The dijkstra's implementation should be bidirectional

[PayasR]

There are still some perf gains to be had if we switch to a bidirectional dijkstra's implementation for the query phase.
(See https://arxiv.org/pdf/1504.05140.pdf , pages 22-23)

[easbar]

The query algorithm already is bidirectional, see here. IMO the easiest way to improve query speed at the moment would be to implement stall on demand, something like this.

[PayasR]

Oh okay, I was looking at this instead. Yes, stalling is needed too.. do you have it on the To-Do list? I don't see any project roadmap available on GitHub..