Matrix results change when locations are added to the query

[aoles]

It has been reported by one of the users that adding locations to the matrix query might alter the results for the existing locations, as in the example below.

library(openrouteservice)

coordinates <- list(
  c(20.29302, 50.14362),
  c(20.30200, 50.14231)
  )
(res1 <- ors_matrix(coordinates)$durations)
#>       [,1]  [,2]
#> [1,]  0.00 48.98
#> [2,] 48.98  0.00

## add 3rd location to the query
coordinates[[3]] <- c(20.27696, 50.15716)
(res2 <- ors_matrix(coordinates)$durations)
#>        [,1]   [,2]   [,3]
#> [1,]   0.00  48.98 230.51
#> [2,]  48.98   0.00 273.73
#> [3,] 230.51 273.73   0.00

## different 3rd location
coordinates[[3]] <- c(20.29904, 50.14285)
(res3 <- ors_matrix(coordinates)$durations)
#>       [,1]  [,2]  [,3]
#> [1,]  0.00 49.01 33.34
#> [2,] 49.01  0.00 15.67
#> [3,] 33.34 15.67  0.00

identical(res1[1,2], res2[1,2])
#> [1] TRUE
identical(res1[1,2], res3[1,2])
#> [1] FALSE

^{Created on 2022-06-24 by the reprex package (v2.0.1)}

Unfortunately, I was unable to reproduce this locally. The problem might be related the use of virtual edges introduced by the additional locations.

[aoles]

I was able to find a reproducible example on the HD test graph where adding a 3rd location increases the original travel time. This happens regardless of the matrix algorithm used, i.e. both for RPHAST and Dijkstra.

library(openrouteservice)

options(openrouteservice.url = "http://localhost:8082/ors")

coordinates <- list(
  c(8.679177, 49.41364),
  c(8.686624, 49.423034)
)
(ref <- ors_matrix(coordinates)$durations)
#>        [,1]   [,2]
#> [1,]   0.00 353.47
#> [2,] 364.28   0.00

coordinates[[3]] <- c(8.679672, 49.413868)
(res <- ors_matrix(coordinates)$durations)
#>        [,1]   [,2]   [,3]
#> [1,]   0.00 353.49   5.96
#> [2,] 364.28   0.00 370.25
#> [3,] 122.96 347.53   0.00

identical(ref, res[-3, -3])
#> [1] FALSE

^{Created on 2022-06-27 by the reprex package (v2.0.1)}