Alejandro Candioti

There might be a case where computing all predecessors can be slower than computing the list. Need to experiment more but that could be easily fix by an internal Boolean...

> There might be a case where computing all predecessors can be slower than computing the list. Need to experiment more but that could be easily fix by an internal...

> I think we can push the optimization even further! > > This path reconstruction technique can be extended to `target is None` cases. We can completely deprecate the old...

Hey @dschult @rossbar do you think you can review this small one? Further improvements require more profiling so I'd prefer to do them in a separate PR

> The high-level motivation is clear and the code changes are (IMO) relatively minor and straightforward to understand. The next step in my book would be to add some explicit...

@dschult @rossbar do you think there’s more work needed here? I sent a follow-up PR to add benchmarks: https://github.com/networkx/networkx/pull/8059. In cases where the shortest path has few hops, the new...

> Thanks for updating the benchmarking results @amcandio ! They give me pause though: the speedups for the long paths are obvious and undisputed, but it also looks like there...

> I don't see changes on the scale of nanoseconds. Am I looking in the wrong place (the shortest path benchmarking PR #8059 ) It seems like some cases are...

Now that I look at the +- variability everywhere. It could be that my poor Chromebook was struggling when I ran the benchmarks

@dschult so it definetely was my old Chromebook! I was running bencharking on a VM on a 2019 chromebook. Ran it with more respectable hardwarde (M1 chip) and I got:...