lower tolerance for band structure segment recognition

I am testing the band-structure plotting functionality with the Octopus calculator and get this:

http://dcwww.camd.dtu.dk/~askhl/files/bandplot-bad-tolerance.pdf

The k-points are parsed from the outputs because probably they could be non-trivially related to the input arguments, but the outputs have a limited printed precision of 6 decimals. However I don't know whether a tolerance of 1e-5, which is what I propose in the MR instead of current 1e-6, is the best. What is a good value in general?

http://dcwww.camd.dtu.dk/~askhl/files/bandplot-good-tolerance.pdf

In the existing GPAW example, typical numbers in the comparison are 1e-2 (new segment) and 1e-16 (otherwise). In Octopus, the latter becomes 2e-6. Presumably k-points are rarely written with smaller precision than that.