map matching uturn trimming fix #2558
Conversation
…occur at the end of the edge
kevinkreiser
changed the title
| } // No trimming needed | ||
| else if (!edge_begin_info.trim) { | ||
| edge_begin_info.distance_along = 0; | ||
| edge_begin_info.vertex = edge_shape.front(); |
There was a problem hiding this comment.
when we are not trimming either the front or the end we need to make sure we still set these things because they are used regardless below in the shape trimming function
| // Add edge shape to the trip and skip the first point when its redundant with the previous edge | ||
| // TODO: uncommment correct removal of redundant shape after odin can handle uturns | ||
| // trip_shape.insert(trip_shape.end(), edge_shape.begin() + !is_first_edge, edge_shape.end()); | ||
| trip_shape.insert(trip_shape.end(), edge_shape.begin() + !edge_begin_info.trim, |
There was a problem hiding this comment.
the commented out line above is the correct code and causes no duplicates to be had in the shape. however at uturns, it seems that odin does not detect the uturn has occured unless the shape point is doubled where the uturn occurs. cc @dgearhart
| EXPECT_EQ(shape.size(), expected_shape.size()); | ||
| for (int i = 0; i < shape.size(); ++i) { | ||
| EXPECT_TRUE(shape[i].ApproximatelyEqual(expected_shape[i])); | ||
| } |
There was a problem hiding this comment.
this test case showed the above behavior so we simply expanded it to check the actual shape was what we thought it should be (this is how we found out that the odin bug with uturn maneuvers exists). we decided not to fix the odin uturn bug at this time and instead opted to leave the extra shape point in (at the uturn point) so that we get the correct maneuver
| @@ -75,7 +75,7 @@ void via_discontinuity( | |||
| // Insert a second discontinuity so the next (opposing) edge is trimmed at the end from | |||
| // 1-dist along to 1 | |||
| vias.insert({path_index + (flip_index ? 0 : 1), | |||
| {{false, snap_ll, 1.0f - dist_along}, {true, PointLL(), 1.0f}}}); | |||
| {{true, snap_ll, 1.0f - dist_along}, {false, PointLL(), 1.0f}}}); | |||
There was a problem hiding this comment.
this was a long time copy paste fail..
when a uturn is detected in a map match we need to trim the edges shape where the uturn occurs. we recently fixed code that always set trimming information even when it wasnt needed. that fix assumed that the downstream code actually cared to whether or not that boolean telling it to trim was set. this was not the case..
this may fix #2552