Time  Nick        Message
10:43 sapier      https://github.com/minetest/minetest/pull/724 added minizip support to mainmenu (NO libarchive)
10:44 sapier      I need the changes done for minetest dir detection as I don't want to redo this plz push those changes or at least tell me where to find them
14:16 kahrl       sapier: I'm currently rebasing a merge of your and my path function additions to master
14:16 kahrl       I have some questions about CopyDir
14:16 kahrl       is there a reason why it uses "r" and "w" as IO modes instead of "rb" and "wb"?
14:17 kahrl       also can it be fooled into an infinite recursion by symlink backreferences, and is that a problem?
14:21 kahrl       also is there any reason you used C style IO instead of streambuffers (not a problem, just wondering)?
14:23 sapier      aren't rb/wb same as rb/wb nowadays?
14:23 sapier      r/w
14:24 sapier      c style no special I just thought c++ classes are a little bit too big for that simple operations
14:24 kahrl       they are different on windows, unless that changed recently
14:25 kahrl       r converts \r\n to \n and w converts \n to \r\n
14:25 sapier      ok I'm not very common to windows development so if it's better to use rb use it
14:25 kahrl       microsoft fail: http://stackoverflow.com/questions/7250720/ofstreamoperatorstreambuf-is-a-slow-way-to-copy-a-file
14:25 kahrl       so stdio is good here
14:26 kahrl       if microsoft had implemented this well it would be as simple as destfile << srcfile.rdbuf();
14:26 kahrl       to copy a whole file
14:27 sapier      not much of a difference as you need to open it first and handle whatever exceptions that might occur if you use c++ code
14:27 kahrl       true
14:28 sapier      failing to install a mod shouldn't make minetest crash :)
14:28 sapier      is anyone working at mod handling?
14:29 sapier      mod functions are distributed throughout minetest which is a real pain
14:29 rubenwardy  what do you mean by "mod handling"?
14:29 rubenwardy  script api?
14:29 sapier      dependency management current modlist, mod datatype
14:29 sapier      s
14:30 sapier      no not the api
14:30 rubenwardy  a ha
14:30 sapier      the mod management
14:30 kahrl       sapier: I think there's a problem in the while(!feof(...)) loop
14:30 kahrl       in case an IO error happens, fread keeps returning zero
14:31 sapier      file line?
14:31 kahrl       in CopyDir
14:31 kahrl       filesys.cpp, dunno which line that is in whatever branch
14:31 sapier      found it
14:32 sapier      shouldn't fread return -1?
14:32 sapier      in case of error?
14:32 kahrl       man page says it returns 0
14:33 sapier      true but I don't se an error?
14:33 kahrl       return type is size_t so -1 doesn't exist
14:33 kahrl       well, if an IO error happens feof(sourcefile) will return 0
14:33 kahrl       so it becomes an infinite loop
14:34 kahrl       ferror(sourcefile) needs to be checked
14:34 sapier      oh true we need to check for ferror too
14:34 sapier      yes
14:34 sapier      do you fix it?
14:34 kahrl       yeah
14:35 kahrl       should the size of readbuffer be BUFSIZ?
14:36 kahrl       I heard it's an optimized value for this kind of operation
14:36 sapier      don't know what value is set there but I don't care about the exact buffersize either
14:39 sapier      doesn't seem to be very special to me ... I guess the optimal buffersize is to dependent on your os and hardwar
14:44 sapier      why does everyone do same thing different in mt community? serverlist is done in plain binary data modlist uses some strange forced json header ..
14:50 kahrl       i'm splitting of the file copy part because it's too deeply nested
14:50 kahrl       off*
14:51 sapier      no problem
14:51 kahrl       what do I call it? I was going for CopyFile but I realized that that is a bad idea
14:52 kahrl       perhaps CopyFileContents
14:54 sapier      why not CopyFile?
14:54 kahrl       CopyFile is a macro in windows.h
14:55 sapier      oh
14:55 sapier      MyCopyFIle? :-)
15:26 sapier      what about moving json code to separate files?
15:27 sapier      translating datastructures from json to internal representations lots of code that only makes functions less readable
15:31 kahrl       CopyFileContents now looks a bit ugly but should handle all error cases
15:31 kahrl       http://paste.dy.fi/5Yi
15:33 sapier      I'm not sure if this is correct
15:34 sapier      and the readbuffer within while is problematic too
15:34 sapier      if(readbytes != sizeof(readbuffer)) this is not a safe indicator for eof
15:35 sapier      I'm not sure if fread could be interrupted by an interrupt
15:35 sapier      returning a partial read in this case
15:39 kahrl       hmm the c standard says: "The fread function returns the number of elements successfully read, which may be less than nmemb if a read error or end-of-file is encountered."
15:39 kahrl       the standard draft I just downloaded, actually
15:40 sapier      I'm not sure but does this exclude temporary read errors?
15:40 kahrl       mathematically speaking it can be less in other situations too (it's an "if and only if"
15:40 kahrl       )
15:40 kahrl       it's not*
15:41 sapier      I'd feel more safe if there was used feof instead of implicit documentation knowledge
15:42 kahrl       I'll replace it with if(feof(sourcefile) || ferror(sourcefile))
15:42 kahrl       what's the problem about the readbuffer?
15:43 sapier      it's pushed to stack on each loop
15:43 kahrl       I don't think any modern compiler works that way
15:43 sapier      you did program it that way ;-) compiler needs to do what you say
15:44 sapier      -needs + has to
15:44 sapier      of course depending on optimization level different compilers will do different things
15:44 kahrl       I'm sure this is optimized even on -O0
15:44 sapier      I don't think so
15:45 sapier      you told him to create a new variable on each loop not doing this is an optimization
15:47 kahrl       it's an optimization that costs nothing, not even causing debug problems
15:48 kahrl       besides this function is IO bound anyway
15:49 sapier      no matter I'd not rely on compiler optimization if I don't have to
15:50 sapier      but of course thats an oppinion
15:55 kahrl       http://paste.dy.fi/5SE
15:56 kahrl       the subq $10016, %rsp is done outside of the loop even on -O0
15:59 sapier      with your gcc
15:59 kahrl       with any modern compiler
15:59 sapier      proove "any" ;-)
16:00 kahrl       define any compiler that doesn't do this as not modern ;)
16:00 Calinou_    inb4 clang
16:01 sapier      do what you can't avoid to do ... but don't blame a compiler doing what you tell him to do
16:01 kahrl       meh. I don't care enough about this to continue to argue
16:01 sapier      me too
16:02 sapier      i guess in 99.9999% of all cases both options will result in same binary code
16:10 kahrl       I just took the diff of two diffs
16:10 kahrl       would that be considered a second derivative?
16:11 sapier      lol
16:16 VanessaE    not to change the subject, but can someone please explain this:  http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2012:12:02%20PM.png
16:16 VanessaE    (namely the horrible performance in this scene - and this is a fresh sign-on, so this view is the only thing currently loaded)
16:17 PilzAdam    VanessaE, hows the same scene with fancy leaves disabled?
16:19 VanessaE    PilzAdam: about the same.  http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2012:18:26%20PM.png
16:19 VanessaE    1 fps difference.  and this is without aniso/mipmap/filters or shaders.
16:20 PilzAdam    well, turn the shaders on
16:20 Calinou     with shaders on I also get this problem PilzAdam
16:20 VanessaE    shaders are usually on.  they cost me a few more fps.
16:20 Calinou     25 FPS with shaders + AA + AF + filtering on with default pack
16:20 Calinou     (570 with nvidia 319.17)
16:21 VanessaE    there's a serious performance regression here.
16:21 Calinou     VanessaE: not a regression, it happens since always
16:21 VanessaE    Calinou: I used to get far better performance than what I get now with scenes like that ^^^^^
16:21 Calinou     ah
16:22 Calinou     I always had 30-ish FPS on your server spawn
16:22 VanessaE    (and for the record, I only displayed the F6 profiler for your reference, not mine)
16:22 PilzAdam    VanessaE, try reducing the nodeboxes in your mods?
16:23 VanessaE    PilzAdam: if you'll look at that scene, there aren't that many.
16:23 VanessaE    and only three low-poly objects by me are even visible there
16:24 VanessaE    (the two types of fences and the lights.  total of about 10 nodeboxes between those three nodes.
16:25 VanessaE    plus you can see moreblocks/stairsplus nodes in there, which as you know only consist of one or sometimes two nodeboxes per object.
16:25 VanessaE    the roofing on the grey stone building consists of four nodeboxes per tile I think.
16:25 VanessaE    no, make that two.
16:26 VanessaE    underground are two long columns of pipes, which consist of about a dozen nodeboxes apiece, but since they're not visible, they shouldn't be affecting the draw time.
16:27 PilzAdam    oh, they do AFAIK
16:27 VanessaE    why?
16:27 PilzAdam    the whole mapblock is rendered
16:27 VanessaE    that's STUPID
16:27 PilzAdam    make it better then
16:27 VanessaE    if I could, I would.
16:27 PilzAdam    if you could then you would know that its not stupid
16:28 PilzAdam    so stop judging things you dont understand
16:28 VanessaE    don't tell me I don't understand.  I was screwing around with 3d rendering stuff before Minetest ever existed.  On machines far less capable.
16:28 VanessaE    there's something going badly wrong here.
16:29 VanessaE    I'm just trying to catch it before it gets any worse.
16:29 PilzAdam    I would rather blame the nodeboxes
16:30 VanessaE    ok, then explain this:
16:31 VanessaE    wait, lemme get a better shot first.
16:32 VanessaE    http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2012:31:24%20PM.png
16:32 VanessaE    there are a metric shitload of nodeboxes in this shot.
16:32 VanessaE    far more than in the sub-30 fps shots
16:32 VanessaE    inside that green building is a large mesecons circuit.  Under the road, pipes.  alpha glass in the background.  signs all over the place.  glass here and there.
16:33 PilzAdam    how many nodeboxes do you have per pipe node?
16:33 VanessaE    lemme re-check
16:34 VanessaE    about 40 per.
16:34 VanessaE    (about half as many visible faces though)
16:34 PilzAdam    and how many pipe nodes were rendered?
16:34 VanessaE    (because many of them are overlapped)
16:34 VanessaE    in that last shot?
16:34 PilzAdam    in the 28 FPS
16:35 VanessaE    in the 28 fps shot there are probably 130 pipes or so, none visible.
16:35 VanessaE    in the 48 fps one about the same number.
16:35 PilzAdam    so, 5200 nodeboxes
16:36 VanessaE    roughly.
16:36 PilzAdam    you want to say me that in the 48 FPS shot are more?
16:36 VanessaE    about the same number - I know what I built under that road.
16:37 kahrl       if stuff is under the road it might be occlusion culled
16:38 PilzAdam    VanessaE, the road are nodeboxes, right
16:38 VanessaE    kahrl: then why this?  http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2012:03:47%20PM.png
16:38 VanessaE    PilzAdam: the sidewalks and roughly 5% of the asphalt.
16:39 kahrl       VanessaE: note the larger number of blocks drawn
16:39 VanessaE    kahrl: yes - but note it's a shot of the sky
16:39 VanessaE    why is it drawing all the stuff that's off-screen?
16:39 kahrl       nearby blocks are always drawn, and occlusion culling isn't always exact
16:40 VanessaE    then I submit that the occlusion culling is broken :)
16:53 VanessaE    kahrl: isn't there *something* that can be done?  It didn't used to be this bad.
16:54 kahrl       I don't remember any recent changes in the occlusion culling code
16:54 PilzAdam    there was a fix in isBlockInSight() method
16:55 kahrl       oh, right
16:56 kahrl       VanessaE: try reverting 1369503aba and compare
16:58 VanessaE    Another comparison:  51fps, sure - but with 223 vertices being drawn when there are precisely 2 vertices visible.  http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2012:56:29%20PM.png
16:59 VanessaE    223k drawn that is
16:59 VanessaE    kahrl: ok
17:01 VanessaE    compiled, waiting for the client to load the map around it
17:02 VanessaE    that's actually slightly worse.
17:03 VanessaE    http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2001:02:28%20PM.png
17:03 VanessaE    I entered the world in exactly the view shown here, so there was never an opportunity to see anything but the one node I'm looking at there.
17:10 VanessaE    strange, shouldn't the mesh processing basically die down after a couple of seconds?
17:10 VanessaE    hm, there is something causing the mesh code to run while near the spawn, but I see nothing changing
17:11 VanessaE    (except the clouds of course)
17:11 VanessaE    num_processed_meshes keeps jumping between 0 and 5 or 6 every couple of seconds
17:11 kahrl       ABMs?
17:12 VanessaE    dunno, I don't see anything changing other than a player that just ran past me
17:12 VanessaE    I know there are abms that run the pipes
17:12 VanessaE    but those just replace one node with another
17:13 kahrl       that causes a mesh update
17:13 kahrl       sapier: I pushed my stuff here: https://github.com/kahrl/minetest/commits/pathfunctions
17:14 sapier      thanks I'll merge to luamainmenu
17:14 VanessaE    I've noticed mesh updates lately have gotten really laggy - they spike the client's draw time
17:14 VanessaE    others have reported the same thing
17:14 VanessaE    hmmmm mentioned it as well
17:17 PilzAdam    AFAIK hmmmm said that mesh genration is genereally slow
17:17 PilzAdam    not that it become worse
17:17 VanessaE    yes
17:18 VanessaE    and I've noticed it getting worse overall.
17:57 VanessaE    oh wow, severe entity duplication bug in pipeworks
17:59 VanessaE    can someone try this circuit please and tell me if it's my code or the engine's entity duplication bug?
18:01 VanessaE    http://digitalaudioconcepts.com/vanessa/hobbies/minetest/screenshots/Screenshot%20-%2006102013%20-%2001:59:26%20PM.png
18:01 VanessaE    every time an item hits a split in a pipe, it is duplciated
18:01 VanessaE    duplicated*
18:01 VanessaE    and it's replicated several times on exiting a pipe.
18:01 VanessaE    this code has not been changed in months
18:19 VanessaE    in fact it seems like any time an item even changes direction at all
18:46 sfan5       celeron55: In case you care: I got minetestserver running on android, chat is buggy (probably because wchar_t is differnt on android) and commands don't work, but gameplay itself should completly bugfree
18:56 VanessaE    actually scratch that thing about the duplication, it's not a bug.  it's something nore did in my code.
18:56 VanessaE    s/nore/Nore/
18:57 ShadowNinja VanessaE: Has he fixed my duplication bug yet?
18:59 VanessaE    dunno
19:00 VanessaE    how does that one manifest itself?
19:01 VanessaE    ->$minetest
19:01 VanessaE    ->#minetest
19:16 sapier      what do you think about this gui for online modstore? http://imgur.com/4h7nrGA
19:17 PilzAdam    there are buttons missing to switch the page
19:18 sapier      no ^ and v are only shown if you're not already at top/bottom ... but I guess this looks strange somehow
19:18 sapier      the "scrollbar" ;-)
21:40 Exio        sfan5: what libc are you using?