Time  Nick         Message
08:46 sapier       thxyz what's typical size of that FMBitset?
08:47 sapier       with size I' meand number of elements in there .... and prior someone notes I know thexyz isn't here right now ;-)
08:48 VargaD       hi sapier
08:50 sapier       VargaD: how's your investigation progressing? :-)
08:52 VargaD       I have compiled freeminer, it is incredibly slow, just like minetest (I'm a 4 year old Atom N450 netbook because of christmas)
08:52 VargaD       but still more playable than minecraft :)
08:53 sapier       :-) yes we think about speeding up build from time to time but by now everyone was to frightended about even trying, especially as there ain't a guarantee it's even possible
08:54 VargaD       yes the build is also slow, but I don't think that it can be faster
08:54 VargaD       building C++ with optimastions is slow
08:54 sapier       reducing header coupling sometimes helps but it's a lot of work with limited result
08:54 VargaD       maybe building with clang make is a bit faster
08:55 VargaD       that is the easiest way :)
08:55 sapier       not maybe this is already prooven
08:55 sapier       it's about 1/3
08:56 VargaD       I don't think it is good to minetest to move reduce the size of headers
08:56 VargaD       it make many optmasatons impossible and LTO is still experimental
08:57 sapier       as I said noone is really working at this (as far as I know)
08:59 VargaD       What do you think about thexyz's performance improvement in ABM?
08:59 sapier       and header size reduction would be more decoupling minetest subsystem from each other adding defined interfaces between them ... minetest actually hase come far from the monolithic block it once was. reducing coupling is a open issue that has to be done by some time in order to cleanup basic design
09:01 sapier       seems interesting, I just wonder why he used a vector, as of speed pov a int array would be even faster, especially as char isn't native datatype on any of our supported architectures
09:02 sapier       of course this would add another 7 bytes of overhead
09:02 sapier       (on 64 bit os)
09:10 VargaD       I'm not convinced that it make minetest faster
09:11 VargaD       I still don't know exactly what does this code do
09:12 VargaD       We need to benchmark that code somehow, to decide which way is faster
09:12 sapier       if I understand thexyz correct he already did this
09:12 VargaD       using bitset istead of set should be much faster
09:14 sapier       that his first commit
09:14 VargaD       yes I see
09:15 VargaD       in the next commit he replaced it to his own bitset implementation
09:16 VargaD       the content size seems compile time constant, so it would be better to use a data structure that is allocated on stack
09:17 VargaD       less pointer use should make to code faster and do less allocations
09:17 sapier       yes that's the part I'm not conviced to be really faster too
09:17 VargaD       FMBitset could be a template with size given in template argument
09:18 sapier       templates aren't liked very much ;-) there are some things they're usefulll but at this very special location?
09:23 VargaD       how do you make a general bitset data structure that has compile time constant size?
09:24 sapier       I'm not sure it is required to be compile time constant
09:25 sapier       guess this is one of the times you have to decide general <-> speed ;-)
09:26 VargaD       How much ActiveABM do we allocate?
09:27 sapier       guess noone is using abms very much as they are known to be slow
09:27 sapier       but that's my personal view only I could be wrong about that
09:29 VargaD       so ActiveABM is somethink a mod can attach to some nodes?
09:30 VargaD       and it is also slow, so we allocate a few of them
09:30 sapier       yes
09:30 sapier       the problem starts once you attach a abm to a common node
09:30 sapier       e.g. dirt, cobble
09:31 sapier       as you can specify a chance and time this can kill performance
09:31 VargaD       when you attach ActiveABM to a common node that class allocated many times?
09:31 sapier       e.g. if you do it once a second with chance 1 on dirt ;-) 16x16x16 calls to that abm if a single block of dirt is active
09:32 sapier       usually there are more blocks active
09:32 VargaD       oh
09:33 VargaD       std::bitset has compile time size and also use less memory, trash cache less, maybe that is a good enough solution
09:34 sapier       possible but thexyz must have tested something, I'd like to know what he found out
09:41 VargaD       wow freeminer has a lot of commits notv merged to minetest
09:41 VargaD       Why do they forked minetest?
09:42 sapier       because proller didn't want to do the work necessary to clean up a commit for merge to minetest ... and to be fair he started his work at a time where minetest development slowed down due to lack of time of core devs
09:43 VargaD       it would be nice to unite development again...
09:43 proller      because average time of pull approve is half year
09:44 sapier       most of prollers commits where interesting, but most contained absolutely unrelated features within s a single pull request, often a usefull one and a controversial one in same
09:44 thexyz       yes what?
09:44 thexyz       std::bitset's count() is O(N), that's why I wrote my very own bitset
09:44 proller      sapier, because i try to do something instead of speaking
09:45 sapier       proller I wont explain this again I already told you community development is only about 50% of coding
09:45 VargaD       thexyz: oh I see
09:46 thexyz       if you feel like improving it then do that but since my profiler isn't complaining about this piece of code anymore and I don't see anything terribad about it I'm not feeling like changing this stuff to allocate on stack
09:46 sapier       thexyz why do you use a vector instead of allocating memory and access per []
09:46 VargaD       thexyz: how do you test performance?
09:46 VargaD       Do you have a mod that triggers this code?
09:47 VargaD       allocation memory and using vector isn't different in preformance
09:47 thexyz       no, I've tested this in a separate app and then just run through profiler with fix and without
09:47 VargaD       it is much safer to use vector
09:47 thexyz       to confirm that bitset's .count is indeed O(n)
09:48 VargaD       the only improvement I see is to make size compile time constant so it isn't allocated at all
09:48 thexyz       sapier: well I just felt like it, there's less code and no difference
09:48 VargaD       but maybe that isn't worth to hassle to use template
09:48 thexyz       VargaD: maybe, I didn't think about using a template
09:49 sapier       what complexitiy is [] for std::vector?
09:49 VargaD       O(1)
09:49 thexyz       constant
09:49 thexyz       what's with this question
09:49 sapier       hmm ok forget about the array :-)
09:50 VargaD       it is a template, the compiler generates exactly the same code, it isn't worth to wite code that handles memory allocations
09:50 VargaD       when you compile it is exactly the same
09:50 sapier       I assume a good compiler will do the int alignment too
09:51 VargaD       only when you ask it
09:51 VargaD       but alignment won't do much here
09:52 VargaD       this can't be vectorased anyway
09:52 sapier       reading a char is in worst case read, shift, mask
09:53 sapier       writing is even worse ... shift read, mask, set, write
09:53 sapier       but this is highly dependent on architecture and compiler
09:53 VargaD       I don't think that works like that, the cpu should handle that well, maybe it is worth to try
09:54 sapier       as I said this depends on architecture and compiler, I know for sure mips for example can't do single byte writes
09:55 sapier       at least the version I work with ;-)
10:00 sapier       wow x86 is quite comforable on this .. if compiler does compile sane code 1 2 4 8 are atomic (as long as aligned correctly for those >1)
10:02 VargaD       do we run minetest server on mips?
10:02 VargaD       :)
10:03 sapier       don't know if anyone has a mobile device ;-)
10:03 sapier       my phone is arm
10:03 VargaD       those are arm
10:03 VargaD       some routers are mips
10:04 sapier       yes mips isn't very common these days
10:04 VargaD       loongson is also mips
10:04 sapier       but usually we use x86 and there it's not an issue
10:09 VargaD       does freeminer and minetest world/network compatible?
10:10 sapier       world yes network ... I think they still are but freeminer may have switched to enet already thexyz can tell more about this
10:10 VargaD       they have a lot of interesting improvements
10:11 sapier       minetest decided to switch to tcp for reliable communication instead of using enet but right now noone is working at it
10:11 proller      .. and newer will..
10:11 VargaD       enet has a udp based reliable communicatino as far as I remeber
10:12 proller      ant its work!
10:12 VargaD       probably, many project using enet
10:12 VargaD       and they work fine
10:12 proller      minetest have wat wtf based communication
10:12 VargaD       that is quite a bad thing :(
10:13 VargaD       it would be nice to port enet to minetest
10:13 proller      my server sometimes resends 100+ pps
10:13 VargaD       sapier: I'm willing to do the port, but will you accept a clean pull request?
10:13 sapier       we discussed this about 100 times the last week I wont discuss again
10:14 sapier       it's not up to me to accept it but I'd start with something less controversial then enet ;-)
10:15 sapier       enet has some issues too ... no ipv6 support, compatibility break, one man show,
10:15 VargaD       oh I see
10:17 VargaD       it is a good idea to use TCP in a realtime game?
10:17 sapier       but enet works way better then current network code, yet breaking compatibility to anything we need to be sure it's the way to go for next years, obviously not everyone is convinced enet is way to go ... and no it's not only my opinion
10:17 sapier       thought is UDP/TCP combination
10:18 sapier       of course TCP only isn't an option
10:18 proller      need in UDP/TCP/SCTP/RTMP/etcMP combination!!11
10:18 proller      we need more protocols!
10:18 VargaD       SCTP isn't widely suported :)
10:19 proller      linux+*bsd okay
10:19 sapier       linux+windows are required  ... bsd macos androind are benefits we take if we have chance
10:20 VargaD       Why don't you use boost or C++11? Sorry for my silly questions sapier :(
10:20 proller      okay. we must implement gre! open ppp session to minetest server an make p2p net with other clients.
10:20 sapier       .... didn't read the bad b word ...
10:21 sapier       c++11 is beeing thought about but for now visual studio 2010 usage stops us from using it
10:21 sapier       proller plz write your tirades to #minetest
10:21 VargaD       yes visual studio is quite bad in C++11
10:23 VargaD       Thank you for your patience sapier
10:25 sapier       I try to help where I can varga but keep in mind minetest is a project with more then one core devs and there are different opinions about some issues
10:27 sapier       as you might have noticed networking code and pull quality are most controversial points right now, 3d engine and security have been others in past ... and using json for everything (maybe code too ;-P) is a all time favourite ;-)
10:29 VargaD       I see there are a lot of developers and many branches on github
10:29 sapier       as I said coding is at most 50% of time ;-) most time you will spend on defending and persuading ppl your variant is really best
10:30 VargaD       minetest is a remarkable application, I'm sure we can solve the issues
10:30 sapier       take it a neverending strugle for perfection ;-)
10:31 VargaD       of course but commits should be clean and perfect :)
12:29 * specing    digs extra:john_minetest
13:34 Jordach      wrong channel john.
13:59 sapier       you can see fps in one of the debug tabs john_minetest
13:59 sapier       it's a little bit hidden
14:02 sapier       where is a win32 build?
14:03 sapier       as far as I know we don't do official builds some "latest" in forum usually is a fake ;-) it's only latest for a small time ;-)
14:04 sapier       but as far as I know that fps fix doesn't have any effect on windows
14:04 * sfan5      puts a kitten on john_minetest's head
14:05 sapier       not the windows one ;-)
14:07 sapier       true
14:11 RealBadAngel if some1 wanna build something then maybe this repo? https://github.com/RealBadAngel/minetest
14:12 RealBadAngel theres code that generates normal maps on the fly
14:19 sapier       ok guys I started working at UDP/TCP protocol handling, I intend to write this in a 100% backward compatible way using dual protocol handling for transition time
14:43 VargaD       sapier: can I help you in implementing the new network protocol?
14:59 sfan5        sapier: does that mean if packetdata.startswith("\x4f\x45\x74\x03"): <old way> else: <new way> ?
15:09 sapier       no it means it will support old as well as new clients
15:10 sapier       transportation method is unrelated to packetdata on proper layered environments
15:10 sapier       thanks vargad but this task isn't big enough to be split in usefull parts
19:04 RealBadAngel sfan5,  can you make win build with this? https://github.com/RealBadAngel/minetest/commit/a48d0e9bed5d478e1b9c468484eb81420ffed28b
19:09 RealBadAngel hi PilzAdam
20:56 Calinou      there is some prediction stuff in Minecraft, but not much
20:56 Calinou      CraftBukkit is very server-side focused
20:56 Calinou      (Bukkit is an API.)
21:08 Calinou      the client can't get new models: they are all _hardcoded_ (yes)
21:08 Calinou      same for textures and sounds
21:08 Calinou      sounds can be played by server at various volumes and pitches, without requiring separate files
21:14 thexyz       yes
21:14 thexyz       also, google is your friend
21:15 thexyz       I see
21:15 thexyz       sadly this is not related to minetest core development
21:15 RealBadAngel afaik both client and server require mods
21:15 Calinou      google isn't your friend
21:15 * Calinou    rolls eyes
21:15 RealBadAngel so for modified server you need modified client
21:15 RealBadAngel with very same mods
21:17 RealBadAngel it may had changed already, im not playing mc for a long time
22:20 iqualfragile RealBadAngel: last time i looked modding support even in craftbuckit was not satisfying
22:29 RealBadAngel so anybody to test: https://github.com/RealBadAngel/minetest/commit/a48d0e9bed5d478e1b9c468484eb81420ffed28b.patch ?
22:43 kaeza        wish I could, but my crappy card does not support normalmaps :(
22:54 RealBadAngel kaeza, this code generates them on the fly
22:57 kaeza        http://pastebin.com/Z0MPAgUe
23:01 RealBadAngel git am ? why not git apply ?
23:03 RealBadAngel http://stackoverflow.com/questions/12240154/what-is-the-difference-between-git-am-and-git-apply
23:05 kaeza        do you have to enable anything besides "Generate normalmaps"?
23:05 kaeza        (and "Shaders")
23:05 RealBadAngel shaders, bumpmapping, generate normalmaps
23:05 RealBadAngel disable filtering (at least for 16x textures)
23:05 kaeza        nope, all black
23:06 RealBadAngel do you have any errors compiling shaders?