Issue

Fixes #5967

Negative turn duration penalties were not being clamped during traffic updates or at extraction time, unlike turn weight penalties which were. This caused edge.data.duration to go negative, producing physically impossible negative route durations.

Was this change primarily generated using an AI tool? Yes — GitHub Copilot (claude-sonnet-4.6) was used to assist with analysis, implementation, and test writing.

Tasklist

• self-review code for correctness and following the coding guidelines
• add tests (see testing)
• update relevant wiki pages
• CHANGELOG.md entry (see how to)
• review
• adjust for comments

Requirements / Relations

No other PRs required.

Changes

src/updater/updater.cpp — core fix (traffic update path):

• In update_edge, clamp the turn duration penalty using num_nodes as the floor (deciseconds), mirroring the existing weight clamping block. Log a warning when the penalty is too negative.
• In updateTurnPenalties, add a logWARNING for negative turn_duration_penalty values (mirrors existing weight warning).

src/extractor/scripting_environment_lua.cpp — extraction-time fix:

• In ProcessTurn, clamp turn.duration to -MAXIMAL_TURN_PENALTY / 10.0 (seconds). This prevents int16_t overflow and keeps durations in a physically meaningful range. Uses the fixed decisecond resolution (÷10), not GetMaxTurnWeight() which is scaled by weight_precision and only applies to weights.

features/testbot/traffic_turn_penalties.feature — tests:

• New scenario reproducing issue Negative route durations #5967: a penalty of -70 on a middle turn that would yield -10s total without the fix, now correctly produces ~40s.
• Updated two existing c→g speed expectations from 72 km/h → 71 km/h. The old code masked the negative penalty via a phantom-node first-edge clamp in routing_base.hpp; with the fix applied upstream the route is 0.1s longer, changing the truncated speed by 1 km/h.

Testing

All 363 testbot scenarios pass with MLD. Turn penalty feature file also verified with CH.