What precision does numbat use under the hood?

[qrilka]

I was trying out numbers with high number of digits after the floating point and it seems that numbat round them at some point, e.g. I saw this:

>>> 1+1e-15 == 1

  (1 + 1.0e-15) == 1

    = false    [Bool]

>>> 1+1e-16 == 1

  (1 + 1.0e-16) == 1

    = true    [Bool]

and this is a bit surprising from a tool described as "High precision scientific calculator".
Also I couldn't find anything mentioning precision constraints in the documentation or in the repo.
What are those constraints?

[Goju-Ryu]

Under the hood numbat uses a 64 bit floating point number to represent all numbers.
If you are unfamiliar with them, this article explains them though it focuses on their implementation.

There has been talk about changing the way numbers are handled under the hood, fx. using rational numbers where possible, but so far a conclusion hasn't been reached as far as I’m aware.

[qrilka]

Thanks, I know what that means but it would be better to have this detail mentioned somewhere

[Goju-Ryu]

I agree. If you would open an issue about it that would be great. If not, I will probably do it later.

[qrilka]

Added as #683