Camelia, the Perl 6 bug

IRC log for #darcs, 2013-02-19

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:00 sm although, there they look like FilePath (from getCurrentDirectory) but ghc tells me they must be Darcs.Util.Path.AbsolutePath
00:02 gh_ quoting http://irclog.perlgeek.de/​darcs/2010-08-22#i_2733193
00:02 gh_ "what do here and cwd actually mean?"
00:02 gh_ "they are paths.  one is fetched when darcs starts and I think the other is the root of the repo that you're working in (not 100% sure)
00:02 gh_ 07:58I think here is the repo, and cwd is the path that you ran darcs i"
00:03 gh_ n
00:04 sm good memory
00:04 sm I wonder if plugging in "." for both will make it work as before
00:05 sm previously, darcsden was doing withCurrentDirectory "." ...command... so I'm guessing yes
00:06 gh_ I recall that many commands don't use neither of these options
00:13 sm oh oh.. withGutsOf no longer exported
00:13 sm I don't like the sound of this
00:14 sm hey, that's from friday.. getting to recent stuff
00:15 sm http://hub.darcs.net/simon/darcsden/src/​DarcsDen/Handler/Repository/Forks.hs#119 is the code using it
00:20 gh_ whithGutsOf did not do anything (on non-OF repos)
00:20 gh_ so you can directly use the function you passed to withGutsOf now
00:21 sm thanks
00:22 sm all compiling now! just some iconv link errors
00:28 sm I installed ubuntu's libghc-iconv-dev, but I think that's for the system ghc 7.4.1 and I'm using a binary-install ghc 7.4.2
00:28 sm whose stuff is in /usr/local/lib/ghc-7.4.2, I think
00:29 sm which does not include iconv. It looks like I can build darcs without iconv maybe ?
00:31 gh_ sm, install libcurl-dev or something like this
00:31 gh_ it's the lack libcurl that triggers the iconv error
00:31 gh_ *of
00:31 gh_ oh, you said it was compiling... so it's not that
00:32 gh_ with the problem I mention it shouldn't even start to compile
00:32 gh_ it's late here, see you another time
00:32 sm ok.. thanks gh_
00:32 sm good night
00:33 sm https://gist.github.com/4982029 is my specific iconv link error
00:38 sm victory!!
00:39 sm ghc flags that build darcsden/darcs/ssh here: https://gist.github.com/4982061
00:43 sm now http://hub.darcs.net:8901/ rebuilds and restarts when I edit the source
01:07 sm is there a way to forcibly obliterate past tags with one command ?
01:24 markstos_ joined #darcs
01:46 donri joined #darcs
01:47 sm I dunno
01:49 sm when interrupted at http://darcsden.com/alex/ssh/b​rowse/src/SSH/Channel.hs#L-170 , ssh kills both the IO redirect thread and the darcs process, eg darcs apply. terminateProcess ( http://hackage.haskell.org/packages/​archive/process/latest/doc/html/Syst​em-Process.html#v:terminateProcess ) is unreliable but if I wait a second and getProcessExitCode , I see Just ExitSuccess. But ps shows the processes still there untill killed manually
01:50 sm alexsuraci, any ideas ?
02:18 markstos_ joined #darcs
02:44 shlevy left #darcs
03:10 Thaalos joined #darcs
04:47 preflex_ joined #darcs
05:13 `nand` joined #darcs
05:17 mndrix joined #darcs
05:45 mulander joined #darcs
06:22 Heffalump sm: so darcs is apparently being sent SIGTERM but not shutting down cleanly?
06:22 Heffalump or indeed at all
06:23 Heffalump what do you do to kill it manually? Just normal kill (i.e. also SIGTERM)?
06:25 Heffalump sm: I wonder if it's worth also doing an hClose pin like in the EOF case
06:45 lelit joined #darcs
07:06 alexsuraci_ joined #darcs
07:15 Igloo joined #darcs
07:24 Igloo joined #darcs
07:53 Lynic joined #darcs
08:03 Lynic left #darcs
08:18 gal_bolle hi all
08:38 amgarchIn9 joined #darcs
08:42 raichoo joined #darcs
08:49 intripoon joined #darcs
08:56 gh_ joined #darcs
09:13 Igloo joined #darcs
09:46 gh_ will darcs run better on ext4 with linux 3.8? http://kernelnewbies.org/Linux_3.8#head-​372b38979138cf2006bd0114ae97f889f67ef46a
10:30 dixie gh_: the best is to check with performance tests :)
10:31 gh_ yeah, so we need a machine with 2 partitions, one, say, ext3, and the other ext4 (and of course linux 3.8 with the above optimization)
10:31 gh_ and a repo with many small files
10:45 dixie probably lot of work :( and not very needed...
10:47 dixie but maybe it can discover some interesting results
10:51 iago joined #darcs
11:21 jyyou joined #darcs
11:56 kmels joined #darcs
13:23 owst joined #darcs
13:50 mizu_no_oto joined #darcs
14:36 sm g'day all
14:36 owst hey sm
14:36 sm Heffalump: you know what, I think that fixes it
14:37 sm good call :)
14:50 sm alexsuraci: could you review and release http://hub.darcs.net/simon/ssh/changes when you get a chance
15:09 sm aand.. fix deployed I believe.. back later
15:14 mizu_no_oto joined #darcs
15:16 markstos joined #darcs
15:49 byorgey how was the hacking sprint?
15:49 byorgey (I assume it is done by now?)
16:15 lelit` joined #darcs
16:22 whaletechno joined #darcs
16:34 iago Heffalump, is rebase considered "stable enough" ? or there are many open bugs?
16:40 owst iago: there are a few slight UI issues, but it's definitely usable
16:40 owst byorgey: the sprint was good!
16:40 iago owst, are you using it?
16:44 owst I have used it, yes
17:28 byorgey owst: excellent!
17:30 kmels joined #darcs
17:47 stepkut joined #darcs
18:20 raichoo joined #darcs
18:30 iago joined #darcs
18:44 amgarchIn9 joined #darcs
19:01 sm I'm ready to close http://hub.darcs.net/simon/darcsden/issue/46 , maybe some darcs hub users can try interrupting their pushes and confirm that subsequent pushes work fine
19:02 sm Heffalump: we have a $10 bounty to split, how would you like your half ? :)
19:03 Heffalump :-)
19:03 Igloo joined #darcs
19:03 Heffalump you can buy the first round if we ever meet :-)
19:03 Heffalump if $5 can even buy a beer now, let alone then
19:04 sm great, agreed and I'll put it towards hosting
19:09 sm I just don't want to declare victory too early, will leave the issue open until it's had more testing
19:09 Heffalump yeah, I have a feeling you might only have closed down the most common failure mode
19:10 Heffalump and that the thing is inherently unreliable given the chance of the darcs apply process itself getting OOMed
19:14 sm it needs to be reliable, as in it should really truly kill the process
19:14 sm which sounds easy..
19:14 sm normally keeping the alive is the problem
19:17 Heffalump what should really truly kill what?
19:18 sm ssh's channel handling loop should really kill the darcs subprocess when the ssh connection is interrupted, http://hub.darcs.net/simon/ssh​/browse/src/SSH/Channel.hs#178
19:18 sm and with the addition of closing the subprocess's stdin as you suggested, it seems to
19:19 sm hmm, a redundant io there
19:23 Heffalump ok, so is the alternative scenario I posed where darcs apply uses lots of memory and gets killed, a real one that we've had?
19:24 sm killed by the OOM killer ? I can't say it's never happened, but it's not a typical failure mode
19:24 sm but possible
19:25 sm I guess darcs is not able to catch that signal in order to clean up
19:25 sm sigTERM or whatever it is
19:27 Heffalump it's a KILL, so no.
19:27 Heffalump ok, cool. I don't really understand why the handle fix works - I had a vague idea that perhaps if the process was blocked on the handle it couldn't receive a signal - but I don't really understand the semantics of this stuff.
19:28 sm me neither
19:28 sm #haskell or stack overflow might explain it if we took the time
19:29 sm and I still have a cron job that will mail me hourly if lock files are left, or will do once I redirect darcsden email
19:30 markstos left #darcs
19:31 sm 19000 messages, great
19:32 sm I think it's time to start fresh
19:35 Igloo joined #darcs
19:46 schlaftier joined #darcs
20:26 favonia joined #darcs
20:35 mizu_no_oto joined #darcs
20:42 iago which is the most stable repository of Darcs supporting rebase?
20:43 iago (I mean, most stable Darcs branch)
20:45 Heffalump I'd recommend screened. There's also reviewed, but AFAIK that has strictly more bugs.
20:45 Heffalump (albeit of course that by definition, the patches just in screened haven't been reviewed and so are more likely than the average to contain bugs)
20:50 carter joined #darcs
20:53 iago thanks!
21:06 Igloo joined #darcs
21:11 favonia joined #darcs
22:34 favonia joined #darcs
22:40 Igloo joined #darcs
22:46 sm with 2.9.7 (+ 76 patches), darcs check just told me The repository is consistent! ... Hash mismatch(es)! ... Bad index.
22:47 sm coincidentally or not, I see a similar error (referencing different files) in the related repo on the other machine, with darcs 2.5.2
22:48 sm is it talking about patch index ? how would 2.5.2 know about that ?
22:50 sm not patch index then.. some older index
22:51 Igloo joined #darcs
22:54 sm I think it's telling me a (clean) working copy file does not have the same hash as its committed counterpart ?
22:57 sm and the fix was to remove _darcs/index and check again
23:47 stepkut joined #darcs

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary