Open vpaturet opened 2 months ago
Does the same problem apply to NeTEx service journeys?
Hard to tell. At least it is not observable with the current test suite. The TripPattern creation logic is also completely different. With NeTEx, TripPatterns exist in the dataset in the form of JourneyPatterns, their NeTEx id can be used as TripPattern id, the direction is specified in the NeTEx Route and the grouping of Trips (ServiceJourneys) in TripPattern is also specified in the dataset. This gives little room for variation. With GTFS data, TripPatterns are constructed as an arbitrary grouping of Trips based on Route, stop sequence, mode, direction. TripPattern IDs are generated with a counter.
Expected behavior
Heuristic results should be stable and should not depend on the order the trips are imported during graph build.
Observed behavior
As illustrated in this commit, changing the order in which GTFS trips are imported during graph build will also change the results returned by the heuristic search. This is demonstatred by a variation in the results recorded by the speed test.
Version of OTP used (exact commit hash or JAR name)
dev-2.x
Data sets in use (links to GTFS and OSM PBF files)
speed test
Steps to reproduce the problem
Run speed test with the commit mentioned above.