Time  Nick       Message
01:15 paramat    game#1450 bugfix for missing sound
01:15 ShadowBot  https://github.com/minetest/minetest_game/issues/1450 -- Default: Add missing 'default_dig_snappy' sound by paramat
08:23 lhofhansl  Just filed #4877
08:23 ShadowBot  https://github.com/minetest/minetest/issues/4877 -- Limit Active Block Range to what player can see. Retrieve a spherical volume. by lhofhansl
08:23 lhofhansl  another optimization based on the player
08:24 lhofhansl  the player's view range. This time of about active blocks.
08:24 lhofhansl  s/of/it's/
08:25 hmmmm      http://dev.minetest.net/Code_style_guidelines#If_statements
08:27 hmmmm      as for the actual PR
08:28 hmmmm      i don't feel like it's an optimization, this is just making the active block range shorter.  that causes different problems..  a better optimization would be to make running ABMs less costly, so that larger ranges can be used
08:29 lhofhansl  OK... "if (range <= 0) range = active_block_range" is really the same as "range = range <= 0 ? active_block_range : range" so it seems one line is fine. Can do two lines.
08:29 hmmmm      for those types of statement MYMIN(), MYMAX() or rangelim() would be preferred
08:29 hmmmm      statements
08:29 lhofhansl  hmmmm: it lets you increase active_block_range and still get decent behavior with clients that do not use the full range.
08:30 lhofhansl  right now active_block_range defaults to 2 which is quite small. With that one can set it higher.
08:30 hmmmm      the client sets the active block range?
08:31 lhofhansl  no the server does. Now that value can be a increase a bit.
08:31 lhofhansl  It's weird when active objects just disappear when you just 40 or so blocks away.
08:32 hmmmm      set higher
08:32 lhofhansl  this aligns it with what you can actually see on the client, up to the setting.
08:32 hmmmm      because with a spherical range it lops off the edges
08:32 hmmmm      ?
08:32 lhofhansl  Same as #4811 which did this for active objects only.
08:32 ShadowBot  https://github.com/minetest/minetest/issues/4811 -- Optimize/adjust blocks/ActiveObjects sent at the server based on client settings. by lhofhansl
08:32 hmmmm      if you increase the range to 3 or something that's still going to be strictly greater than a cuboid of 2
08:33 hmmmm      well that PR is different from this PR
08:34 lhofhansl  Yep. There're parts: (1) retrieve a sphere (saves about 40% of objects to process), and (2) take the clients view range into account and do not process things that client cannot see anyway.
08:34 hmmmm      the ABM catchup code doesn't account for blocks outside of the active block range, does it
08:34 hmmmm      only blocks loaded from an unloaded state
08:34 lhofhansl  right
08:35 hmmmm      so your chief complaint was that the client would be able to see a block become inactive even though it's still within sight range
08:35 lhofhansl  so active_block_range is upper limit set by the server. If the client sees less it's not processed.
08:36 lhofhansl  Yep... And there's no one good setting for the server, because it really depends on how the client can see things.
08:36 hmmmm      okay so this strictly is an optimization in that kind of sense
08:36 lhofhansl  So either things disappear or they are unnecessarily processed.
08:37 hmmmm      for block sending, active object sending, and now you want active block modifiers, you want it so that if the player can't see it, you don't need to do it
08:37 hmmmm      whereas if the client has a view range greater than the max set by the server config, they can still see it
08:37 hmmmm      but what about latency
08:37 lhofhansl  no the server sets the max.
08:37 hmmmm      players can move around quite fast
08:38 lhofhansl  client can only reduce it
08:38 hmmmm      so if you have a max active block range of say 3
08:38 hmmmm      and the player has a viewing range of 1
08:39 lhofhansl  same problem as it is now only that the server is "hardcoding" the setting for all clients regardless of how far they can see stuff.
08:39 hmmmm      and the player moves really quickly a block in some direction
08:39 lhofhansl  (it's map blocks, but yeah same point)
08:39 hmmmm      because of the optimization, abms will not have run for the player within that frame of time and there'd be the latency in getting it to start running again
08:39 lhofhansl  that's how it would work, yes
08:40 hmmmm      it's an unfortunate side effect
08:40 lhofhansl  agreed
08:40 hmmmm      but the real problem exists in the slow block sending and ABM running
08:40 lhofhansl  the ABM might be in Lua. can be arbitrarily slow.
08:40 hmmmm      oh of course
08:40 hmmmm      but that's capitalism for mods
08:40 lhofhansl  :)
08:41 hmmmm      they need to stay fast or else they'll get overtaken by the competition
08:41 lhofhansl  At least we should process stuff in a sphere only, not the cuboid.
08:41 hmmmm      or at least in theory... in reality it's still kind of difficult for server administrators to see which mods are causing the lag
08:41 hmmmm      anyway ABMs are usually not that slow
08:42 hmmmm      but what is slow is the node scanning
08:42 hmmmm      iirc i profiled this a while back and the node scanning (which is not lua) takes up a solid 55% of cpu time
08:42 lhofhansl  it you set active_block_range to 5 or 6 it gets quite slow I find.
08:42 hmmmm      but that's with a vanilla minetest_game
08:42 hmmmm      55% of cpu time spent in ABM processing* to clarify
08:43 lhofhansl  If we retrieve a sphere at least, we can default active block range to 3 with about the same number of objects processed.
08:44 lhofhansl  maybe that's all I should change, and get rid of the view_range stuff.
08:44 hmmmm      at such a small radius, these spheres really look like diamonds
08:44 hmmmm      you might be chopping off more blocks than you think
08:44 hmmmm      small integer radius anyway
08:44 lhofhansl  I counted the blocks. Does 93 instead of 125 with the default of 2.
08:45 hmmmm      that's a respectable difference
08:45 hmmmm      125 is the amount with radius = 2, right?
08:45 lhofhansl  yep
08:45 hmmmm      because it's 2+1+2 = 5 for each direction
08:45 hmmmm      right ok
08:46 lhofhansl  exactly 5^3
08:46 hmmmm      mmm
08:46 lhofhansl  the sphere is a bit less than 4/3*pi*3^3 in that case.
08:46 lhofhansl  (due to the <= check)
08:47 hmmmm      you did the same if (foobar) foobar = 5; thing in the other PR
08:47 hmmmm      dammit
08:47 lhofhansl  yeah
08:47 lhofhansl  happy to change it
08:47 hmmmm      it's not really a big enough deal to warrant a new PR
08:48 hmmmm      it'll get cleaned up in somebody's omni-cleanup commit
08:48 hmmmm      also what's up with making POD typed parameters "const"
08:48 hmmmm      ...did nerzhul tell you to do that?
08:49 lhofhansl  somebody commented on that PR to that extend.
08:49 lhofhansl  forgot who
08:49 hmmmm      that's dumb
08:49 hmmmm      it's nerzuhl
08:49 hmmmm      he can't contribute by doing real reviews so he focuses on trivial personal preferences
08:50 hmmmm      i don't review often because reviewing takes real time and effort if you want to be worth more than a static code analyzer
08:50 lhofhansl  what's dumb about it?
08:51 OldCoder   hmmmm, and yet it's necessary
08:51 hmmmm      because who cares if you modify the formal parameter of a function
08:51 hmmmm      nobody is getting hurt by it
08:51 hmmmm      the only time you need to use the const qualifier (with our code standards anyway) is if passing a reference
08:53 lhofhansl  went by the style of the methods around. don't really care
08:53 hmmmm      i feel like i'm wasting my time by discussing such trivial code style matters tbh
08:53 hmmmm      maybe there should be a script that checks submissions
08:53 hmmmm      so i don't need to point anything like this out and instead work on logic
08:54 lhofhansl  it's not wasted time, I think.
08:54 lhofhansl  With a reasonably sized code base these thing become important.
08:56 hmmmm      well i did find at least one real error in your PR
08:56 hmmmm      static const blah foobar = g_settings->getBlah();   <--- see any problems?
08:56 lhofhansl  That's what we reviews for :)
08:56 lhofhansl  but that line I just moved :)
08:57 hmmmm      touche
08:57 hmmmm      you could fix a bug with this PR as well then
08:57 lhofhansl  yep
08:58 lhofhansl  wait... when can this be changed. Not in-game, right?
08:58 hmmmm      no
08:58 hmmmm      hmm
08:58 lhofhansl  I was also told reading from GlobalSettings is slow (probably not an issue here, since it's called once/sec)
08:59 hmmmm      right, it's only inside of loops
08:59 hmmmm      that's the case you should worry about
09:00 lhofhansl  alright... very late here... need to catch some sleep.
09:00 hmmmm      you can modify active_block_range from within a mod or by plizadam's everything-config-page
09:01 hmmmm      with the former you can definitely expose this bug
09:01 hmmmm      ahh no it happens with the latter as well
09:01 hmmmm      it has to modify the active settings in order to get saved to the config file
09:02 hmmmm      so yeah.  in two cases you'd be able to modify active_block_range, restart the game, and the old setting will be in effect until you actually close the application and restart.
09:03 lhofhansl  I'll fix that on the way... also adjust the layout of the ifs.
09:03 hmmmm      sure
09:04 hmmmm      you're also missing a const on a parameter passed as a reference btw
09:05 lhofhansl  for the &ranges?
09:05 hmmmm      yes
09:05 hmmmm      you're not modifying ranges, are you?
09:06 lhofhansl  Again I went by the surrounding code. Not sure what the wanted style is when everything around it does it differently
09:06 hmmmm      good thing there's a wiki page for it http://dev.minetest.net/Code_style_guidelines
09:07 lhofhansl  I'll fix that too.
09:07 lhofhansl  OK... TTYL... Need to sleep.
10:47 Wuzzy      Does the Feature Freezeâ„¢ imply a string freeze?
12:06 Wuzzy      Are questions about strings (for translation) on-topic here?
12:40 celeron55  has something changed in occlusion culling?
12:40 celeron55  because i think this is way more broken than i remember this being
12:41 celeron55  i just dug a straight 3x4 tunnel and MT absolutely refuses to render it
12:42 celeron55  this is stupid
12:45 sfan5      there were some changes afaik
12:46 sfan5      https://github.com/minetest/minetest/commit/2b21cac1d80454bd23c5f60c4570b30edba62584 this?
12:49 celeron55  seems to be broken before that too; i guess this is just a very unfortunate place for a tunnel
12:51 celeron55  i wish there was a setting to just plain disable occlusion culling if i want to
12:54 celeron55  well this is easy to add for sure...
12:55 jin_xi     50 MESE blocks on fps gain
12:58 celeron55  well in what i'm doing it's 60fps with, 60fps without
13:00 celeron55  if i run on my dedicated card which allows more than 60, it more than halves the fps (from 130 to 60)
13:01 celeron55  doesn't matter
13:01 celeron55  i just want to see this damn tunnel
16:05 VanessaE   celeron55: you sure it isn't just your view range being too short?
16:06 celeron55  that's not a relevant question when just disabling the culling already fixed it
16:06 * VanessaE shrugs
16:09 Fixer      probably celeron selected unlucky spot for tunnel that is prone to occlusion mismatch
16:18 lhofhansl  Updated #4877, I think there's some misunderstanding around it.
16:18 ShadowBot  https://github.com/minetest/minetest/issues/4877 -- Limit Active Block Range to what player can see. Retrieve a spherical volume. by lhofhansl
16:22 sfan5      spherical sounds fine, increating the active block range would be wonderful
16:22 sfan5      being able to increase*
16:23 lhofhansl  Maybe I split it into two. Do the spherial in one (which also increased active_block_range's default to 3 map blocks.
16:23 lhofhansl  In the other I do the view-range. Then you can shred the 2nd to pieces :)
18:10 lhofhansl  So is there interest in just doing the spherical part of #4877?
18:10 ShadowBot  https://github.com/minetest/minetest/issues/4877 -- Limit Active Block Range to what player can see. Retrieve a spherical volume. by lhofhansl
18:17 sfan5      i would +1 that
18:36 lhofhansl  See #4881 (also increases active_block_range, so that may cause some discussion)
18:36 ShadowBot  https://github.com/minetest/minetest/issues/4881 -- Process AMBs in a spherical volume instead of a cuboid by lhofhansl
18:36 VanessaE   what's an AMB? :)
18:37 lhofhansl  Oops :)
18:41 sfan5      lhofhansl: two minor things: nodes not blocks, you can change any settings at runtime via /set
18:41 sfan5      the question is whether it makes sense for setting changes to go into effect immediately
18:42 est31      the problem is that it cant be asked for every server step
18:42 est31      because retrieving settings is very slow
18:42 est31      the string is not interned
18:42 lhofhansl  that's just called once/sec, though.
18:44 juhdanad   By the way, there are two kinds of ABM: the ones which have a small chance and the ones which are always happen (for example lava cooling).
18:45 juhdanad   The latter should be implemented with node timers IMO.
18:45 lhofhansl  not emotionally attached to the static const change :)
18:46 sfan5      node timers are specific to a single node though
18:47 sfan5      you'd need to add/remove them everytime the liquid moves
18:47 lhofhansl  any concerns about increasing active_block_range to 3. Even with the other changes it will double the active blocks to process. (I think that's fine, but wanted to call to that out)
18:47 lhofhansl  s/blocks/nodes/
18:48 sfan5      no no
18:49 sfan5      active_block_range is in blocks
18:49 lhofhansl  Still mostly new Minetest.
18:49 sfan5      but in "A 5x5x5 cube (range = 2) has 125 blocks." it's nodes
18:49 lhofhansl  I see.
18:49 sfan5      http://dev.minetest.net/Basic_data_structures
18:51 lhofhansl  Oh OK. I have been referring in my mind to these as blocks and map blocks. Will use nodes.
18:51 juhdanad   sfan5: what I meant: lava flows->lava gets an update->if there is water, calls minetest.after(0) or sets a timer to set itself to stone.
18:51 lhofhansl  the 5x5x5 are map blocks, though, no? The range is given in map blocks.
18:52 juhdanad   water flows->water gets updated->lava gets updated because a neighbor changed->lava transforms itself
18:52 sfan5      lhofhansl: one map blocks is 16x16x16
18:52 sfan5      active_block_range is the range in mapblocks
18:53 sfan5      juhdanad: you would still need to remove add node timers everytime something changes with the liquids
18:54 lhofhansl  right, in that loop we are enumerating map blocks.
18:54 sfan5      yes
18:54 juhdanad   It is still faster than scanning trough all nodes in each second.
18:54 sfan5      how do you know?
18:55 lhofhansl  (anyway, not that important, I get the difference between a node and a map block, had to deal with that in #4811 and #4724, just didn't use the right term here)
18:55 ShadowBot  https://github.com/minetest/minetest/issues/4811 -- Optimize/adjust blocks/ActiveObjects sent at the server based on client settings. by lhofhansl
18:55 ShadowBot  https://github.com/minetest/minetest/issues/4724 -- Improve number of blocks sent and occlusion culling. by lhofhansl
18:56 juhdanad   I only think it can be faster. It is very rare that more than one sixth of all loaded nodes are actually changing liquids.
18:57 juhdanad   And you only need to check one node timer per tick: the one which will be first executed (so you can sort them by execution time).
18:58 juhdanad   So it is effectively checking one node per program cycle if nothing changes.
18:58 lhofhansl  juhdanad: what about you #4847? Seems to be an important fix.
18:58 ShadowBot  https://github.com/minetest/minetest/issues/4847 -- Fix unnecessary block loading by juhdanad
18:58 sfan5      it's checking zero nodes if a node timer doesn't fire
18:59 juhdanad   And ABMs not only check all nodes, but their neighbors too.
19:00 juhdanad   This is why I think converting to node timers might worth a try.
19:02 lhofhansl  I found when active_block_range is larger ABMHandler.apply becomes the most expensive operation. That's what prompted me to think about the sphere volume and to limit by view-range.
19:02 juhdanad   lhofhansl: it is only important if you are exploring. If all blocks are generated around you then it doesn't make a difference.
19:02 juhdanad   I mean #4847 ^
19:03 ShadowBot  https://github.com/minetest/minetest/issues/4847 -- Fix unnecessary block loading by juhdanad
19:03 lhofhansl  I see. Will the reloading also create new active objects as Rogier-5 suspects?
19:04 sfan5      it shouldn't
19:04 sfan5      however if the objects fall out of the world they might not get deleted when the block is reloaded
19:04 juhdanad   I'm testing it right now.
19:04 sfan5      i doubt minetest forgets to delete objects when (re-)loading a block
19:04 lhofhansl  OK. Then it does no harm other than increase disk reads (and those would likely come from the page cache anyway)
19:12 juhdanad   I found it! When the server loads a mapblock, it does not clear static objects.
19:12 sfan5      it doesn't do that at all or just in this specific case?
19:13 juhdanad   So because not generated blocks with items were loaded multiple times, the loaded items were duplicated.
19:13 juhdanad   sfan5: I don't understand your question.
19:14 sfan5      does it never clear static objects?
19:14 sfan5      or does this only happen at the world border?
19:15 juhdanad   It never clears objects. You can have duplicated monsters if they happen to be in a half-generated area.
19:16 juhdanad   I'll push a fix to #4847 soon.
19:16 ShadowBot  https://github.com/minetest/minetest/issues/4847 -- Fix unnecessary block loading by juhdanad
19:29 juhdanad   Pushed!
19:34 lisac      Hello. Is there currently a feature to make gravity pull towards a block in MT?
19:35 sfan5      no
19:35 lisac      :(
19:35 lisac      Thanks.
19:35 juhdanad   Should the player walk on the side or the bottom of the block, then?
19:36 lisac      Well, I just saw this mod https://forum.minetest.net/viewtopic.php?f=9&t=15933 and wondered if one could make it possible to actually walk on the planets.
19:37 juhdanad   Well, I recommend Spheretest then: https://www.youtube.com/watch?v=ztAg643gJBA
19:39 garywhite  Question: Jeija made that, and although it says about MC, it looks like MT textures
19:40 juhdanad   This is a Minetest fork, the title means 'Minecraft-like world'
19:42 garywhite  ok hang on, does that mean that if it's still basically Minetest under the hood, does that mean that someone could create a Spheretest world that MT clients can join?
19:43 juhdanad   It is Minetest, just the client has a different renderer.
19:44 juhdanad   Oh, and the world is finite.
19:44 garywhite  Hang on, it's just a copy of 0.4.14-dev with a demo world
19:44 garywhite  0.4.14-dev with the old MSVC renderer
19:45 garywhite  & it looks like a freakin cartoon
19:46 juhdanad   Did you follow readme.mt?
19:46 garywhite  no
22:53 Fixer      !tell paramat are you planning to add flowing water sound to that fire sound mod?
22:53 ShadowBot  Fixer: O.K.
23:11 paramat    maybe
23:46 paramat    #4847 could perhaps be considered for release as it fixes 2 bugs
23:46 ShadowBot  https://github.com/minetest/minetest/issues/4847 -- Fix unnecessary block loading and duplicating objects by juhdanad
23:48 Fixer      managed to cought reproducable crash after some --- with saved earlier map
23:49 Fixer      will make backtrace
23:54 Fixer      have we changed mapblock format recently?
23:54 paramat    i doubt it
23:55 Fixer      seen error about mapblock format: you probably made map in never version of minetest
23:55 Fixer      2016-12-11 01:55:26: ERROR[Main]: ServerError: World data version mismatch in MapBlock (0,0,11)
23:55 Fixer      2016-12-11 01:55:26: ERROR[Main]: World probably saved by a newer version of Minetest.
23:56 Fixer      indeed i played it in never version
23:57 Fixer      will post a new crash issue not related to this