Bug: realtime parser returns duplicate vehicles

[jamespfennell]

When I parse the most recent GTFS realtime feed for the NYC L train I see many duplicate vehicles in the result. One vehicle has IsEntityInMessage: false and the other has IsEntityInMessage: true. Presumably the parser sees the first vehicle attached to a TripUpdate and the second attached to a VehiclePosition. It should de-duplicate these - in fact, the IsEntityInMessage: false vehicle should just be dropped in this case.

Observed after jamespfennell/transiter#110 was submitted to Transiter mainline. In Transiter I see lots of debug logging warning of duplicate vehicles in the NYC subway feed, and I think the root cause is here.

CC @cedarbaum.

[jamespfennell]

Started working on this in https://github.com/jamespfennell/gtfs/tree/duplicate-vehicles

[jamespfennell]

Ahhhhh the root cause is pretty subtle. When we're parsing entities we de-duplicate vehicles using a map keyed on VehicleID whose current definition is:

type VehicleID struct {
	ID           *string
	Label        *string
	LicencePlate *string
}

The problem is that even if the values of the ID fields match after following the pointer, Go considers them distinct from a hashing perspective because the pointers themselves are different. I think the simplest solution would be to remove the pointers:

type VehicleID struct {
	ID           string
	Label        string
	LicencePlate string
}

and interpret a field being "" as equivalent to the field not being provided in the GTFS realtime feed. In theory the feed could have Label: &"" but I think this edge case is not a big deal.

[cedarbaum]

Great catch, thanks so much for fixing this!