Minetest logo

IRC log for #minetest-dev, 2013-11-14

| Channels | #minetest-dev index | Today | | Google Search | Plaintext

All times shown according to UTC.

Time Nick Message
00:04 OWNSyouAll joined #minetest-dev
00:22 bcnjr5 left #minetest-dev
00:27 VanessaE now there's someone I haven't seen in a while.  Too bad he picked the wrong channel.
00:43 sapier Ok I think I got VanessaE's crash. If a ResultQueue job times out those functions just catch the timeout exception and leave. This way the RequestQueue itself is deleted
00:43 sapier result queue of course
00:44 sapier yet the Request is still queued. within that request a now invalid pointer to the original request queue is stored
00:44 sapier once this Request is processed memory is randomly messed up
00:44 VanessaE ok, here goes nuthin'
00:44 VanessaE nope.avi
00:44 sapier https://github.com/sapier/minetest/tree/fix_RequestQueue
00:44 VanessaE crashola.
00:45 sapier where does it crash for you vanessa?
00:45 VanessaE looks like it's in the same place.  lemme get a backtrace
00:46 VanessaE http://pastebin.ubuntu.com/6413551/
00:48 sapier hmm ok there's still one possible reason left I was concerned
00:48 sapier if a job is processed after I did clear the queue that one will be in front of queue
00:49 sapier if I'm right that's easy to fix
00:50 VanessaE ok
00:52 sapier ok try again
00:54 VanessaE trying...
00:54 VanessaE sapier: success!
00:54 sapier I fixed only one occurance so this might now happen at one of the other 2 locations ;-)
00:54 VanessaE that did the trick
00:54 sapier yes but it's ugly
00:55 sapier I need to think about it tomorrow it's 2 o'clock in the morning now ;-)
00:55 VanessaE ugly it may be, but I'll bet it significantly narrowed the search for the root cause
00:55 sapier no this IS the root cause
00:56 sapier the whole concept with result queues beeing local objects passed to another thread is ugly
00:56 RealBadAngel joined #minetest-dev
00:57 sapier I practicaly made them persistent by adding the "static" at cost of possible outdated jobs queued within them
00:58 VanessaE oh ok
00:58 VanessaE in that case, you FOUND the root cause, and now fixing it properly will be 10x easier
00:59 VanessaE (to my ming the hardest part of debugging is identifying the cause.  coding a fix is usually trivial once you get to that point)
00:59 VanessaE mind*
00:59 sapier that one was tricky true ... yet I always had that queue handling in mind
01:00 sapier something seemed to be wrong there I just didn't know what
01:01 ShadowNinja sapier: #1003 has a bunch of extra newlines at the end of the files. And should that be in jthread?
01:02 sapier I think so yes .. I'll fix that newlines later I need to go to bed now
01:03 VanessaE WOW
01:03 VanessaE sapier:
01:03 sapier vanessae I just updated that branch to fix the other two possible crashes too ... but that one needs cleanup ... new eclipse did a lot of automatic whitespace changes to those files
01:03 VanessaE your patch not only stops the crash but it also speeds up the texture rendering process somehow.  or so it seems
01:04 VanessaE oh sure
01:04 VanessaE lemme pull that real quick
01:04 RealBadAngel sapier. hold on a sec
01:04 sapier I don't have an idea how that should speedup texture processing but as that branch is a RequestQueue and ResultQueue cleanup it might be possible
01:04 sapier yes rba?
01:05 VanessaE nope, already up to date, so it says.
01:05 RealBadAngel im uploading now texture pack you liked, wanna link?
01:05 VanessaE sapier: hang around for a few more mins please.
01:05 sapier sorry forgot to push did only commit now it should be there
01:06 VanessaE oh ok, sec.
01:06 sapier rba yes but I can't test it right now
01:06 sapier ok vanessa 5 min
01:06 sapier I have to work today ;-)
01:06 VanessaE sapier: ok, got the update.  testing it.
01:06 RealBadAngel upload will be ready in circa 3minutes
01:07 VanessaE sapier: ok, no crash and no immediate change when connecting to the previously-offending server.  good.
01:08 VanessaE sapier: anything I should check in particular?
01:08 sapier no the queue fixes fix things that have been that wrong they never could've been used
01:08 sapier fix fixes :-)
01:09 VanessaE ok
01:09 sapier this issue might have been cause for various strange crashes
01:10 VanessaE sapier: testing stuff, one or two more mins.
01:11 VanessaE wow
01:11 RealBadAngel sapier, https://forum.minetest.net/viewtopic.php?pid=117843#p117843 ,with some new screenshots
01:11 VanessaE Survival is several seconds faster to log in, and opening the inventory immediately on connect no longer crashes the cliebnt.
01:11 sapier btw vanessa you should see in infostream "Waiting for shader timed out."
01:11 VanessaE sapier: you may have just fixed #831
01:12 RealBadAngel sapier, have you disabled shaders when creating inv items?
01:12 sapier lol :-) guess this might fix some other "unconfirmed" bugs too
01:12 VanessaE ok here comes the acid test...
01:13 VanessaE sapier: seems good.
01:13 sapier no I just fixed the queue handling but as timeout now no longer currupts random memory locations this might result in missing shader instead of crash
01:13 RealBadAngel that's my issue with shaders. inv items are lit and shadowed with weird angles
01:14 VanessaE better a missing shader than a seemingy-random crash
01:14 RealBadAngel so shaders *should* be disabled there
01:14 VanessaE RealBadAngel: I've noticed this.
01:14 sapier do you use my branch rba?
01:14 RealBadAngel no, thats issue of my branch
01:15 sapier then no I didn't disable shaders ;-)
01:15 VanessaE sapier: doesn't help the slowness of the extrude code in the extreme cases, this is where kahrl's disable-extrude patch is needed.
01:15 VanessaE anyways, it's good.  go to bed :)
01:15 RealBadAngel but you could if youre already here
01:15 sapier thx for permission ;-P
01:15 sapier good night
01:15 sapier left #minetest-dev
01:15 RealBadAngel cya
01:47 Miner_48er joined #minetest-dev
02:12 zork_ joined #minetest-dev
02:29 VanessaE celeron55: your entity dupe patch (#1000)... is it safe to use?
02:39 paramat joined #minetest-dev
02:43 paramat left #minetest-dev
02:46 VanessaE well anyway, it's been deployed to all my servers.  /clearobjects is in progress on all 5 as I speak.
02:47 VanessaE (and it's on my client)
06:09 darkrose joined #minetest-dev
07:00 NakedFury joined #minetest-dev
07:17 darkrose joined #minetest-dev
07:32 KingsleyT joined #minetest-dev
07:53 OldCoder joined #minetest-dev
08:32 Weedy_lappy joined #minetest-dev
08:40 SmugLeaf joined #minetest-dev
08:44 RealBadAngel joined #minetest-dev
08:44 e1z0 joined #minetest-dev
08:44 celeron55 joined #minetest-dev
08:45 hax404_ joined #minetest-dev
08:46 OldCoder joined #minetest-dev
09:35 darkrose joined #minetest-dev
10:14 proller joined #minetest-dev
10:15 ImQ009 joined #minetest-dev
11:05 proller https://github.com/proller/minetest/compare/indev  - is any objections ?
11:18 sapier joined #minetest-dev
11:25 sapier Proller no idea what you change there but I assume you know what you're doing
11:26 proller https://github.com/minetest/minetest/issues/1005
11:34 sapier More interesting structures sounds great but I can't test atm
11:52 sapier Btw my fix for vanessae's crash isn't 100 % correct. It still may assert if two timouts follow directly while first request is processed in between. It's a invalid assert only that case is perfectly valid.
11:55 Akien joined #minetest-dev
12:43 pitriss Hi guys, I want to ask.. can default tree growing mechanism adjust growing speed based on lenght of day/night cycle? Or this is just constant interval?
13:01 proller https://github.com/proller/minetest/commit/8427d78c5a49ad633875fdb61ffe6f5e0d70fd2c
13:03 proller pitriss, i have idea for future: step by step growing, with adjustable speed
13:04 pitriss proller: ah.. I got that idea when on our server was farming plus growing too fast.. And having this also for trees would be really cool..
13:10 EvergreenTree joined #minetest-dev
13:16 kahrl joined #minetest-dev
13:24 hmmmm joined #minetest-dev
14:09 sapier Proller growing trees mod already has step by step growing trees ;-)
14:10 proller wow, need to try
14:41 Zeitgeist_ joined #minetest-dev
15:12 zat joined #minetest-dev
15:27 iqualfragile joined #minetest-dev
15:27 EvergreenTree joined #minetest-dev
15:33 proller joined #minetest-dev
15:52 BlockMen joined #minetest-dev
15:53 damiel joined #minetest-dev
15:53 BlockMen should i make a pull for that or would someone agree that i can push directly? https://github.com/BlockMen/minetest/commit/e67dea94a2d8a5810bdfd46d8d2baa7c2853fee9
16:00 OldCoder joined #minetest-dev
16:02 sapier1 joined #minetest-dev
16:07 NakedFury joined #minetest-dev
16:08 Jordach joined #minetest-dev
16:35 PilzAdam joined #minetest-dev
16:47 sapier joined #minetest-dev
16:55 sapier proller growing trees is quite old
16:55 sapier pre ltrees
16:59 Calinou joined #minetest-dev
17:13 celeron55 sapier: i was looking at your patch this morning and it seemed like you removed the timeout functionality completely
17:14 celeron55 and that whole got_invalid_name thing is quite dubious
17:15 celeron55 you should at least document it properly, it doesn't make any sense
17:17 sapier yes the got invalid_name thing is to be removed it's useless there still is a situation where the assert fails for no reason
17:17 sapier actually the timeout was source of error
17:18 sapier but I didn't quite remove it, the requesting thread will still continue after timeout
17:19 sapier and the processing thread will process it too (as it did before)
17:19 sapier but now the result will be discarded on next request ... before this patch it was just written to random memory once it was processed
17:21 sapier the timeout never did abort the request it just resulted in noone waiting for result and therefore invalid result queue pointer
17:33 Akien joined #minetest-dev
17:38 djdduty joined #minetest-dev
17:40 rubenwardy joined #minetest-dev
17:43 bas080 joined #minetest-dev
17:46 sapier a generic question are spaces at lineend intended in minetest or can I fix them?
17:48 sapier celeron55 this is the actual fix https://github.com/sapier/minetest/commit/fcf807b4306767776bac9a2528abb1cb571c9331 (and some whitespace fixes done by eclipse ... I'll remove them if they aren't ok)
17:54 BlockMen sapier, are you ok with https://github.com/BlockMen/minetest/commit/e67dea94a2d8a5810bdfd46d8d2baa7c2853fee9 ? it fixes a mistake i made with background clipping in formspecs
17:54 sapier seems to be correct yes ;-)
17:56 sapier VanessaE could you check the new version of my patch, it should fix a error in the version you're using atm
17:58 BlockMen ok, i gonna push it then
18:17 Miner_48er joined #minetest-dev
18:20 VanessaE sapier: sufficient to just pull from your branch?
18:24 sapier yes I force pushed the changes
18:24 VanessaE k
18:25 sapier celeron I wonder if JEvent should be moved to it's own file in fact it's 95% identical to my jsemaphore implementation with one difference, event is a non counting semaphore
18:26 damiel joined #minetest-dev
18:26 damiel joined #minetest-dev
18:28 ImQ009 joined #minetest-dev
18:28 VanessaE sapier: ok, got it.  got a test case for the thing you fixed?
18:29 sapier you now will see a error message if something times out
18:29 VanessaE oh by the way:
18:30 VanessaE can someone please fix the debug-build's unit tests to try ports other than 30000?
18:30 VanessaE (else it fails to bind, since I have a server on that port, and then aborts)
18:31 sapier I don't see any use for it unittests do fail anyway if run in debugger
18:31 VanessaE I'm not using the debugger in this case :)
18:31 VanessaE merely a debug-enabled build
18:32 sapier you shouldn't even use debug builds on production server ;-)
18:32 VanessaE this ain't a server :)
18:32 VanessaE this is the client; I have a regular production server already on that port.
18:35 sapier ok ok if I have nothing to do and a good idea how to fix it correct ... no use to use port 30001 for unit test as it'd just fail in a different case
18:35 VanessaE right, I figure it should count upwards from 30000 until it finds a free port.
18:35 VanessaE even on Oldcoder's system, it would only take like 17 tries :P
18:47 sapier that's fine if there is a working port but if there's really a bug in there it'll run forever
18:55 VanessaE good point.
18:55 VanessaE well, easy enough to just disable the unit tests at the command line.
19:00 OldCoder joined #minetest-dev
19:42 artur99 joined #minetest-dev
19:43 OldCoder Hello. artur99 here lives in Romania. He may offer to produce a Romanian translation.
19:43 OldCoder If he makes such an offer, to whom should he be directed?
19:43 OldCoder artur99, wait for an answer. If there is none, try again in a day or two.
19:43 PilzAdam http://translate.minetest.ru/projects/minetest/core/
19:44 OldCoder artur99, review that link and PM me if you have further questions. PilzAdam thank you.
19:48 sapier joined #minetest-dev
19:51 OldCoder sapier, Hi. Animals are not producing errors despite the sdata nil problem. But I have not observed them yet.
19:51 OldCoder Should I have seen cows by now?
19:52 sapier yes there should be cows
19:52 sapier they don't spawn in already generated areas
19:52 sapier unless there have been cows before they should remain there
19:52 OldCoder sapier, thank you
19:53 OldCoder http://translate.minetest.ru/projects/minetest/core/ro/translate/?checksum=686e697538050e4664636337cc3b834f
19:53 OldCoder PilzAdam or others, artur99 is starting on Romanian translation
19:53 OldCoder We wish simply to ask if he is doing it right. Is that page what is needed?
19:54 PilzAdam OldCoder, looks good so far
19:54 OldCoder PilzAdam, thank you!
19:56 OldCoder sapier, One more question for today (I am busy at work)... If worlds exist and are large, there may be no cows? Is there any way to add cows to existing places?
19:56 iqualfragile joined #minetest-dev
19:57 troller joined #minetest-dev
19:57 proller joined #minetest-dev
19:57 sapier yes there's a method called secondary spawning to populate generated areas it's disabled for performance reason but it may be ok to enable it for some time
19:58 OldCoder sapier, where should I add this?
19:58 * OldCoder wants to see cows and fish and so on
19:58 OldCoder Moo!
19:59 sfan5 meow!
19:59 sapier it's a setting in gui ;-) and not very common used i'm gonna check how to enable per config file directly
20:00 OldCoder sfan5, sapier has not yet made cats, has he? We should politely request this
20:00 sapier and I need to check if there is a secondary spawner specified for cows ... not all mobs have it
20:00 OldCoder sapier, thank you
20:00 sfan5 Jordach once did a model for a cat in blender
20:00 OldCoder Hmm... if you see him, ask him about it
20:01 PilzAdam note that this channel should be used for engine development
20:01 OldCoder PilzAdam, yes
20:01 OldCoder PilzAdam, I am not always clear about what is engine
20:01 OldCoder Ah
20:01 OldCoder Ahnimals
20:01 OldCoder sapier, respond in PM when you can
20:01 OldCoder PilzAdam, but translations are engine?
20:01 PilzAdam yes
20:01 OldCoder Very well
20:01 * OldCoder returns to work
20:03 sapier PilzAdam do you have time to review the request queue fixes?
20:03 PilzAdam no, I will have more time at weekend again
20:10 kahrl sapier: is there a way to do it without making the result queue static?
20:11 sapier complete redesign?
20:11 kahrl hmm
20:12 OldCoder kahrl, if I may ask, is httpfetch likely to merge this month?
20:12 troller https://github.com/proller/minetest/compare/master...make
20:12 kahrl OldCoder, I think so
20:12 sapier pushing a pointer to a stack pointer into queue transfering it to another thread is fairly bad design this only didn't cause more problems because usually no timeouts occur
20:12 sapier -stack pointer + stack object
20:13 OldCoder kahrl, thank you
20:13 kahrl OldCoder: sapier has been working on a merge of the async lua events and httpfetch, I haven't had the time to look at it in depth yet but it seems good so far
20:14 OldCoder Very well
20:14 * OldCoder will be happy to provide testing services for all of this; as in run multiple worlds with the new stuff
20:15 kahrl sapier: maybe make the result queue reference counted?
20:16 kahrl it starts off at 2, then the producer and the consumer thread each decrement it by 1 once they are done with it
20:16 sapier we just removed smartpointers from game ;-)
20:17 sapier but yes that'd work at cost of increased overhead
20:19 kahrl does the result queue actually have to be a queue? if it's only used for one result some other overhead could be removed
20:20 artur99 joined #minetest-dev
20:20 kahrl using a semaphore or condvar to wait for the result (instead of calling sleep() in a loop) would remove some other overhead too
20:21 artur99 hi! romania has this charachers : a-z and ăîâșț
20:21 sapier yes seen that too it's next thing on my list
20:21 artur99 when i make the translation i must use the ăîâțș ?
20:21 sapier but I don't want to do everything at once
20:23 Jordach sfan5, don't ping me for nothing
20:23 PilzAdam artur99, yes, I guess so
20:23 artur99 ok, thanks
20:23 sapier imho making those queues static is change with lowest risk until a more complete design fix can be done
20:24 kahrl yeah and I'm pretty sure it works right now
20:24 kahrl (since there is only one thread other than main that makes requests)
20:25 sapier hmm I don't se a way how the new version could fail at all ... but it's not very good design of course
20:26 kahrl if there were several threads sharing the same result queue, they could compete with each other to clear each other's results
20:26 kahrl and those results would be lost forever (unless I'm missing something)
20:26 sapier true but is that function used by multiple threads?
20:26 kahrl atm it is not
20:27 sapier I assumed it to be thread specific but you're right that asumption is risky
20:27 kahrl maybe once somebody wants to try a pool of mesh update threads we need something more sophisticated
20:28 sapier imho a queue should be handled by some sort of control instance only responsible to start/stop consumers as well as providers
20:28 sfan5 Jordach: I am so sorry
20:28 Jordach sfan5, use J0rdach
20:28 sapier this way control instance can enforce validity of all involved subcomponents
20:29 damiel joined #minetest-dev
20:29 sapier but no idea how much changes would be required to modify it this way
20:34 bcnjr5_ joined #minetest-dev
20:34 kahrl actually... I noticed something
20:35 kahrl we wouldn't need a result queue at all, just some kind of indication that the main thread finishing producing the result
20:35 kahrl since in all three places where ResultQueue is used, the main thread places the result into a cache anyway, so the result could be fetched from there
20:36 kahrl s/finishing/finished
20:36 sapier that cache is a result queue ;-)
20:36 sapier but a global one
20:36 kahrl right
20:37 sapier would be fine with it too but there may be issues with scaling once more consumers/providers are involved
20:38 sapier current design was meant to have one result queue per request thread thus having constant time to fetch a result
20:38 sapier within a global queue result fetching time depends on number of already queued results
20:38 werwerwer_ joined #minetest-dev
20:39 sapier this wouldn't be an issue if it didn't matter what result is returned first
20:39 sapier async works this way
20:40 sapier btw the mutex queue isn't actually perfect for it ... it uses a events only (non counting semaphores)
20:41 EvergreenTree joined #minetest-dev
20:41 kahrl I thought it didn't even use events
20:41 bas080 joined #minetest-dev
20:42 sapier let me have a look I saw events when looking for this bug but I may have been in another thread by that time
20:43 sapier but I don't intend to completely rewrite this for 0.4.8 I was hoping to get 0.4.8 done by end of this month
20:44 kahrl as far as I'm concerned, anything that fixes the memory corruption on timeout and works otherwise is fine for 0.4.8
20:45 sapier you're right there isn't a event used in mutex queue at all
21:05 RealBadAngel joined #minetest-dev
21:26 Zeitgeist_ joined #minetest-dev
21:31 * VanessaE is back
21:43 VanessaE I just received a request, which I can't do but others might:  an option has been requested to force objects that are drawn with allfaces_optional to be drawn, client-side only, with plantlike instead.  E.g. all tree leaves.
21:44 VanessaE the rationale is that plantlike won't lower the framerate like allfaces_optional does when fancy leaves are enabled, and looks better than a solid, opaque leaf block when that feature is disabled.
21:44 VanessaE I can force this server-side with a very simple change to a couple of mods, but this seems like it would be more appropriate if done with a client-side config option.
22:08 KingsleyT joined #minetest-dev
22:15 ecube joined #minetest-dev
22:26 hmmmm hey guys
22:26 VanessaE hi
22:26 hmmmm yesterday 0.4.8 was supposed to be released but I was out all day
22:27 hmmmm kahrl, sapier, you guys have something really important getting cooked up at the moment
22:27 hmmmm ?
22:27 VanessaE I thought it was delayed till the end of November?
22:27 VanessaE (feature freeze)
22:27 hmmmm who said that
22:27 hmmmm the feature freeze is for a week
22:27 VanessaE I don't recall now
22:28 sapier I did request but noone did answer vanessae
22:28 hmmmm just FYI we can release as often as we'd like
22:28 hmmmm soooooooooog
22:28 hmmmm I personally see no reason why not to release right now while we're past a week of feature freeze and nothing experimental has been added
22:28 sapier hmmmmm for FYI we can't release a broken version at all ;-P
22:28 hmmmm why is this version broken
22:28 hmmmm last time i checked it was fine
22:29 sapier it's broken because it's corrupting memory, invalid lua error handlers and stalls within modmgr
22:29 hmmmm are you kidding
22:29 hmmmm omg
22:29 sapier everything could be patched right now if someone did agree ;-)
22:29 bcnjr5 left #minetest-dev
22:30 sapier no I'm not kidding we really do random memory corruptions in current master
22:30 hmmmm I take it the corrupted memory errors are really uncommon?
22:30 hmmmm because I haven't come across any related crashes yet
22:30 sapier have a look in issue list there are various strange crash reports for years
22:30 hmmmm for 'years'?
22:30 sapier yes
22:31 hmmmm well that makes it a bit different
22:31 hmmmm you made it sound like it was the most urgent thing ever
22:31 hmmmm do you know what's causing the problem?
22:31 sapier I actually have a patch ready
22:31 sapier yes I know why it happens
22:31 hmmmm who is disagreeing with it then
22:31 sapier nobody yet it hasn't been reviewed
22:32 hmmmm then?/
22:32 hmmmm by all means let's see it
22:32 ShadowNinja 0.4.8 should come after pcall_errfunc and sqlite_rollback IMO...
22:32 sapier how heavy is that sqlite rollback thing?
22:32 hmmmm sqlite_rollback is a new feature not a bugfix...
22:33 ShadowNinja sapier: What do you mean be heavy?
22:33 hmmmm again, only bugfixes
22:33 sapier https://github.com/sapier/minetest/tree/fix_RequestQueue this branch fixes two issues with request queue, the last commit is the random memory corruption fix
22:33 ShadowNinja hmmmm: It is sorta a bugfix since it fixes Minetest eating GBs of memory and taking it's sweet time on rollback and rollback_check.
22:33 hmmmm i'll take a closer look after i'm done eating
22:33 sapier both are bugfixes but first one was that wrong noone could've ever used this
22:34 hmmmm shadowninja, that's not a bugfix.  that's a new feature.  the current version is just a poor way of doing it
22:34 hmmmm it can wait
22:35 sapier Shadow as of rollback I think same way as hmmmm
22:35 hmmmm yeah everybody's trying to get their features in after the feature freeze and a day after it's supposed to have been released
22:35 hmmmm I'll check over those commits and then we'll have to ship that
22:35 ShadowNinja Alright, well pcall_errfunc then.
22:36 hmmmm yes, pcall_errfunc is fine
22:36 hmmmm I'll check that over too after
22:36 hmmmm by any luck it'll be out tonight
22:37 ShadowNinja I have a tweak for it though that should make it much smaller (sugested by sapier).
22:37 sapier you should fix this prior hmmmm reviews
22:38 ShadowNinja With a shell script it will take only a minute, but without it will take a hour. And I don't know sed well enough.
22:40 sapier come on :-) I'd have fixed that manually in 10 min ;-)
22:41 sapier we're talking about the non required parameter flip right?
22:45 ShadowNinja sapier: Yes, but that has to be changed in about 20-50 places.
22:46 ShadowNinja ~40 to be more precise.
22:46 sapier while we've been talking you could've fixed it  ... and fliping parameters without a good reason isn't quite usefulll .. no imho avoiding one "NULL" isn't a good reason (the other one is equalized by writing the default value) ;-P
22:48 ShadowNinja There, I got a working sed command...
22:48 ShadowNinja And I understand why they should be that way and agree.
22:49 hmmmm alright now
22:49 hmmmm hrmm, from your last commit in that branch it seems like the problem was that each caller would grab the same results, and you solved the issue by having each pass a key or something?
22:49 sapier no
22:50 sapier the problem was that a pointer to a stack object was pushed to another thread
22:50 hmmmm whaa
22:50 hmmmm who coded that
22:50 sapier once a timeout occured that stack object wasn't valid any longer
22:50 hmmmm oh, i see, request was global
22:50 sapier I fixed it by enforcing the pointer to be valid
22:51 sapier not best solution I know
22:51 sapier but as long as there's only one thread doing the requests thats ok
22:51 hmmmm why don't you just use heap memory
22:51 sapier because someone needs to clean up
22:51 hmmmm why can't the caller do that after he's done
22:52 sapier the caller quits by timeout
22:52 hmmmm this whole thing looks fubar
22:52 sapier while the request may still sit within the queue
22:53 sapier it is but complete redesign is way to heavy to fix the bug for 0.4.8 ... yes it has to be done and I intend to do so for 0.4.9
22:53 hmmmm well that's a relief
22:53 sapier if you have a look at the previous commit in that branch you see multicaller support is completely messed up too
22:54 hmmmm result_queue.pop_front() takes what, number of MS before timeout or something?
22:54 hmmmm i'm looking at the other commits now
22:54 sapier 1 second
22:55 hmmmm actually, hold on, I can't really read this on the github webpage
22:55 hmmmm need more context
22:55 sapier yes someone is doing terrible things if processing requires more then a second but corrupting memory isn't an option anyway :-)
22:55 hmmmm so you're starting to add doxygen comments to things
22:56 hmmmm zz
22:56 sapier not even celeron55 didn't remember what this thing is intended to do ;-)
22:56 hmmmm ahhhhhhhhhhhh THAT lump of code
22:57 hmmmm oh what a bunch of crap
22:57 hmmmm wasn't that the thing that EmergeThread used for queued block requests before I rewrote it?
22:57 sapier actually it's intention was quite good but it didn't ever do what it was intended to do
22:57 hmmmm see I figured it was just better to.... not.... attempt to fix it
22:57 hmmmm hell I didn't even understand what the problem was exactly
22:58 sapier hmmmm you should've fixed it now we have two queue implementations in there
22:58 hmmmm my implementation isn't generic though
22:58 sapier and a third one in async thread ... if it's possible I'm gonna merge them for 0.4.9
22:58 hmmmm it's not like a queue is something so advanced that it needs to be modularized
22:59 hmmmm no it's fine, it's fine, don't bother
22:59 hmmmm if it were more advanced code then I'd say you have a point
22:59 sapier no it's not fine ;-) we need to fix 3 queues instead of one ;-)
22:59 hmmmm but it doesn't do much, and if you try to generalize a single implementation, it won't be as well suited to all the other things
22:59 sapier and you can do a lot of things wrong in a queue ;-)
22:59 hmmmm also my queue implementation is not broken
23:00 hmmmm soooooooo
23:00 sapier :-) this one did work quite some time without everyone recognizing how broken it is ;-)
23:00 hmmmm don't break my shit
23:01 sapier :-) I'll ask again once I had a look if it is reasonable ... I don't know by now
23:01 sapier but for now you should just chek if this fix can be added
23:02 hmmmm so now there's no timing out
23:02 sapier it is
23:02 hmmmm you just keep looping until you get your item
23:02 sapier no
23:02 hmmmm I don't see any other breaks
23:02 sapier timeout throws the exception
23:03 hmmmm oh an exception
23:03 hmmmm ahh, ItemNotFoundException
23:03 sapier has been done this way befor
23:03 hmmmm that's pretty stupid
23:03 hmmmm people really need to stop playing with exceptions
23:03 sapier yes it is but I didn't want to invest that much time to fix something I need to rework anyway
23:03 hmmmm right, all this is temporary
23:04 hmmmm sure
23:04 hmmmm I can go for this fix for now
23:05 ShadowNinja hmmmm: This error occurs a few times, can you fix it? http://pastebin.ubuntu.com/6418430/
23:05 sapier actually in this situation a exception isn't that bad ... if I'm gonna rewrite this I may remove the timeout ... for what sane reason should we miss to process a request?
23:06 hmmmm let's say something gets TOTALLY screwed up
23:07 hmmmm is it better to go around in an infinite loop and lock everything up
23:07 hmmmm or just bomb out with an error
23:07 sapier "TOTALLY screwed up" is synonym to assert for me ;)
23:07 hmmmm let's be honest, if it times out there's something wrong anyway
23:07 hmmmm yes, weren't there asserts there anyway?
23:07 sapier wrong location
23:07 ShadowNinja pcall_errfunc updated with sapier's tweak.
23:07 sapier the asserts did check result.key == name
23:08 sapier quite useless
23:08 hmmmm oh
23:09 sapier they git by pure chance sometimes :)
23:09 sapier -git+hit
23:09 hmmmm ShadowNinja, iirc I do a lot of the same thing all over
23:09 hmmmm isn't there more than 1 of those warnings?
23:10 ShadowNinja hmmmm: Yes, I got two I think.
23:11 hmmmm that's a really stupid warning
23:12 ShadowNinja hmmmm: Actually that's the only one I got, but I didn't "make clean" first.
23:12 hmmmm does encountering that warning disable strict aliasing rules?
23:13 hmmmm if so then there'd be a reason to change it
23:13 hmmmm or does it disable strict aliasing for that group of expressions?
23:13 hmmmm shrug
23:14 kahrl since the called function is in a different translation unit, the warning can't change how it is compiled
23:14 sapier the strict aliasing rules warning is serious
23:14 sapier this may result in very strange errors
23:15 hmmmm that's true
23:15 hmmmm in this case I really don't know if it's worth fixing though, it's a stupid error
23:16 hmmmm do they want me to start using templates for enums?
23:19 sapier no idea I didn't have a closer look at that warning by now
23:19 kahrl just change the type of rot to int
23:19 hmmmm I know what it wants me to do but it's needlessly verbose
23:19 hmmmm int roti;
23:19 hmmmm pass roti
23:19 hmmmm Rotation rot = (Rotation)roti;
23:20 hmmmm I'm sure I would've fixed that warning if I had it on my compiler though
23:20 sapier ok I need to sleep a little bit more early today so good night :-)
23:20 kahrl since you only assign it to dschem.rotation anyway you can skip the last of those
23:23 hmmmm gotta take a nap now
23:23 hmmmm later
23:40 sapier1 left #minetest-dev
23:51 NakedFury joined #minetest-dev

| Channels | #minetest-dev index | Today | | Google Search | Plaintext