FFI floats comparison

[Matrix89]

It seems that when comparing two ffi floats only the integer part is checked eg. 0.1 and 0.5 are equal.

LuaJIT 2.1.1710088188 -- Copyright (C) 2005-2023 Mike Pall. https://luajit.org/
JIT: ON SSE3 SSE4.1 BMI2 fold cse dce fwd dse narrow loop abc sink fuse
> ffi = require("ffi")
> a = ffi.cast("float", 0.1)
> b = ffi.cast("float", 0.1)
> = a == b
true
> b = ffi.cast("float", 0.5)
> = a == b
true
> b = ffi.cast("float", 1)
> = a == b
false
>

Is this intentional or a quirk of how floats actually work?

[GitSparTV]

Does it work the same if you use ffi.new?

[Matrix89]

Yep

LuaJIT 2.1.1713773202 -- Copyright (C) 2005-2023 Mike Pall. https://luajit.org/
JIT: ON SSE3 SSE4.1 BMI2 fold cse dce fwd dse narrow loop abc sink fuse
> ffi = require("ffi")
> a = ffi.new("float", 0.1)
> b = ffi.new("float", 0.2)
> = a == b
true
> b = ffi.new("float", 1)
> = a == b
false
>

BTW. I've only tested on x86

[MikePall]

Please never box scalar numbers with ffi.new("int"), ffi.new("float") or similar! This is only useful for passing arguments to vararg functions. Ditto for ffi.cast().

Explicitly boxing scalars does not improve performance or force int or float arithmetic. Or whatever the intention possibly was.

It just adds costly boxing, unboxing and conversions steps. And it may lead to surprise results, because cdata arithmetic on scalar numbers is always performed on 64 bit integers.

I've added this to the documentation. Thanks!