Time  Nick      Message
20:59 Krock     nerzhul: what kind of "updates on the server" do you mean?
21:36 Krock     nerzhul: new toy #7569
21:36 ShadowBot https://github.com/minetest/minetest/issues/7569 -- Send custom floats (log2) by SmallJoker
21:54 p_gimeno  sorry for the dumb question, but what's wrong with just using F32?
21:54 p_gimeno  I mean, actual single-precision floats
21:54 Krock     as in memcpy?
21:56 p_gimeno  yes, or union integer and hton/ntoh
21:56 Krock     the memory representation is not defined, so it would be required to disassemble each feature of the float
21:56 Krock     and the bitop everything together again so that it's ensured to be the correct format
21:56 Krock     see also: https://github.com/minetest/minetest/pull/6312
21:59 Krock     I cannot test whether all by Minetest supported architectures also will work with this idea - missing computers, time and motivation to get the stuff together
22:00 p_gimeno  you don't need floats to work the same in all architectures in order to transmit them through a network, and while the C++ standard does not specify floats to be IEEE-754 compliant, you can add a configure check and either refuse to compile, or have fallback code for IEEE conversion
22:01 p_gimeno  I'm pretty sure there must be simple functions with a usable license that convert float to IEEE and vice versa
22:04 p_gimeno  it seems you did a good job getting close to that, already
22:06 Krock     thanks. However, it's not perfect. Special numbers (nan/inf/-inf) and those close to the maximal float value aren't handled correctly
22:09 p_gimeno  special cases are trivial to handle: if (isnan(f)) return 0x7FC00000; if (isinf(f)) return 0x7F800000 | (f < 0) << 31;
22:11 p_gimeno  well, the other way around has the danger that the implementation aborts on division by zero instead of returning NaN/Inf, but I doubt that's a problem
22:12 p_gimeno  that's assuming you want to transmit these values through the network in the first place
22:13 Krock     yes. at least have the regular values correctly and as precise as possible transmit
22:14 Krock     will check logs
22:17 p_gimeno  a configure check (or even runtime) seems pretty trivial: https://stackoverflow.com/questions/5777484/how-to-check-if-c-compiler-uses-ieee-754-floating-point-standard
22:34 p_gimeno  here's one such IEEE conversion project: https://github.com/MalcolmMcLean/ieee754
22:35 p_gimeno  unfortunately that one doesn't specify a license, guess you could bug the author in an issue to see if he would add one
22:45 nerzhul   Krock interesting PR