make Irrational <: AbstractIrrational

[stevengj]

As I mentioned in #24238, adding new irrational types, e.g. to represent exact multiples of π, would be vastly easier if we had Irrational <: AbstractIrrational <: Real rather than Irrational <: Real, because most of the methods for promotion etc. would be shared with the existing Irrational type.

This is a pretty simple change that I think we should make immediately.

[nalimilan]

Can you explain why subtyping Irrational doesn't work? That's not clear to me even after reading the issue.

[stevengj]

@nalimilan, what do you mean? Our current Irrational is not an abstract type, so it is impossible to subtype.

[fredrikekre]

Could be added to the "Abstract number types" section of doc/src/stdlib/numbers.md

[stevengj]

Okay, done.

[stevengj]

Travis failure seems unrelated?

The command "make -C moreutils mispipe" exited with 0.
0.08s$ make $BUILDOPTS -C base version_git.jl.phony
make: Entering directory `/home/travis/build/JuliaLang/julia/base'
sh /home/travis/build/JuliaLang/julia/base/version_git.sh /home/travis/build/JuliaLang/julia/base > version_git.jl.phony
fatal: bad revision '^^d6638e8434659c80ee967a60eb0858b1ec5af20'

[fredrikekre]

Should be unrelated. I wonder why AppVeyor did not run on this?

[stevengj]

Another case where I would like to use this: to define an irrational type with different promotion rules for the ChangePrecision package (see https://discourse.julialang.org/t/change-default-precision/6520/51?u=stevengj).

[StefanKarpinski]

This ready to merge, @stevengj?

[Jutho]

What would now be the minimal interface/set of methods that a new subtype of AbstractIrrational needs to implement? Conversion to Float64 and BigFloat? I have a use case for square roots of rational numbers, which often occur in physical constants (e.g. Wigner symbols in my use case).

[garrison]

I created a new issue for "Next steps/guidelines for AbstractIrrational" (#26550)