Time  Nick       Message
01:33 paramat    ninja! =)
08:31 ANAND      Looking to use std::tuple to return two closely related values, but I notice tuple isn't used anywhere else in the code.
08:32 ANAND      I hope we don't have any no-tuple guidelines :)
08:33 ANAND      Bc I'm not too happy if I have to create a struct just to return two values together.
08:38 p_gimeno   std::pair?
08:41 ANAND      Nice
08:41 ANAND      I'll give it a shot
09:06 ANAND      std::pair suits my use-case well
09:07 ANAND      Now I'm trying to make an ObjectRef method return two values. Is there any special way to do this, or would simply calling lua_pushnumber and lua_pushboolean suffice?
09:09 ANAND      I could just take a look at another API method that returns more than one value, but I can't think of any at the moment. :)
09:21 nerzhul    ANAND yeah we don't have explicit tuples (std::map are using tuples anyway ^^)
09:26 ANAND      Right, thanks. I'm using std::pair as p_gimeno suggested. Is that ok?
09:52 ANAND      Looking at get_eye_offset; the function simply calls push_v3f twice for first and third.
09:52 ANAND      Quite simple for a change... :)
13:14 rubenwardy p_gimeno: have you seen this https://github.com/EvidenceBKidscode/ffiopt/blob/master/ffiopt.patch
13:14 rubenwardy and this: https://forum.minetest.net/viewtopic.php?f=9&t=22555
16:13 p_gimeno   rubenwardy: YUCK. Integer for a pointer?
16:16 p_gimeno   rubenwardy: this was discussed in #6863 and paramat said it was not worth it, because in most Lua mapgens the transfer was a relatively small part of the processing time compared to dealing with the data
16:16 ShadowBot  https://github.com/minetest/minetest/issues/6863 -- Add LuaJIT FFI-friendly memory-intensive functions
16:17 rubenwardy I'm not sure about that
16:17 rubenwardy LuaJIT is quite fast
16:31 p_gimeno   yeah, in my maze generator (see third post of the thread), transfer time dominated, but that generator has a very small processing time. When there's noise to process in the Lua side, the processing time is almost necessarily bigger than the transfer time, to the point that it typically dominates.
16:32 p_gimeno   Someone posted a profiler run of mapgen on IRC recently, let me see if I can find it.
16:35 p_gimeno   http://irc.minetest.net/minetest/2019-04-05#i_5524921
16:36 p_gimeno   "setting the data back to the VM" takes 90 of 630 ms
16:39 p_gimeno   I did some profiling with paramat's riverdev and watershed and found that the speed gain was negligible
16:40 p_gimeno   well, maybe not negligible, but hard to measure due to the dominance of the processing time (as opposed to transfer time, which is what the patch addresses)
16:41 p_gimeno   I still liked the idea of having the buffers out of Lua memory, for memory and general Lua performance reasons
16:46 p_gimeno   wait, does that mod expose FFI for all mods?
16:46 p_gimeno   that's crazy