2018-7-11

New version with support for Julia 0.7; dropped support for previous Julia versions.

2018-1-25

The package now has proper documentation (built with Documenter).
A wider variety of types should be supported; e.g. your distance matrix can have Rational arguments, initial paths can be AbstractVectors, etc.
simulated_annealing will no longer sometimes return a worse path than you pass it.
lowerbound checks that your problem instance is symmetric before using spanning tree bounds.

solve_tsp accepts an optional quality_factor keyword specifying (without units or guarantees) the trade-off between computation time and solution quality.

Julia version 0.4 is no longer supported.
Introduced helper function repetitive_heuristic(distmat, heuristic; keywords...) for repeating a heuristic with many starting cities. The repetitition is parallelized with a @threads loop.
Using the repetitive keyword for nearest_neighbor or cheapest_insertion is deprecated. Instead, use repetitive_heuristic(distmat, heuristic; ...).
The firstcity keyword should now be an Int, not a Nullable{Int}. (For now, backwards compatibility is maintained). If the first city is not specified, a random city is chosen.
The available path generation heuristics now include farthest insertion. Small experiments suggest farthest insertion is the fastest of the generation heuristics and performs decently well on Euclidean instances.

The previous update broke Julia 0.4 compatibility. Thanks to tkelman for the find and fix. I expect to drop 0.4 support when Julia 0.6 is released.