Add more methods to Base.:(==)

[hyrodium]

Overview

Before this PR

julia> using Quaternions

julia> quat(1) == 1
true

julia> quat(1) == complex(1)
ERROR: promotion of types Quaternion{Int64} and Complex{Int64} failed to change any arguments
Stacktrace:
[...]

julia> quat(1,1,0,0) == complex(1,1)
ERROR: promotion of types Quaternion{Int64} and Complex{Int64} failed to change any arguments
Stacktrace:
[...]

After this PR

julia> using Quaternions

julia> quat(1) == 1
true

julia> quat(1) == complex(1)
true

julia> quat(1,1,0,0) == complex(1,1)
false

Why add methods?

Base has the following ==(::Real, ::Complex) and ==(::Complex, ::Real) definitions.

https://github.com/JuliaLang/julia/blob/e7345b89fd4eb15e8f395395701e19be705d7b06/base/complex.jl#L247-L249

Quaternions.jl should have methods similar to these, instead of == by type promotions.

How about compatibility with the Complex type?

There were the following choices for the design:

• ==(::Complex, ::Quaternion) always throws an error.
  • The current behavior.
  • I think checking the equality of a complex number and a quaternion is kind of weird, and throwing an error would be a good design.
  • However, most evaluations of == do not throw an error (e.g. "foo" == :foo), so it would be consistent with Base if we avoid throwing errors for that.
• ==(z::Complex, q::Quaternion) returns true if c.re==q.s && c.im==q.v1 && q.v2==q.v3==0, otherwise returns false.
  • This is not good because we don't have compatibility with Complex (Remove compatibility with Complex #113).
• ==(::Complex, ::Quaternion) returns true only if these are real and equal, otherwise throws an error.
  • Similar to the current behavior, throwing an error should be avoided.
• ==(::Complex, ::Quaternion) returns true only if these are real and equal, otherwise returns false.
  • The proposed method definition in this PR.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (2055de5) 100.00% compared to head (670aa2a) 100.00%.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #139   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            2         2           
  Lines          164       168    +4     
=========================================
+ Hits           164       168    +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sethaxen]

In my opinion, any code that mixes complex numbers and quaternions is doing the wrong thing. We're already explicit that there's no natural embedding of complex numbers in the quaternions, and I think it's perfectly fine if we don't have an equality method between quaternions and complex numbers at all, so that even pure real complex numbers and quaternions are unequal.

If every hypercomplex number type needed to add an equality method like this for every other hypercomplex number type, we would have combinatorial explosion and probably lots of unintended ambiguities.

[hyrodium]

Thank you for the comment! I agree with that and close this PR.