GEO instances should truncate values

[rafaelmartinelli]

Currently, the package is rounding the calculated distances for GEO instances. Following the description on the documentation, the distances should be truncated.

For example, instance ulysses16 has distance 1.0 for every arc (i, i), and the optimal value found is 6867.0 instead of 6859.0.

[rafaelmartinelli]

I've already fixed this. I'll create a pull request.

[nathanPro]

Hello everyone!

I have the impression that the current implementation does not match what TSPlib describes in the documentation.

In particular, the documentation states that, when computing the latitude and longitude in radians, we need to do:

deg = nint( y[i])

and then, when computing the distances, we need to do

dij = (int) ( RRR * acos( 0.5*((1.0+q1)*q2 - (1.0-q1)*q3) ) + 1.0);

If I am not confusing the conversion between C and Julia, this amounts to using round in the first part, and trunc in the second.
The current implementation uses trunc in both:

TSPLIB.jl/src/distances.jl

Line 77 in 6f77574

degs = trunc.(hcat(x,y))

TSPLIB.jl/src/distances.jl

Line 89 in 6f77574

dist[i,j] = dist[j,i] = trunc(dij)

Honestly this is more consistent, but I wonder if it changes the graphs computed from the data.
I do not know if this change instances; I am comparing some experiments with code that uses this library, so I used static_cast<int> in both cases.

I guess my main question is: am I missing something? Have you guys noted that TSPlib uses both round and trunc and deliberately chose the current implementation?