Int vs Decimal

[lavish crater]

Guys, when it comes to monetary transactions. Do you advise using integers or decimal? I know the difference between one and the other, but I don't know which would be the best and why it would be the best. Remembering that I need to deal with decimal places.

[random slate]

Decimal has an advantage in form that it do not limit you to single monetary unit (not all currencies divide into 100s)

[mental spruce]

never use floats for money, that’s the only solid advice. there are some elixir packages that handle money like https://hexdocs.pm/money/Money.html

[random slate]

Also, in some monetary computations you need fractions of minimal monetary unit

[strange dragon]

nice

[lavish crater]

I didn't know about this lib. I will look

[strange dragon]

thing is, Decimal requires a certain standardization in the projects. These days I did the following: #Decimal<0.5> == #Decimal<0.5000> and got a false as a result

with int numbers we don't have this problem since all right 0s are ignored after the comma

but I found out Decimal.equal?

rs

[mental spruce]

Floating point comparisons will fail in that way for nearly every language. https://0.30000000000000004.com/

[strange dragon]

yep

that's true about float, but not for int

[mental spruce]

Yeah, that’s why in my initial comment I said to avoid using floats (decimals are a float) for money

[strange dragon]

btw, a very famous digital bank here in Brazil had problems with this recently

it was impossible to transfer 17.99R$

the system rounded the value to 17.98 every time

😂

[random slate]

No, Decimal library is not float

[mental spruce]

i’ve never used this specific package, but given that decimal in every context to me means a float, this strikes me as weird. can’t reason why #Decimal<0.5> == #Decimal<0.5000> being false then

[random slate]

No, decimal in any context mean decimal, floats in general mean IEEE 754 binary32/binary64 format. And the reason why #Decimal<0.5> == #Decimal<0.5000> is not equal is the same why %Foo{a: 1, b: 2} != %Foo{a: 2, b: 4}, their internal representation is different and that is why you cannot use == there and you must use Decimal.equal?/2

[livid lance]

floating point is only one way of handling decimal numbers

[random slate]

If anything, decimals can mean IEEE 754 decimal format, which is similar to what Decimal uses internally, but decimals are perfectly fine for money storage, as it avoid most of the issues related to the rounding coming from the binary encoding

The problem with IEEE 754 binary for money comes not from the fact that it is floating point, but from the fact that it has limited precision and that it uses base 2 for computations. If you have unlimited precision floating points in base 10, then there is no issue

[mental spruce]

TIL. looked into the Decimal package itself, it uses some pretty clever tricks to avoid ever dividing numbers and multiplies with multiplicands and their base10 exponents to compute new significant figures and exponents

[random slate]

Yes, because division is super costly in general sense. In x86-64 assembly it is faster to add hundreds of integers than to divide just two.

And that is even without SIMD magic

[ruby depot]

I prefer this library, it has excellent integration with cldr, knows all the details and nuance of all currencies, like rounding rules and stuff: https://hexdocs.pm/ex_money/Money.html