precision function

[edljk]

.. does not seem to be available anymore for MP:

julia> precision(Float64x4)
ERROR: MethodError: no method matching _precision_with_base_2(::Type{MultiFloat{Float64, 4}})
The function `_precision_with_base_2` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  _precision_with_base_2(::BigFloat)
   @ Base mpfr.jl:957
  _precision_with_base_2(::Type{BigFloat})
   @ Base mpfr.jl:962
  _precision_with_base_2(::Type{Float64})
   @ Base float.jl:872
  ...

[dzhang314]

Hey @edljk, thanks for pointing this out! This is because the Julia authors have once again changed the way that Base.precision(::Type{T}) is internally implemented. It looks like Base._precision has been renamed to Base._precision_with_base_2.

It's annoying that these changes occur with no notification to package authors. I will publish a fix in the forthcoming MultiFloats.jl v3.0 release, which will also include new arithmetic algorithms fixing #42. In the meantime, you can use this patch to restore previous behavior:

using MultiFloats

@inline Base.precision(::Type{MultiFloat{T,N}}) where {T,N} =
    N * precision(T) + (N - 1) # implicit bits of precision between limbs

Please be aware that MultiFloat types work with a fundamentally different representation than IEEEFloat and BigFloat types, and their precision requires special interpretation. The gap between x and the next representable MultiFloat value depends on the exponent of the trailing (least significant) limb of x, as opposed to the leading (most significant) limb. Base.precision(Float64xN) is intended to provide an average-case estimate; it is neither an upper or lower bound.

[edljk]

Perfect, thanks!