IRC log for #minetest-dev, 2014-04-22

Time	Nick	Message
00:03		PilzAdam joined #minetest-dev
00:05	PilzAdam	Megaf, #1243
00:05	ShadowBot	https://github.com/minetest/minetest/issues/1243 -- /clearobjects spams errrors
00:06	Megaf	PilzAdam, what can we do about it?
00:09	Megaf	now, how do I compile the sapier's android branch for android?
00:38	MegafODroid	well, it still spamming the terminal
00:56		kaeza joined #minetest-dev
00:59		Ritchie joined #minetest-dev
01:00		Anchakor_ joined #minetest-dev
01:24		Anchakor joined #minetest-dev
02:42		smoke_fumus joined #minetest-dev
03:39		stormchaser3000 joined #minetest-dev
03:53	MegafODroid	holly molly
03:53	MegafODroid	I just made a really bad loop
03:53	MegafODroid	using a not gate
03:53	MegafODroid	what can I do now?
03:55	VanessaE_	uninstall mesecons, dig the offending node
03:55	VanessaE_	and figure out why it didn't overheat.
03:55	VanessaE_	and keep mod chat to #minetest
03:58	stormchaser3000	hey vanessa
03:58	stormchaser3000	have you ever seen this error?
03:58	stormchaser3000	22:25:19: ERROR[ConnectionSend]: con(6/1)RunTimeouts(): Peer 2 has timed out. (source=peer->timeout_counter)
03:58	VanessaE_	ask sapier about that one.
03:58	VanessaE_	I believe he and megaf here had to deal with that one
03:58	stormchaser3000	but sapier is not on
03:58	Anchakor	heh I forgot I got this chan on autojoin
03:59	Anchakor	how has minetest developed in past year or so?
04:03	MegafODroid	VanessaE_: managed to take the not gate before the mod start running (restart the server)
04:03	VanessaE_	that works too
04:03	VanessaE_	but take this to #minetest
04:03	VanessaE_	this is not the place for it.
04:04	MegafODroid	yep
04:04	MegafODroid	stormchaser3000: that's a workaround we found
04:04	MegafODroid	it means that a client disconnected and the server can reach it
04:05	stormchaser3000	hmmm
04:05	stormchaser3000	wierd
04:05	stormchaser3000	i can't use my server when that happens though
04:06	stormchaser3000	simaler to the draw hotbarr = NULL thing
04:06	MegafODroid	stormchaser3000: hold on
04:06	VanessaE_	update your minetest build then
04:06	VanessaE_	all that shit has long since been fixed.
04:06	stormchaser3000	i just built a few hours ago
04:06	stormchaser3000	actualy like an hour ago
04:07	MegafODroid	VanessaE_: that's a different thing
04:07	MegafODroid	#1170
04:07	ShadowBot	https://github.com/minetest/minetest/issues/1170 -- A few bind_address fixes by kahrl
04:08	VanessaE_	that doesn't seem related.
04:08	MegafODroid	my problem was the #1231
04:08	ShadowBot	https://github.com/minetest/minetest/issues/1231 -- ServerThread ServerEnv & Server::ProcessData() ERRORS
04:09	MegafODroid	stormchaser3000: are you using ipv6:
04:09	MegafODroid	?
04:09	stormchaser3000	yeah
04:09	MegafODroid	hm
04:10	MegafODroid	stormchaser3000: pastebin a bit of the log
04:10	stormchaser3000	ok
04:12	stormchaser3000	http://pastebin.com/Kr9hgJeg
04:14	MegafODroid	stormchaser3000: do you have a github account?
04:14	stormchaser3000	yeah
04:14	MegafODroid	stormchaser3000: I mean, when that error is happening, that bit of log
04:14	MegafODroid	oh wait
04:15	MegafODroid	and you said you cant use you server when you get that error?
04:15	stormchaser3000	yeah
04:16	stormchaser3000	my client never makes it to the world
04:16	stormchaser3000	but downloads all the data
04:16	MegafODroid	stormchaser3000: well, you can post a reply here https://github.com/minetest/minetest/pull/1170
04:17	MegafODroid	describing very well what's going one, from whem is your git build and a link to pastebin
04:17	MegafODroid	VanessaE_: or should he open a new issue?
04:17	VanessaE_	I'd say a whole new issue but ....
04:17	VanessaE_	too late.
04:19	stormchaser3000	ok done
04:20	MegafODroid	stormchaser3000: you better talk to sapier tomorrow
04:21	VanessaE_	I was gonna ask for the full log of the "I can't get into my server after this point" event.
04:21	VanessaE_	because what you posted to that bug is, honestly, useless.
04:22		nore joined #minetest-dev
04:23	stormchaser3000	which i did post after
04:23	VanessaE_	nope.avi
04:23	VanessaE_	what you posted is basically one line worth of error.
04:23	stormchaser3000	look again
04:23	VanessaE_	where's the rest of the event?
04:24	stormchaser3000	i just posted a patebin
04:24	VanessaE_	I just did.
04:24	VanessaE_	you linked to this: http://pastebin.com/YN0xTGK5
04:24	VanessaE_	there's exactly one error shown there
04:24	VanessaE_	where's the rest of the event?
04:24	stormchaser3000	that is because that is the only error there is
04:25	VanessaE_	and why is the IRC mod loading after this error?
04:25	VanessaE_	(or connecting, rather)
04:25	VanessaE_	ditch the IRC mod and try again
04:26	stormchaser3000	bye i gtg
04:26	VanessaE_	...
04:26	MegafODroid	That's not good
04:26	* MegafODroid	can see smoke comming from VanessaE's head
04:28	MegafODroid	Good night VanessaE_
04:50		OldCoder joined #minetest-dev
05:17		jin_xi joined #minetest-dev
06:15		cj joined #minetest-dev
06:18		Garmine joined #minetest-dev
06:33		troller joined #minetest-dev
06:48		Guest99516 joined #minetest-dev
07:15		grrk-bzzt joined #minetest-dev
07:17		grrk-bzzt joined #minetest-dev
07:28		ImQ009 joined #minetest-dev
07:31		rsiska joined #minetest-dev
07:48		darkrose joined #minetest-dev
07:48		darkrose joined #minetest-dev
08:08		CraigyDavi joined #minetest-dev
08:08		PenguinDad joined #minetest-dev
08:47		darkrose joined #minetest-dev
08:47		darkrose joined #minetest-dev
08:53		rsiska joined #minetest-dev
08:56		restcoser joined #minetest-dev
09:48		Calinou joined #minetest-dev
09:56		PenguinDad joined #minetest-dev
10:08		diemartin joined #minetest-dev
10:17		iqualfragile joined #minetest-dev
10:40		Jordach joined #minetest-dev
10:50		zyxwvuts joined #minetest-dev
10:51		ockomock joined #minetest-dev
10:59		PenguinDad joined #minetest-dev
11:27		RealBadAngel joined #minetest-dev
11:28	RealBadAngel	hi
11:29	RealBadAngel	https://github.com/minetest/minetest/pull/1244
11:29	RealBadAngel	any comments?
11:33		rsiska joined #minetest-dev
11:55		werwerwer_ joined #minetest-dev
12:56		kaeza joined #minetest-dev
12:57		hmmmm joined #minetest-dev
13:17		VargaD joined #minetest-dev
13:57		PenguinDad joined #minetest-dev
14:17	CiaranG	Has anyone ever considered/worked on implementing non-axis-aligned collision box detection for entities?
14:42	celeron55	i am pretty sure no
14:43	celeron55	i think that would be a right point to pick a small physics library instead of exercising NIH
14:46	sfan5	NIH?
14:46	celeron55	http://en.wikipedia.org/wiki/Not_invented_here
14:46	CiaranG	Well I'm not sure about that. It would be a minor addition to the existing code.
14:50	CiaranG	Also, obviously the vast majority of collision boxes in minetest are axis-aligned. And for most of the remainder, it does't matter. So using a library that isn't optimised for that case would probably be a bad move.
14:55		zat joined #minetest-dev
14:58		BlockMen joined #minetest-dev
15:25		Calinou joined #minetest-dev
15:37		rdococ joined #minetest-dev
15:41		sapier joined #minetest-dev
15:43		rsiska joined #minetest-dev
15:45	sapier	"CiaranG Has anyone ever considered/worked..." considered ... yes, really worked at it ... no
15:47	sapier	stormchaser3000 the timeout error message was made visible recently but it's been there before too just at debug loglevel. basicaly it's telling your client doesn't respond for about 5 seconds and is dropped for that reason
15:51	ShadowNinja	~tell RealBadAngel 1244 seems good.
15:51	ShadowBot	ShadowNinja: O.K.
15:56	ShadowNinja	VanessaE_: The later IRC mod connection is intentional. It prevents server starts -> IRC mod connects -> other mod after IRC mod crashes -> script restarts server -> repeat loops, causing flooding.
15:59	sapier	hmm so we don't know anything about that bug despite for some reason client times out
15:59	Anchakor	has any progress by anyone (public on not) been done on moving minetest past minecraft-clone to its-own-game?
16:00	sapier	once again what's the question?
16:02		PilzAdam joined #minetest-dev
16:04		tomreyn joined #minetest-dev
16:04	ShadowNinja	Anchakor: There are some games which are very different from MC. But by it's nature MT can't be made very different.
16:05		grrk-bzzt joined #minetest-dev
16:05		Jordach_ joined #minetest-dev
16:05	sapier	can someone please check #1240 ... If noone complains I'm gonna merge it in a couple of hours. I know it may cause trouble if I missed something but the current way of doing it causes even more strange issues. So please READ the code and try to find bugs in there.
16:06	ShadowBot	https://github.com/minetest/minetest/issues/1240 -- Fix broken local formspec handling by sapier
16:06	Calinou	Minetest is not a Minecraft clone, Anchakor
16:06	sapier	Calinou: :-) that's what everyone tells, but without additional information it's quite hard to understand.
16:07	Calinou	it's a bad generalization
16:07	sapier	The additional information I talk about is minetest not beeing a game at all but only an engine. The games themself define if it's a clone or something different.
16:07	Calinou	hence, don't call it a Minecraft clone, for Sam's sake (:D)
16:08	sapier	guess if someone wants to do clone minecraft there might be little difference
16:09	Calinou	there are games that try doing it like realtest
16:09	Anchakor	well the point of the game is the same no? build awesome shit from blocks with other people
16:09	Calinou	both games have no real goals, but Minecraft has some "end" (a fairly bad one...) and there's more activities to do
16:09	sapier	as of my pov the game is writing/improving the engine ;-)
16:10	Calinou	Minecraft servers have more varied objectives
16:10	Anchakor	I'm instested in some adventure mode gameplay, like roguelikes
16:10	Calinou	(eg. "Hardcore" PvP, I almost never saw that on Minetest)
16:10	sapier	Anchakor following this logic any third person shooter would be a doom clone ;-)
16:11	Anchakor	sapier: they are, but later there was too many and it was necessary to classify them more properly :)
16:11	sapier	minetest provides a scripting api you're free to write a game like the one you requested
16:11	Anchakor	I know
16:12	Anchakor	I want to know if anyone has done / is trying to do with MT engine something else then a building game
16:13	sapier	I suggest asking this at minetest (if you didn't already do it)
16:15		twoelk joined #minetest-dev
16:21	celeron55	actually it's not trivial to write a game like that using minetest
16:21	celeron55	i'm guessing roguelikes have a more active world than what the engine currently is locked to do
16:22		Amaz joined #minetest-dev
16:22	celeron55	or, like, it's unable to really run anything in the world outside of the player's close range
16:23	sapier	but does it really need to do those things immediatly?
16:23	celeron55	adding something that allows for a game to do something like that in some reasonable way will be accepted
16:23	celeron55	if the things move around, then yes
16:23	celeron55	otherwise they won't affect anything along their path
16:24	celeron55	or do something else that can have global effects
16:24	sapier	so you're actually talking about the forceload entities that are requested every now and then
16:24		Amaz joined #minetest-dev
16:24	celeron55	and global effects implies that the world probably has to have a limited size in that usage
16:24	celeron55	sapier: it's part of it, kind of; but simply forceloading entities may require too much processing
16:24	celeron55	there's designing to do
16:25	sapier	well you can "emulate" quite a lot of things by using global step and external data storage but yes that's gonna be quite complex
16:26	celeron55	Anchakor: you may already guess but i did start a project that actually involves a bit of something else, but i freezed the project quite quickly because i have to focus on my main project (which is not minetest nor minetest-related)
16:27	twoelk	couldn't some NPC's be designed as bots that use some basic client of their own? Maybe having some sort of authentification that shows they come from the game and not from outside.
16:27	Anchakor	celeron55: cool, will keep an eye for that
16:28	celeron55	twoelk: that's a huge overhead
16:29	sapier	twoelk basicaly mobf mobs built from forceload entities would be almost exactly what you request (without causing the network overhead)
16:29	celeron55	twoelk: you're limited to same amount of them as there are players and even less due to running the clients
16:29	sapier	yes celeron but that's true for all active entties
16:29	sapier	active as of doing things themselfs
16:30	twoelk	I don't think that usefull for a herd of sheep but rather for maybe five or so enemies of a shooter type game
16:31	sapier	hmm e.g. if you've got some archers and guards attacking a player?
16:31	twoelk	sort of
16:31	sapier	that's already possible ;-)
16:32	twoelk	so the mob/npc can traverse terrain not generated by my client but on its own right
16:33	sapier	they're not very smart but yes for terrain not beeing to complex they can
16:33	twoelk	so we need a game that showcases that
16:34	twoelk	game as in minetest subbgame
16:34	sapier	but quite less people did try this by now so expect some bugs in there
16:36	twoelk	I guess making them smarter (process more code) will make them perform slower
16:36	ShadowNinja	hmmmm: It seems that MGv6 ignores the caves/nocaves flags.
16:37	sapier	yes twoelk that's the basic problem, more smart usually causes more load and as long as mods have to do all their processing within server step this increases lag
16:38	sapier	guess I'm gonna make async api available for mods soon
16:38	twoelk	eh? so we need to find a way to bypass server step
16:39		rubenwardy joined #minetest-dev
16:39	sapier	basically yes ... but we need to keep in sync to it too so it's a little bit tricky
16:40	sapier	any comments to #1241? it's supposed to cleanup mainmenu and provide base for simplified android menu
16:40	ShadowBot	https://github.com/minetest/minetest/issues/1241 -- Add formspec toolkit and refactor mainmenu to use it by sapier
16:41	twoelk	maybe prediction and analysis should be done in several parallel threads with a junk collection of sorts to get rid of threads not used as dessision went another thread
16:42	sapier	problem isn't the prediction and analysis itself but all of it needs map access ... that's the bottleneck
16:43	sapier	of course analysis can be done smart or stupid too but as long as it's part of server step that's not a difference
16:45	twoelk	I dont knoso a mod can never disconnect from the pulse the server supplies?
16:46	sapier	not yet
16:46	twoelk	-i dont kno
16:46	twoelk	so this is where the bot jumps in
16:46	sapier	no
16:47	sapier	a bot is some external thing that's why you have to run through full network stack causing way more overhead then necessary
16:47	twoelk	grrr
16:47	sapier	of course you can do it with a bot, no question
16:47	sapier	but imho this isn't best way to do it
16:48	twoelk	although if I have one bot that moves several mobs from the same connection?
16:48	sapier	can't be done
16:49	sapier	one connection one player/bot
16:49	twoelk	it definitly is the wrong way to do it! The best way would be to write that into the engine, but I dont see that any time soon
16:50	sapier	you'd be surprised how much of that work is already done
16:50	twoelk	well not "that" but some sollution
16:50	twoelk	oh
16:50	sapier	async is there, just not enabled for mod api
16:50	twoelk	hope appears on the horizon
16:50	sapier	yes it's not gonna provide full set of features
16:51	sapier	e.g. map modification has to be done from server step by now
17:00	twoelk	I'm still wondering wether the quasi physical map and the semi sapient entities should not be processes of their own, working each in their own container of som sort. Maybe not as complete botclient but something that can crash and not take down the map with it and yet not jumping through the big loop of a full network stack but coming from somewhere closer.
17:01	sapier	if async operation would gain map access you'd have exactly this but I don't wanna predict how much overhead map locking would cause in this scenario
17:10	ShadowNinja	Pushing soon: http://sprunge.us/TZPe?diff
17:10	twoelk	actually if the mapper could connect to minetest as some client it would be possible to make maps while the world is running, maybe there are other uses for a sort of client access that 3rd party programs can use
17:10		Zeitgeist_ joined #minetest-dev
17:10		Zeitgeist_ joined #minetest-dev
17:12	sapier	better not do that twoelk, mapper would cause full map to be generated ;-)
17:12	twoelk	oops
17:12	sapier	64kx64kx64k nodes ;-)
17:12	twoelk	need avoid_unloaded function
17:13	sapier	that's forceload ;-)
17:13	sapier	wait
17:13	sapier	misread your message
17:13	sapier	but there's no usecase for that feature except of mapper
17:14	twoelk	at the moment, offer it and someone will find more uses :-P
17:15	sapier	so we provide a feature to double map size requirement because of someone may find a way to use it? ;-)
17:15	twoelk	actually zombies could avoid unloaded areas as there are no players to eat there
17:15	sapier	zombies wont even exist there because unloaded areas aren't loaded ;-)
17:16		proller joined #minetest-dev
17:16	twoelk	? why would that double the size?
17:16	sapier	that's been a suggestion for a new completely unrelated feature
17:16	* twoelk	is rereading what was said
17:17	sapier	as you said once provided someone will find a usecase for it ;-)
17:17	sapier	because I don't know of what use a feature to double map size requirement could be :-)
17:18	sapier	imho mapping should be done on server
17:18	twoelk	ok I want my program to avoid unloaded area so to not generate new chunks.....why does that make the map bigger?
17:19	sapier	the feature "double map size" is not related to your feature unload area ;-)
17:20		proller joined #minetest-dev
17:20	* twoelk	is confused, must have missed something
17:20	sapier	g
17:20	sapier	well maybe I haven't written it in best way
17:20	sapier	I was aiming at your request to add a feature without a real usecase
17:21	sapier	minetest is suposed to be a game engine not a map database
17:21	twoelk	well then I request fancy smart cars for everyone
17:22	twoelk	no the map database is a sqlite file
17:22	sapier	a feature to not load blocks from a client would be only usefull for a single (argueable) usecase ... and it's quite a lot of work to provide it
17:22	twoelk	minetest is a database server, in parts, at the moment
17:23	sapier	nope
17:23	sapier	minetest uses a database
17:23	sapier	it's just gonna send to client what client needs to know
17:23	twoelk	that is why the mapper is not allowed to read the database, because minetest has read/write access when running
17:24	sapier	yes ... but you should do a backup by time anyway so mapper could run when doing that backup
17:24	sapier	and for client maps client could build them upon it's cache which most likely is what user needs to know anyway
17:25	twoelk	sqlite allows only one user at a time, in our case this is a server that channels several clients into a single user process, if I understood sqlite correctly
17:25	sapier	if you wanna do a backup you need to stop server anyway
17:26	twoelk	gah you cant write while I#m typing :-D
17:26		proller joined #minetest-dev
17:30		rsiska joined #minetest-dev
17:34		EvergreenTree joined #minetest-dev
17:35		rsiska joined #minetest-dev
17:37		rsiska joined #minetest-dev
17:38		rsiska joined #minetest-dev
17:42		RealBadAngel joined #minetest-dev
17:43		rubenwardy_ joined #minetest-dev
17:56	RealBadAngel	so, going to push 1244 now
18:03	sapier	ok
18:07	RealBadAngel	done. i am browsing now old issues and checkin whats worth merging whats closing
18:08	sapier	I'm merging 1240 as noone wants to check it anyway and current code is quite broken
18:08	sapier	noone except blockmen
18:09	RealBadAngel	its too complex to just read and say its ok
18:09	sapier	I know that's why it's important to read and at least confirm it's not obviously wrong
18:10	sapier	the old bug wasn't seen by anyone for weeks ... yet it's quite obvious once you know where to look
18:10	BlockMen	RealBadAngel, http://dev.minetest.net/Git_Guidelines point 2.
18:10		proller joined #minetest-dev
18:11	RealBadAngel	BlockMen, https://github.com/minetest/minetest/pull/795
18:11	RealBadAngel	i wasnt original author of it
18:11	BlockMen	git rebase?
18:12	BlockMen	you can rename commits
18:12	sfan5	you can use git commit --amend to rename itr
18:12	sfan5	-r
18:13	RealBadAngel	i will remember that
18:33	RealBadAngel	also, going to push https://github.com/minetest/minetest/pull/1237
18:33	RealBadAngel	metaducky is right
18:43		rsiska joined #minetest-dev
18:46	ShadowNinja	twoelk: The C++ mapper works while the world is running, although it's a bit slower since it has to wait for the server to release any locks it sets.
19:05	sapier	I'm gonna merge #1240 now and then I'm gonna duck and wait for bugreports ;-)
19:06	ShadowBot	https://github.com/minetest/minetest/issues/1240 -- Fix broken local formspec handling by sapier
19:07		EvergreenTree joined #minetest-dev
19:10		OldCoder joined #minetest-dev
19:45	twoelk	ShadowNinja: I could have sworn... now when was that fixed ... bah talking rubbish 'cause I'm not up to date.
19:46		EvergreenTree joined #minetest-dev
19:47	ShadowNinja	twoelk: Be carefull though, I don't know if Minetest handles SQLITE_BUSY well.
19:48	Megaf	sapier: don't
19:48	Megaf	did you do it already?
19:49	sapier	yes
19:49	sapier	what's wrong?
19:51		proller joined #minetest-dev
19:51	Megaf	Uh, nothing, I would test it for you before you merge, but that's a bug on the windows build, so I can't thest
19:52	sapier	well the actual bug was on linux too but it didn't happen to cause an error (by now)
20:21		Fresh_me_ joined #minetest-dev
20:24	Megaf	sapier: Wher's the apk?
20:24	sapier	I knew I did forget something
20:28	Megaf	sapier: If you could publish somewhere on how to build mientest for android thet would be great. I got the whole android SDK plus emulators en Eclipse already
20:29	Megaf	and I managed to export apps to my android phone too
20:29	sapier	build/android and then enter make
20:29	Megaf	Or write very briefly and I write a tutorial
20:29	Megaf	is that simple?
20:30	sapier	almost
20:30	sapier	you need to have sdk as well as ndk somewhere on disk, make will ask you on first run where to find the
20:30	sapier	m
20:31	sapier	it's only tested for arm by now
20:31	Megaf	ok
20:31	sapier	guess to build x86 android makefile requires some adjustments
20:31	Megaf	right
20:31	Megaf	I will see that when I get home
20:32	sapier	there's some portability code in there but it's not complete
20:33	Megaf	I'm checking the code, looks cool
20:34	sapier	for some of the makefiles you can specify the build architecture but it's not done for all
20:37	sapier	http://animalsmod.comuf.com/downloads/Minetest-debug.apk link for everyone who doesn't want to compile
20:38	sapier	I've already managed to merge (or at least split to independent commits) most of the andorid changes
20:38	sapier	remaining code is almost completely separated from core
20:42		Exio4 joined #minetest-dev
20:49		jin_xi joined #minetest-dev
21:01		Amaz joined #minetest-dev
21:25		BlockMen left #minetest-dev
21:28		ImQ009 joined #minetest-dev
21:30		rsiska joined #minetest-dev
21:30	hmmmm	you know, what if the mod model needs to be fundamentally reinvented
21:30	VanessaE_	oh G8d
21:30	hmmmm	instead of having mods process things in the server step, what if there was a separate mod thread
21:30	VanessaE_	G*d
21:31	VanessaE_	well
21:31	VanessaE_	many have been proposing just that
21:31	VanessaE_	but that would be fundamentally hard to do
21:31	VanessaE_	and it would probably break LOTS of things
21:31	VanessaE_	it's a good idea though
21:31	hmmmm	any sufficiently good innovation would need to break things
21:32	VanessaE_	true
21:33	VanessaE_	it would finally solve the core issue of mods just plain locking up the server
21:34	VanessaE_	(mostly)
21:34		OldCoder joined #minetest-dev
21:34	VanessaE_	though that would require introducing locks or so
21:35	VanessaE_	and I guess it was decided that modders couldn't be trusted with that
21:36	hmmmm	well the idea is to have a queue of pending mod actions and every server step those will be performed
21:37	hmmmm	only big problem i can forsee is fighting for access of the map for reads
21:37	hmmmm	what if map access was done with some kind of async io
21:37	hmmmm	Lua can do closures, right?
21:37	hmmmm	so it doesn't even need to be callback based
21:37	VanessaE_	it can but I don't understand how they work.
21:38	VanessaE_	(that's one programming concept I've never quite gotten my head around)
21:38	VanessaE_	now as for read access, what about read priorities instead?
21:38	VanessaE_	async like you are thinking, but with stacked priority
21:39	hmmmm	well the priority would be the hundred million things that need it inside of server thread
21:39	hmmmm	if serverthread wants to read map, it locks it, but a mod wanted to read the map as well, so you still get the same sort of laggyness
21:39	hmmmm	...it's just that instead of being lualocked it'll be waiting on a maplock
21:39	VanessaE_	like if two processes ask for this one node at the same time, and no priority is set, well then the result would be first come first served (random, probably). if a mod asks the engine for priority access to the node, then it gets first crack at it.
21:40	hmmmm	it's not that simple.
21:40	VanessaE_	hm.
21:40		proller joined #minetest-dev
21:40	VanessaE_	well I suppose it wouldn't be particularly simple, no
21:41	VanessaE_	but the reason it was decided against doing some kind of multithreaded Lua was just exactly this issue, of multiple mods potentially reading or writing the same node or mapblock at the same server step
21:41	VanessaE_	or so I thought
21:42	VanessaE_	and that can only be solved either by locking or by some kind of priority system (even if it's static)
21:44	VanessaE_	(I'm counting one Lua thread versus the engine as "multithreaded" for this discussion)
21:44	sapier	hmmm you don't need to reinvent full moding api if you stick to async engine ...yet the map locking issue will still be there
21:44	VanessaE_	(because it'll eventually lead to truly multithreaded Lua eventually)
21:45	sapier	lua isn't capable of doing multithreading, no chance to get this done
21:46	VanessaE_	(also I work for the Department of Redundancy Dept. at the Bureau of Excess Parenthesis)
21:46	VanessaE_	sapier: it isn't right now, no. But anything is possible in the future.
21:46	sapier	well if you intend to rewrite a multithreaded lua interpreter maybe
21:47	VanessaE_	in the old days, we called it multitasking/multithreading even if all you had was a couple of IRQ sources to drive your threading engine.
21:47	sapier	imho providing a callback based async mechanism is way more promissing
21:48	VanessaE_	so it's really a matter of terminology, if you'd rather call it async, even if the effect is the same (two or more Lua mods seem to run concurrently), then that's fine.
21:48	sapier	no it's not the mods running concurrently it's only jobs requested by lua mods
21:48	VanessaE_	I know.
21:49	VanessaE_	I'm talking about the effect the end-user sees.
21:49	sapier	it's only gonna help for some classes of tasks to be done, you wont get a mob to react to a punch this way
21:49	VanessaE_	true.
21:49	sapier	but you can make a mob find a path in parallel to other things
21:50	sapier	or a spawn algorithm find a suitable position
21:50	sapier	or mapgend generating map
21:50	VanessaE_	indeed.
21:51	VanessaE_	send moretrees/plantslib off to populate a block while the rest of the system goes on to generate another block, etc.
21:51	VanessaE_	(without having to resort to saplings mode)
21:51	sapier	for example
21:52	VanessaE_	hmmmm is right though that Lua needs to be more "detatched" from the engine
21:52	sapier	yes
21:53	sapier	but main issue isn't game but map
21:53	VanessaE_	yeah
21:53	sapier	we need to enable it to read data without needing to lock it
21:53	sapier	once this is done async calls could be allowed to do read operations only resulting in asny things not blocking server
21:54	VanessaE_	but here's the thing, if Lua's running in it's own thread, it's still gonna be event-based, right?
21:55	sapier	of course
21:55	sapier	what else?
21:55	VanessaE_	that means mapgen callbacks of some kind. So what would lock the map?
21:55	VanessaE_	in other words, what would be left to solve at that point?
21:56	sapier	well the async threads could access map at any time while server thread (with it's own lua instance) does other callbacks
21:56	VanessaE_	(I'm ignoring just for the moment, hmmmm's idea of queueing up Lua events)
21:56	sapier	that'd be a problem
21:56	VanessaE_	right.
21:57	sapier	well async jobs are queued anyway
21:57	sapier	you can see this if you enter modstore in mainmenu
21:59	sapier	hmmmm making async available for mods is quite a minor patch, the big deal is providing safe map access
22:00	VanessaE_	that begs the question:
22:00	VanessaE_	what's the absolute worst thing that can happen if it's left "unsafe"?
22:00	sapier	total corruption of internal game data
22:00	VanessaE_	suppose you just randomly opened up async access to all, with access to the map through the usual ..... oh
22:01	VanessaE_	"Okay, that's bad. Important safety tip, thanks Sapier."
22:01	sapier	in best case player actions are only lost, in worst case whole database is messed up
22:01	VanessaE_	wait a second here...
22:01	VanessaE_	just how low-level is the access?
22:02	sapier	depends, if our database backend does proper locking (I assume it doesn't) we may only get a deadlock
22:02	* twoelk	's discussion topic scanner locks in on something promising
22:02	VanessaE_	well that's what I mean...modders don't need access to nearly THAT low level
22:03	VanessaE_	(and anyway a modder could delete a server's map.sqlite anyway)
22:03	sapier	it's not what modders need it's what you need to do to get that data
22:03	sapier	if you set a node from multiple async threads as well as main server thread, result will be undefined
22:04	VanessaE_	well the obvious case there is that each of those threads should write the database and whatever was last written is what takes effect.
22:04	sapier	in best case db is consistent at last one to set it, in worst case (without db backend having proper locking) internal db data is corrupted
22:04	VanessaE_	if "undefined" is possible, then it's badly written.
22:05	VanessaE_	but that's something that happens at a level lower than what the modder is coding at.
22:05	VanessaE_	that's what I'm talking about
22:05	VanessaE_	what the modder is supposed to have access to
22:05	sapier	yes but the modder is able to cause it
22:06	VanessaE_	yes, but that would suggest a bug in the database code, not in the Lua interpreter etc.
22:06	sapier	no
22:06	VanessaE_	yes
22:06	sapier	if database code is not supposed to be used multithreaded it's not abug
22:06	VanessaE_	the modder doesn't have access to the database (even vmanip doesn't get that close)
22:06	sapier	same as a lua stack isn't meant to be handled by different threads
22:07	VanessaE_	which means if you open up async access, you have to harden the database engine against it.
22:07	VanessaE_	failing to do so would be a bug
22:07		rsiska joined #minetest-dev
22:07	VanessaE_	that's not really the same thing as, say, failing to lock some API function such as set_node()
22:07	sapier	of course we can add our own locks in upper layer ... that's most likely what we need to do but that's the tricky thing if we do it wrong we don't get any performance gain in good case and deadlocks in bad
22:08	VanessaE_	well the idea isn't necessarily a performance gain as much as consistent performance.
22:08	VanessaE_	however
22:09	VanessaE_	if you have multiple cores, and you really are running Lua on its own thead as far as the CPU is concerned, then there surely will be a performance gain
22:09	sapier	in case of server thread waiting for a lock taken by async we dont win anything
22:09	VanessaE_	right
22:10	VanessaE_	the idea with async is to stop a multi-second lockup because of some slow-by-necessity mod from lagging two dozen users :)
22:10	sapier	lua isn't anything to run, it's just a language to do things, think of it as core functions as of threading the only difference to c++ functions is their data is limited to a single thread only
22:10	VanessaE_	when I say Lua in that context, I mean the interpreter :P
22:11	sapier	well of course we could do locking to make sure a single thread can access a lua context by time only
22:11	VanessaE_	the code that runs the code, don't split hairs
22:11		OldCoder joined #minetest-dev
22:12	sapier	no it's not even a interpreter activity is still core itself, there's no magical transition to lua the one doing is still the one doing core code
22:12	VanessaE_	mmm, true
22:12		rsiska joined #minetest-dev
22:13	sapier	think of some worker switching tools once he's doing it with c++ next time with lua
22:13	VanessaE_	I get tha
22:13	VanessaE_	that*
22:13	VanessaE_	though some would say the time it takes to switch tools is pretty horrendous :P
22:13	VanessaE_	the idea here is to give the worker a couple more hands
22:13	sapier	while the lua tool can be used by a single worker per time only while c++ could be used by multiple workers
22:14	sapier	as long as those c++ workers don't try to use same c++ tool handles
22:14	VanessaE_	at any rate, I better shut up now
22:15	sapier	well my suggestion is adding more lua tools and workers ... but to get this done you have to package work and pass it from one worker to another
22:16	twoelk	and have some coordinater with fixed rules ;-)
22:16	sapier	yes
22:16	sapier	but the additional lua tools can't do all of those things the current lua tool can do
22:16	sapier	so only special types of work can be packaged
22:17	twoelk	maybe that expertize is not needed for every job
22:18	sapier	that's what I'm hoping ... an doing it this way we can implement the additional tools quite simple and improve them step by step
22:18	twoelk	interacting with nodes may be different to just moving around
22:18	sapier	I assume this to be way less error prone then rewriting all at once
22:21	sapier	but first I wanna merge android port ;-)
22:22	twoelk	have let my niece try a debug apk version on her new smartphone
22:22	twoelk	water looked wierd
22:23	sapier	well android devices aren't as fast as pc's so all fancy stuff is turned off
22:23	twoelk	but movements looked a lot smoother than on my pc
22:23	sapier	you can disable all of those things in settings too ;-)
22:24	twoelk	let me reword: water looked so wierd I did not recognize it as water
22:24	sapier	it's not transparent there
22:25	twoelk	and strange color. not blueish at all but some dirty grey
22:25	twoelk	might be here device though
22:25	sapier	ok that's strange
22:27	twoelk	I dont use a smartphone, so I cant investigate and she is not into testing things
22:29		sapier left #minetest-dev
22:30	twoelk	I did notice though that I could hardly read any text. I would definitly have to change some settings concerning that if I were to play
22:30	twoelk	too late
22:43		EvergreenTree joined #minetest-dev
23:03		us`0gb joined #minetest-dev
23:33		lisacvuk joined #minetest-dev
23:39	hmmmm	the solution here is and always was RCU
23:40	hmmmm	the fundamental problem is that the map is this simply gigantic lock
23:40	hmmmm	and everybody needs it
23:46	VanessaE_	mmh
23:48	VanessaE_	well I guess as long as it's fast.
23:49	VanessaE_	(I had to look up RCU. I've used that back in the stone age, but we didn't call it that back then)
23:49	VanessaE_	(or something similar anyway)

← Previous day | Channels | #minetest-dev index | Today | Next day → | Google Search | Plaintext