongr45gp@unaffiliated/alezakos) |
| 20:15.19 | blast007 | hmm, it's dying at the end of the deconstructor for URLFetchHandler in the bzfs API |
| 20:16.29 | blast007 | "free(): invalid pointer" |
| 20:25.37 | JeffM | probably my crap then |
| 20:49.47 | *** join/#bzflag Marzipan (~Markus@77.117.246.16.wireless.dyn.drei.com) |
| 21:08.31 | I_Died_Once | I thought it was /diff - whats the command to see which server settings have been changed? |
| 21:13.57 | JeffM | hey blast007 did the callstack show anything past the fetchhandler destructor? |
| 21:14.07 | JeffM | like KillCurrentJob, or removeAllJobs? |
| 21:19.14 | *** join/#bzflag Markus__ (~Markus@178.115.248.11.wireless.dyn.drei.com) |
| 21:44.00 | blast007 | nope |
| 21:44.48 | blast007 | JeffM: http://pastebin.com/1biPu75q |
| 21:45.32 | JeffM | oh hmm, it's called from world event manager, the destructor calls world event manager |
| 21:45.34 | JeffM | that's probably why |
| 21:48.43 | JeffM | blast007, take out the call to worldEventManager.removeEvent(bz_eTickEvent,this); on line 2949 in bzfsAPI.cxx and see if it dosn't crash |
| 21:48.52 | blast007 | k |
| 21:48.59 | JeffM | the fetcher is static, so it dosn't ever need to remove itself |
| 21:49.11 | JeffM | I think the distructors are fighting |
| 21:56.49 | blast007 | bleh. I'm getting different backtraces from two different builds on the same system.. |
| 21:57.09 | blast007 | one of which is a slightly modified build, but nothing that I'd think would affect this |
| 21:57.56 | blast007 | the modified server makes mention of URLFetchHandler, and the other one does not |
| 21:58.39 | blast007 | but otherwise identical, short of the pointer values of course |
| 21:59.03 | JeffM | fun |
| 21:59.18 | JeffM | I wonder if curl is dead allready |
| 21:59.34 | JeffM | if curl dies before the static goes out of scope we don't have to clean up anything |
| 22:00.16 | blast007 | the server where the backtrace does not mention URLFetchHandler still dies even when I comment out the line you mentioned above |
| 22:00.57 | JeffM | comment out the entire guts of the desctructor |
| 22:01.00 | JeffM | what the hell :) |
| 22:01.06 | JeffM | shotgun debuging |
| 22:01.53 | blast007 | of URLFetchHandler? |
| 22:01.58 | JeffM | ja |
| 22:02.00 | blast007 | k |
| 22:02.18 | JeffM | your quitting anyway it won't leak |
| 22:02.31 | blast007 | yeah, just kinda 'eww' :) |
| 22:02.53 | JeffM | this is mostly to see if it's even involved |
| 22:03.30 | JeffM | the thing to do would be to make the handler be a pointer not a static and remove it before the rest of the system is killed |
| 22:03.41 | JeffM | then it won't try to shut down things that are gone |
| 22:04.01 | JeffM | actualy, no I think I know what's up |
| 22:04.12 | JeffM | the world event handler is trying to delete a static |
| 22:04.30 | blast007 | the backtrace is a bit different now |
| 22:05.12 | JeffM | yeah thats why we had the "autodelete" thing |
| 22:05.29 | blast007 | http://pastebin.com/zY0T8PCb |
| 22:08.54 | JeffM | yeah I'm making a change now |
| 22:10.16 | blast007 | alright |
| 22:12.02 | CIA-106 | BZFlag: 03JeffM2501 * r22496 10/trunk/bzflag/src/bzfs/bzfsAPI.cxx: allocate the url hander for the API so the event system can delete it |
| 22:12.58 | JeffM | try that |
| 22:15.10 | JeffM | now that I see that the "invalid pointer" makes sense, you can't delete something that was not newed |
| 22:17.17 | blast007 | different crash now :) |
| 22:17.27 | JeffM | well that's something |
| 22:17.41 | blast007 | http://pastebin.com/wjAt2iJZ |
| 22:18.19 | JeffM | ok that one, I got nothin |
| 22:18.24 | JeffM | probalby need to valgrind it |
| 22:18.52 | blast007 | is there anything useful I could print out from that function? |
| 22:21.06 | JeffM | checking something |
| 22:22.31 | CIA-106 | BZFlag: 03JeffM2501 * r22497 10/trunk/bzflag/src/bzfs/bzfsAPI.cxx: call the default constructor for the API url hander so we can init it as NOT being in a plugin. |
| 22:22.37 | JeffM | blast007, try that |
| 22:23.36 | blast007 | different crash ;) |
| 22:23.42 | JeffM | whoot |
| 22:23.53 | JeffM | gotta catch em all |
| 22:24.04 | blast007 | http://pastebin.com/q4eSwp0y |
| 22:26.51 | JeffM | change line 57 in WorldEventManager.cxx to if (!theEvent || eventList.empty()) |
| 22:27.26 | blast007 | and this is a valgrind from r22497 without that change http://64.31.37.34/valgrind22497.log |
| 22:28.24 | JeffM | ewww |
| 22:29.21 | blast007 | and here's a backtrace with that change: http://pastebin.com/hEKHckAF |
| 22:50.17 | JeffM | meh just comment out the destructor for worldEventManager |
| 22:50.22 | JeffM | it's a static, let em die |
| 22:51.30 | JeffM | all it realy needs is a cleanup method, it can't be deleted by anyone and there is only ever one, it should be a singleton |
| 23:01.37 | blast007 | well, that does solve the crash, of course :) |
| 23:02.56 | JeffM | well that tells me what was happening |
| 23:03.09 | JeffM | and you get no crash if we never call a URL handler? |
| 23:03.18 | JeffM | with no code mods |
| 23:04.07 | JeffM | maybe the API needs a fake plugin to hang it's events onto |
| 23:04.20 | JeffM | I shall ponder |
| 23:14.40 | blast007 | I had several other plugins loaded that didn't call a URL, and those didn't crash |
| 23:15.20 | blast007 | and the plugin I had noticed it with initially didn't call a URL until a player joined, and with that, the server would not crash unless a player joined |
| 23:17.40 | JeffM | yeah, it's that static handler |
| 23:17.54 | JeffM | the event manager assumes that all events are tied to a plugin |
| 23:18.04 | JeffM | and the URL manager isn't |
| 23:18.26 | JeffM | so maybe I'll make an "API" plugin that is always in scope that we can hang stuff on |
| 23:18.39 | JeffM | or not tie into the tick event |
| 23:19.27 | JeffM | I'll lookinto it when I get home |
| 23:19.45 | JeffM | has anyone looked into why distcheck won't work? or is that a non-issue? |
| 23:36.57 | *** join/#bzflag short_circuit (~james@c-98-195-162-191.hsd1.tx.comcast.net) |
| 23:44.00 | blast007 | JeffM: seems like it might work if a system curl dev package is not installed |
| 23:48.12 | blast007 | or maybe not.. |
| 23:58.39 | JeffM | I don't think it has to do with curl, it has to do with the API event handler that uses curl |
| 23:59.46 | *** join/#bzflag sirquine (~quine@c-67-176-80-111.hsd1.co.comcast.net) |