Time  Nick       Message
02:28 ANAND      I'm going to definitely try to implement SSCSM - it's an awesome feature and I don't want it to die.
02:28 ANAND      I noticed that nerzhul has made some progress in this direction (https://github.com/nerzhul/minetest/tree/csm_mod_sending)
02:29 ANAND      Am I allowed to work on top of his code?
04:01 ANAND      Hmm... not much progress has been made on his branch, unfortunately
04:01 ANAND      https://github.com/nerzhul/minetest/compare/077f231...a1dd0d3
05:07 ANAND      Added `set_fov = true` to `minetest.features` in #7557
05:07 ShadowBot  https://github.com/minetest/minetest/issues/7557 -- Expose player FOV to Lua API by ClobberXD
07:16 nerzhul    ANAND good new to hear
07:16 nerzhul    if you have a design proposal before code propose it and i will amend
07:17 nerzhul    my work is very very raw
09:22 ANAND      Thanks for the encouragement ;)
09:23 ANAND      I have no idea whatsoever regarding the implementation details. Perhaps I might get a better outline once I understand the all mod-related code
09:27 nerzhul    the main problem is the serialization and transport of mods
09:41 ANAND      True
09:42 pmpp       aren't in memory mod data structures to be replicated on client and server ?
09:42 pmpp       i asked about rpc earlier and i think it's maybe related
09:55 ANAND      Regular mods are purely server-side, and the client doesn't have any knowledge about them.
09:55 pmp-p      so csm mod transfer does not include any gamestate ?
09:56 ANAND      Oh I thought you were talking about regular mods, sorry
09:57 pmp-p      no i'm interested on client side only for now, server looks fine to me
09:57 ANAND      At present, the plan is to serialize the mod's files and send them over to the client.
09:57 pmp-p      as bytecode or source ?
09:57 ANAND      Probably as bytecode, for security reasons
09:59 ANAND      SSCSMs could use mod-channels (with convenience wrappers) behind the scenes for communication with their server-side counterpart.
09:59 pmp-p      i've read somewhere that luajit has machine specific features, won't it impact on bytecode used ? ( i don't know lua yet )
10:00 ANAND      Not sure about that...
10:00 ANAND      It shouldn't affect the mods
10:01 pmp-p      as i'm currently developping a stackless vm for python bytecode i'm very interested in projects having already to move both bytecode and gamestate over the wire to see how it's done
10:03 pmp-p      i like the lua vm because it can be preempted too
10:03 ANAND      Likewise. I'm interested to see how other projects implement mod sending without any problems.
10:05 ANAND      Stuff like endianness and text encoding might pose a problem in this regard.
10:06 pmp-p      well i've seen yet quite a lot about gamestate serialization and i could say i retain Arrow or Capnproto but it's still a bit far from a running along a VM+bytecode
10:06 pmp-p      they say they are machine independant data structure but i've yet to see how it fits
10:11 pmp-p      and as client VM must be very well sandboxed, i was asking for RPC system used actually (if any as wiki is not clear about that)
10:17 p_gimeno   I don't see any gain in sending bytecode over source, and lots of inconveniences
10:18 pmp-p      i tend to agree, but bytecode does not need to compile on client side
10:18 pmp-p      and it could be faster to transmit
10:19 pmp-p      ideally both should be available
10:19 p_gimeno   Lua is quick to compile. Lua and LuaJIT bytecode are incompatible. LuaJIT 32 and 64 bit bytecodes are incompatible.
10:19 pmp-p      source while vm is not 100% safe, bytecode when optimized later
10:20 pmp-p      p_gimeno: thx for confirming that, i was not sure about portability of lua bytecode
10:20 p_gimeno   Lua-to-Lua is OK. LuaJIT-to-LuaJIT would be OK, except for the 64-bit transition.
10:22 pmp-p      why not something like lua-wasm ? wasm is fast and makes it easy to sandbox
10:22 p_gimeno   But Minetest clients can use either. I believe Windows ships with Lua by default and others with LuaJIT, I may be wrong though.
10:23 p_gimeno   Sending it gzipped would reduce bandwidth.
10:25 p_gimeno   Lua does not compile to wasm.
10:26 pmp-p      yeah not lua-jit but standard lua 5.3 seems to
10:26 p_gimeno   MT uses 5.1 only
10:26 p_gimeno   (for compatibility with LuaJIT)
10:26 pmp-p      they are so different ?
10:27 p_gimeno   yes
10:27 pmp-p      :(
10:27 sfan5      pmp-p: even the server sandbox does not allow bytecode for security reasons
10:27 sfan5      there is no way we would allow this on the client
10:28 sfan5      p_gimeno: the windows build ships with luajit since a few years
10:28 pmp-p      sfan5: i agree but some modders in other games would not want to share code bits sometimes
10:29 pmp-p      minetest pleased me because it does not look "closed"
10:29 sfan5      you can still obfuscate code
10:30 sfan5      compare this with javascript, nobody has a problem there either
10:30 p_gimeno   sfan5: I said it reversed. Windows with LuaJIT and others with Lua.
10:30 pmp-p      dunno i try to never do javascript :p
10:30 p_gimeno   Not sure about the others though.
10:32 sfan5      p_gimeno: that would also be wrong since at least debian, ubuntu and arch ship with luajit
10:32 p_gimeno   ah okay, that's what I was not sure about, thanks
10:32 pmp-p      indeed i had to choose beetween luajit5.1 and lua 5.2 or 5.3
10:41 p_gimeno   LJ is THE choice. Sadly the 64 bit version has not been officially released, and MT developers are reticent to ship it by default. But LJ is the fastest interpreter out there, and many servers depend on its speed.
10:42 pmp-p      yeah for servers it makes perfect sense, not sure about the client vm for mods though
10:42 p_gimeno   (by fastest, I mean among open source interpreted languages)
10:43 pmp-p      is performance so important on the client ( which is expected to give full power to the game ) ?
10:46 p_gimeno   I can't answer that, but having a different interpreter for the server and for the client makes little sense to me, and unless explicitly disabled in compilation, the client includes a server.
10:49 pmp-p      depends on if client is doing physics or the server i guess
10:51 rubenwardy I think that LuaJit is not-secure, so probably shouldn't be used for the client-side VM
10:51 p_gimeno   not secure in what sense?
10:52 rubenwardy trying to find where I read i
10:53 pmp-p      https://github.com/GoogleBot42/IronBox claims to be a sandbox for luajit
10:53 pmp-p      but it's unclear about jit compilation
10:53 rubenwardy our client VM kinda sucks in that there's no isolation from the Minetest process
10:53 p_gimeno   FFI is not secure, but FFI needs 'require'. There is a hashing attack that affects both Lua and LuaJIT (and I believe similar attacks apply to Python).
10:56 pmp-p      seems the safest is also the slowest...
10:56 p_gimeno   yeah, securing things is slow :)
10:57 p_gimeno   The attack applies to any hash algorithm which does not have cryptographically secure randomization, and merely consists of deliberately causing collisions in order to slow down the VM or possibly cause a memory overflow.
10:58 pmp-p      p_gimeno: i would not chose ffi for cpython it *is* slow, but would compile code instead with nuitka+cython but same as luajit that would not be so portable
11:02 p_gimeno   I'm not aware of any security problem with LuaJIT vs. Lua other than through use of FFI.
11:09 p_gimeno   btw rubenwardy I've been thinking about #7621. My thought is that it should not be called 'require', it should rather be a Minetest-specific function that is mod-aware.
11:09 ShadowBot  https://github.com/minetest/minetest/issues/7621 -- Add secure require() function by rubenwardy
11:10 rubenwardy that defeats the whole reason I'm adding it
11:10 rubenwardy so that you can embed pure Lua libraries in builtin or a mod without having to modify them
11:10 rubenwardy because Lua libraries use `require()`, not dofile
11:11 p_gimeno   ah
11:11 rubenwardy it's not for loading mod files
11:11 p_gimeno   got it
11:15 p_gimeno   I think MT needs a new mod API
11:18 rubenwardy lol
11:19 p_gimeno   it wouldn't be too hard to do on top of what there is now actually
11:19 p_gimeno   an object-oriented API where you do: local my_mod = minetest.register_mod('my_mod')
11:20 rubenwardy OO in Lua is noticiably slower for not much real benefit
11:20 p_gimeno   then all calls would go through it: my_mod:setpos(...) etc.
11:20 rubenwardy for certain things, there is a very real benefit, eg: https://github.com/CapitalismGame/capitalism_game/blob/master/mods/capitalism/company/company.lua
11:21 p_gimeno   well, for one it would allow a very clear distinction of what mod called a particular API function
11:21 rubenwardy oh, by OO I guess you mean giving each mod an API table which it calls with :?
11:21 p_gimeno   yes
11:21 rubenwardy that's not that much slower
11:22 p_gimeno   it would also get rid of the need of using globals to share a mod's API functions
11:27 p_gimeno   ... as well as get rid of using get_modpath just to detect if a mod is installed
11:44 sfan5      p_gimeno: actually, I accidentially shipped gc64 luajit with 5.0
11:44 sfan5      but there's been no complains so I didn't revert it
12:03 rubenwardy "accidentally" ;)
12:04 pmp-p      "i love when a plan comes together"
14:26 ANAND      Hmm, where have the Travis XCode builds gone?
14:26 rubenwardy disabled by sfan as broken
14:26 ANAND      Oh nice
14:33 ANAND      #8682 is ready for re-review.
14:33 ShadowBot  https://github.com/minetest/minetest/issues/8682 -- Make players immortal using immortal=1 by ClobberXD
18:57 paramat    merging #8702
18:57 ShadowBot  https://github.com/minetest/minetest/issues/8702 -- Mgfractal: Make non-fractal terrain optional by paramat
19:44 paramat    merging #8705
19:44 ShadowBot  https://github.com/minetest/minetest/issues/8705 -- Re-order mapgens in mainmenu and 'all settings' mapgen selection by paramat
19:47 Krock      sfan5: my laptop has some old mobile ATI radeon card installed. it surely uses OpenGL ES 2.0
19:48 Krock      https://i.imgur.com/A8YiHdp.png
19:49 Krock      it's slow as hell but seems to work just fine
19:50 sfan5      huh
19:51 Krock      can still check the glxinfo output but would need to start it .. meh
20:29 sfan5      when do my PRs get merged? ;)