fixed eltype for interval

[lucaferranti]

Description

changed eltype so that now eltype(1..2) returns Interval{Float64} instead of Float64. This was causing issues when doing e.g. matrix*interval multiplication.

Since a function doing what eltype was doing before can be useful, I added numtype to do that. Happy to use another name if you don't like it

Before

julia> eltype(1..2)
Float64

julia> A = [1 2;3 4]
2×2 Matrix{Int64}:
 1  2
 3  4

julia> b = 1..2
[1, 2]

julia> A*b
ERROR: MethodError: no method matching Float64(::Interval{Float64})
Closest candidates are:
  (::Type{T})(::Real, ::RoundingMode) where T<:AbstractFloat at rounding.jl:200
  (::Type{T})(::T) where T<:Number at boot.jl:760
  (::Type{T})(::AbstractChar) where T<:Union{AbstractChar, Number} at char.jl:50

After

julia> eltype(1..2)
Interval{Float64}

julia> A*b
2×2 Matrix{Interval{Float64}}:
 [1, 2]  [2, 4]
 [3, 6]  [4, 8]

Related issues

• fix Fixing broadcast requires changing eltype  #397
• fix eltype should return the full Interval type (?) #487

[codecov-commenter]

Codecov Report

Merging #496 (7a62674) into master (7e3f494) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #496   +/-   ##
=======================================
  Coverage   91.08%   91.09%           
=======================================
  Files          25       25           
  Lines        1783     1785    +2     
=======================================
+ Hits         1624     1626    +2     
  Misses        159      159           

Impacted Files	Coverage Δ
src/IntervalArithmetic.jl	100.00% <ø> (ø)
src/intervals/arithmetic.jl	97.24% <100.00%> (ø)
src/intervals/functions.jl	91.57% <100.00%> (+0.04%)	⬆️
src/intervals/hyperbolic.jl	100.00% <100.00%> (ø)
src/multidim/intervalbox.jl	89.47% <100.00%> (+0.18%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7e3f494...7a62674. Read the comment docs.

[lbenet]

Thanks @lucaferranti for addressing this! I'll try to review this PR tomorrow or over the weekend... sorry, I'm having some very busy days.

[lucaferranti]

No problem at all Luis! Thanks for looking into this :)

[mforets]

This was causing issues when doing e.g. matrix*interval multiplication

Maybe add such test? (examples in the linked issues).

Apart from that, the proposed changes lgtm 💯

[mforets]

i think this PR introduces a potentially breaking change. (apparently, tests for eltype were lacking).

[lbenet]

LGTM! Thanks for solving this. I have only added a minor comment/suggestion, which is probably not needed...

[lbenet]

Similar to what it's done with Interval, I guess it's a good idea to define numtype for IntervalBoxes...

[lucaferranti]

I think you are right that it can be useful, added it!

[lbenet]

I propose to merge #497 first (if possible with tests for the M1 processor), then this one (it may need a rebase), and finally register the new version.

[lucaferranti]

Agreed! It's a good idea