Time  Nick             Message
02:05 paramat          4 hrs bug tracing, getting close: https://github.com/minetest/minetest/issues/7570#issuecomment-406762640
07:25 nerzhul          i push an android build update
07:26 nerzhul          merging #7563
07:26 ShadowBot        https://github.com/minetest/minetest/issues/7563 -- Disable shaders GUI on unsupported drivers by numberZero
10:10 lhofhansl        Just merged #7555 (sorry forgot to send a note here first)
10:10 ShadowBot        https://github.com/minetest/minetest/issues/7555 -- Reduce ABM checks. by lhofhansl
10:12 lhofhansl        Also updated #7544
10:12 ShadowBot        https://github.com/minetest/minetest/issues/7544 -- Allow an optional readonly base database by lhofhansl
11:44 p_gimeno         Krock: your coding lacks an implicit '1' bit at the start of the mantissa, that's probably why you can't encode large numbers. You have redundant encodings, so you can't encode every float number that way.
11:48 Krock            the mantissa and exponent are both signed integers in this case. mantissa with 23 bits and the exponent with 9 bits
11:50 Krock            and there's a loss because the exponent is not shifted by 127 as in the IEEE format
11:51 Krock            the sign bit in the IEEE float format is already included in the mantissa here
12:05 p_gimeno         by redundant encodings I mean that you can encode 1 as e.g. mantissa 1 exponent 0, or mantissa 2 exponent 1
12:05 p_gimeno         err
12:05 p_gimeno         I mean or mantissa 2 exponent -1
12:06 p_gimeno         gtg now, back in ~1h
12:12 Krock            oh, I see what you meant. Combinations that result in the same value
13:27 p_gimeno         yes, meaning you can't have a 1:1 mapping
13:30 p_gimeno         the implicit '1' in the mantissa solves that issue, it's only absent for denormals
13:30 p_gimeno         whether to support denormals is a different question... it's probably not worth the trouble
16:13 lhofhansl        Any ideas on #7573?
16:13 ShadowBot        https://github.com/minetest/minetest/issues/7573 -- Android fog color does not match sky color
16:15 lhofhansl        Looks like the far cloud box is simply not drawn. Looking at the code there's no difference there for Android.
16:35 snowmondaykernel Good day! I have mod update that is impotant but wasn't accepted by mod developer. I have no github account. Can someone host a mod for me on github?
16:40 red-001          snowmondaykernel, talk about that sort of stuff on #minetest , #minetest-dev is for core dev
16:40 red-001          and second you can just make an account
16:40 red-001          all you really need is an email address and 5 minutes
16:43 snowmondaykernel red-001, where can you get an email address without using phone number?
16:59 Krock            p_gimeno: I saw the git repo and the SO question, but I'm not sure whether we want such a sophisticated way to transmit float values. It's also a matter of speed
20:32 paramat          game#2150
20:32 ShadowBot        https://github.com/minetest/minetest_game/issues/2150 -- Beds: Add white simple and fancy beds by paramat
20:45 p_gimeno         Krock: I understand, but the point here is that such code would be a fallback for machines where floats are not IEEE-compliant, which would probably be zero of the machines supported by Minetest. The speed impact would apply to machines that don't have IEEE floats. I think it's a fair compromise, given how unlikely it is that any such machine can even run Minetest.
20:49 p_gimeno         See https://stackoverflow.com/questions/2724359/are-there-any-modern-platforms-with-non-ieee-c-c-float-formats about machines without IEEE floats. Summary: some embedded systems (where Minetest can't run), some old GPUs (where Minetest can't run), and that's about all.
20:50 Krock            how about fast math?
20:51 Krock            I'm however not sure what tricks GCC uses to speed things up
20:51 p_gimeno         I'm thinking in terms of network exclusively. If you mean whether fast math uses IEEE floats, yes it does. It may omit denormal support, though, among other things.
20:52 p_gimeno         But lacking denormal support would mean that you would never get a denormal to transmit over a network, and that if you got a denormal from a network, it would be automatically truncated to zero.
20:54 p_gimeno         One area of concern is endianess. You'd need to have several different functions to handle several different kinds of endianess.
20:54 p_gimeno         er, endianness*
20:57 p_gimeno         I think there are 4 combos to support: 0 1 2 3, 3 2 1 0, 2 3 0 1, 1 0 3 2
20:58 Krock            the network code already handles this for integers
20:59 p_gimeno         the situation is a bit worse for floats, in that they don't necessarily have the same endianness as integers, and there's no hton for floats