Inconsistent return types (and error) with arithmetic on LinSpace

[yuyichao]

Trying to do simple arithmetics on the LinSpace object returns inconsistent result (or error).

julia> typeof(linspace(0, 10, 20))
LinSpace{Float64}

julia> typeof(linspace(0, 10, 20) + 1)
FloatRange{Float64}

julia> typeof(linspace(0, 10, 20) - 1)
Array{Float64,1}

julia> typeof(linspace(0, 10, 20) * 1)
Array{Float64,1}

julia> typeof(linspace(0, 10, 20) / 1)
Array{Float64,1}

julia> typeof(+linspace(0, 10, 20))
LinSpace{Float64}

julia> typeof(-linspace(0, 10, 20))
ERROR: MethodError: `-` has no method matching -(::LinSpace{Float64})
Closest candidates are:
  -{S,T}(::Range{S}, ::Range{T})
  -{S,T}(::AbstractArray{S,N}, ::Range{T})
  -{S,T}(::Range{S}, ::AbstractArray{T,N})
  ...

OTOH, doing the same on ranges seems to be pretty consistent.

julia> typeof(0:2:20)
StepRange{Int64,Int64}

julia> typeof((0:2:20) + 1)
StepRange{Int64,Int64}

julia> typeof((0:2:20) - 1)
StepRange{Int64,Int64}

julia> typeof((0:2:20) * 1)
StepRange{Int64,Int64}

julia> typeof((0:2:20) / 1)
FloatRange{Float64}

julia> typeof(+(0:2:20))
StepRange{Int64,Int64}

julia> typeof(-(0:2:20))
StepRange{Int64,Int64}

P.S. what's the difference between a range and a linspace?

[yuyichao]

Maybe this is a better comparison...

julia> typeof(0:2:20.)
FloatRange{Float64}

julia> typeof((0:2:20.) + 1)
FloatRange{Float64}

julia> typeof((0:2:20.) - 1)
FloatRange{Float64}

julia> typeof((0:2:20.) * 1)
FloatRange{Float64}

julia> typeof((0:2:20.) / 1)
FloatRange{Float64}

julia> typeof(+(0:2:20.))
FloatRange{Float64}

julia> typeof(-(0:2:20.))
FloatRange{Float64}

anyway... The fact taht linspace(0, 10, 20) + (-1) and linspace(0, 10, 20) - 1 returns different types and that -linspace(0, 10, 20) throws an error is quite weird.

The latter causes a test failure on LsqFit.jl.

[Gnimuc]

OrdinalRange is a subtype of Range.
FloatRange is a subtype of Range.
LinSpace is a subtype of Range.
But there is no - operation on LinSpace in range.jl (line 535).

## linear operations on ranges ##

-(r::OrdinalRange) = range(-r.start, -step(r), length(r))
-(r::FloatRange)   = FloatRange(-r.start, -r.step, r.len, r.divisor)

maybe Stefan forgot to overload it 42db285

[jhlq]

This is probably the cause of some mysterious deprecation warnings that
initially said a range constructed from ints with subtraction of ints was
not a range of ints...
On Apr 30, 2015 8:41 AM, "Gnimuc Key" notifications@github.com wrote:

OrdinalRange is a subtype of Range.
FloatRange is a subtype of Range.
LinSpace is a subtype of Range.
But there is no - operation on LinSpace in range.jl (line 535).

linear operations on ranges

-(r::OrdinalRange) = range(-r.start, -step(r), length(r))
-(r::FloatRange) = FloatRange(-r.start, -r.step, r.len, r.divisor)

maybe Stefan forgot to overload it 42db285
42db285

—
Reply to this email directly or view it on GitHub
#11049 (comment).