Decimal/Floating point number arithmetic

[Kolterdyx]

This issue is here to discuss decimal number arithmetic (floating/fixed point numbers) implementation.

Current implementation is whacky at best. It uses fixed point numbers and it has trouble with big numbers because it has to give up digits for decimal values. Addition and subtraction work fine, but multiplication and division cause a lot of trouble since either one or both of the numbers in the operation have to be scaled up or down by 10 to the power of however many digits of precision we have, which causes even more overflow issues. The only good thing is that it's almost as efficient as integer arithmetic (because it technically IS integer arithmetic). The implementation was severely limited by the fact that commands can't return floating point values, which means I couldn't store actual float/double values in storage, and then scale them up or down when needed. This is sadly working as intended so I doubt I'll be able to fix this without a huge overhaul implementing my very own data type.

[WOODEN-CHEST]

I'll create a datapack implementation of the Decimal type which I previously talked about and share it.

I found some ways to optimize it and make it use fewer commands, and most commands it uses are scoreboard and execute if commands, which are rather fast according to tests I did, so it should be sufficient for datapacks.

[Kolterdyx]

I have also found this project, which looks really good in terms of features, but I don't like that it does everything by using an entity and applying transformations to entities. We could maybe use your implementation for fast basic operations and then use gibbsly's library for complex operations which will be heavy anyways

[branyoto]

Wow the project of gibbsly is wild O.O

I agree with you some operation like multiply, negate, square, ceil and floor could be done without entities but rather with the scale factor available on execute store... or data get

[Kolterdyx]

That was my first though, but the issue is that `data get` can only return integers, so we'd lose all precision on that operation. It's a limitation of the command engine Minecraft uses so I doubt it will be changed in the future. That could still be used for integer operations, but for floating point numbers it won't work

…

On Thu, Aug 22, 2024, 11:52 branyoto ***@***.***> wrote: Wow the project of gibbsly is wild O.O I agree with you some operation like multiply, negate and floor could be done without entities but rather with the scale factor available on execute store... or data get — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMMRKZHV477ZQIYRJ3MBRSDZSWYFBAVCNFSM6AAAAABGQEOFMKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBUGI2DSMJXGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[branyoto]

Oh yeah sorry completely forgot that store result ... convert anything in int before applying his scale 🤦

Could still be used for ceil and floor but this is not as helpful as I first thought sorry >.<

[Kolterdyx]

actually, fuck it, I'm going to just use gm for all math, can't be bothered to do a simpler version of math

[Kolterdyx]

Done! Seems to work fine