…o NO (#3111)

* Low priority private access

* Low priority delivery/destination access

* Check weight limit

* In JTSTriangulator, use DelaunayTriangulationBuilder instead of Confo…

* Simplify property saving

* Allow to store multiple segments

* Allow to load multiple segments

* Add test case for high amount of props

* Use try-with-resources

* bytePos iter

* Check capacity

… some feeds

…dge-based CH alternatives (#3139)

* test

* fix

* remove test generator

[compacted](https://wiki.openstreetmap.org/wiki/Tag:surface=compacted?uselang=en-GB) surface type should have similar speed to gravel and ground.

* do not apply smoothness if penalty was already done via surface

* or pick the smallest of surface vs smoothness?

* review comments

* cycleway encoded value

* docs

* implement simple heuristic to automatically force curbside 'right' for bigger roads to avoid crossing the road

* move one method up

* comment

* remove MOTORWAY; is not used for curbside and might be confusing

Add the ability for map matching to U-turn at directed candidate pairs
(same snap point, opposite directions) with a configurable cost penalty,
instead of only at intersections. This produces shorter, more accurate
matched routes when the GPS trace reverses direction near a candidate.

Also add comprehensive debug output (MatchDebugInfo) and a standalone
HTML visualizer that shows candidates, transitions, U-turns, and the
matched route on a Leaflet map. Remove normalization constants from
log-probability distributions for cleaner debug values.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Move the debug visualizer template HTML from test resources and
web-bundle into map-matching/tools/, alongside a new Python CLI tool
(mapmatch.py) that posts GPX files to the map-matching API and
visualizes results in the browser. The CLI supports chunking large
files, selecting point ranges, and --debug mode which uses the
template to render the full debug visualization.

Remove the duplicate debug.html from web-bundle (was a jQuery-based
variant of the same visualizer) and simplify the test to only assert
on debug info correctness without generating HTML files.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Rename file after successful download

* Use a fixed user agent

…hortcut weights (#3297)

* Scale up weights by x10 + round

* Fix most tests that simply need to expect x10 weights now.

* All algos need to use query graph weighting to route on query graph

* We have to use x10 whenever we use setMaximumLMWeight

* Adjust heuristic in alternative path algos (weights are up x10 -> scale factor down x10)

* Scale down CH shortcut weights

* Remove weight tolerance and lower time tolerance

* use random chains for strict checks with edge-based routing and finite u-turn costs

* Implement weight and time adjustment for virtual edges

* green

* misc

* make corrections smaller

* issue number

* comment

* reject NaN weights and infinite shortcuts

* remove low weight shortcut

* increase shortcut version

* what if we did not adjust weights?

* Revert "what if we did not adjust weights?"

This reverts commit 1701eb6.

* Speed up Weighting#roundWeight

Verifies that QueryGraphWeighting forbids u-turns at inner virtual
nodes (snap points along an edge) while allowing straight-through
traversal at zero cost.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…-turn costs

WeightDetails.isEdgeDifferentToLastEdge() was setting edgeId to the
current edge *before* passing it to calcWeightWithTurnWeight as the
"previous edge" parameter. This caused every weight detail segment to
include a u-turn penalty (current edge -> same edge), inflating
per-segment weights by the u-turn cost (e.g. ~1800s).

Fix follows the same pattern as TimeDetails: pass the previous edge ID
first, then update it to the current edge.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The turn cost for transitioning from edge A to edge B is included in
edge B's weight detail segment. The first segment never includes a turn
cost (no previous edge). This matches the behavior of TimeDetails.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>