round(Int,1+0im) throws an error while convert does not (Int,1+0im) in 0.4.1

[dlfivefifty]

The following seems inconsistent:

julia> convert(Int,-1+0im)
-1

julia> round(Int,-1+0im)
ERROR: MethodError: `round` has no method matching round(::Type{Int64}, ::Complex{Int64})
Closest candidates are:
  round{T<:Integer}(::Type{T<:Integer}, ::Integer)
  round{T<:Integer}(::Type{T<:Integer}, ::Float16)
  round{T<:Union{Signed,Unsigned}}(::Type{T<:Union{Signed,Unsigned}}, ::BigFloat)
  ...

[stevengj]

What would round(Int, -1+2im) return? Since the round function is not sensible for most points on the complex plane, I'm skeptical that we should define it on the real axis.

[dlfivefifty]

The same thing as convert: it would through an error

Sent from my iPhone

On 5 Dec 2015, at 2:38 AM, Steven G. Johnson notifications@github.com wrote:

What would round(Int, -1+2im) return? Since the round function is not sensible for most points on the complex plane, I'm skeptical that we should define it on the real axis.

—
Reply to this email directly or view it on GitHub.

[stevengj]

Is there any precedent in any other language for a round function that works for complex z?

[dlfivefifty]

I'm more just advocating that round and convert are almost the exactly the same thing, and so should behave the same.

So maybe remove convert(Int,1+0im)

Sent from my iPhone

On 5 Dec 2015, at 6:59 AM, Steven G. Johnson notifications@github.com wrote:

Is there any precedent in any other language for a round function that works for complex z?

—
Reply to this email directly or view it on GitHub.

[dlfivefifty]

Also in Mathematica

Round[1 + 0.1 I] == 1

On 5 Dec 2015, at 7:15 AM, Sheehan Olver solver@mac.com wrote:

I'm more just advocating that round and convert are almost the exactly the same thing, and so should behave the same.

So maybe remove convert(Int,1+0im)

Sent from my iPhone

On 5 Dec 2015, at 6:59 AM, Steven G. Johnson <notifications@github.com mailto:notifications@github.com> wrote:

Is there any precedent in any other language for a round function that works for complex z?

—
Reply to this email directly or view it on GitHub #14261 (comment).

[StefanKarpinski]

We could define round(z::Complex) = Complex(round(real(z), round(imag(z))).

[StefanKarpinski]

Doesn't quite address this issue.

[jiahao]

We could define round(z::Complex) = Complex(round(real(z), round(imag(z))).

@StefanKarpinski we already do: see #8291

[stevengj]

Yes, the Round function in mathematica is analogous to our current round(z), not the proposed round(Int, z). e.g. Round[1 + 1.1 I] gives 1+i.

[dlfivefifty]

I suppose it's not clear what the following should return:

round(Int,1+0im)
round(Int,1+eps()im)
round(Int,1+0.1im)
round(Int,1+0.6im)
round(Int,1+2im)

I think if all returned 1 it would make sense, if we think of the definition of round(T,x) as convert to type T, rounding if necessary.

Sent from my iPhone

On 5 Dec 2015, at 10:33 AM, Steven G. Johnson notifications@github.com wrote:

Yes, the Round function in mathematica is analogous to our current round(z), not the proposed round(Int, z). e.g.Round[1 + 1.1 I]gives1+i`.

—
Reply to this email directly or view it on GitHub.

[laborg]

Has been addressed in #50812

julia> round(Int,-1+0im)
-1

[dlfivefifty]

I didn't see a unit test added for this case. Can we add a @test round(Int,-1+0im) === -1?