In writing a cucumber test for a node map

| a | b | c |
| d |   | e |
| f | g | h |

with ways

ab
ad
bc
ce
df
eh
fg
gh

where all ways are highway=primary, and equidistant, I discovered that routes ac, ch, af, fh were returning different speeds, where I would have expected them to all return the same speed.

This bug also manifests itself in the following node map:

|   | i |   | j |   |
| k | a | b | c | l |
|   | d |   | e |   |
| m | f | g | h | n |
|   | o |   | p |   |

(with the same ways, in addition to ai, ak, fm, fo, hn, hp, cj, cl, so that all four intersections were full four-way intersections) — in this case again ac, ch, hf, fa all return 59 km/h, wherease ca, hc, fh, af return 57 km/h.

This log shows numbers logged out of the edge_based_graph_factory for the first case. (The first two way names are name_table.GetNameForID(edge_data1.name_id) and name_table.GetNameForID(edge_data2.name_id). The turn angle is apparently being calculated from a possible preceding turn, despite the fact that there is 0 distance traveled on the preceding way — no turn is made as the route starts at an intersection, but a turn penalty is being factored into the first instruction nonetheless.

Bit of an edge case, but 🐛 regardless… : ] @MoKob