Stop medics and techs waiting forever for a patient they cannot reach

[akacross]

Refs #184.

Problem

Once a medic or tech picks the player to heal (rvAIMedic::TakePatient), it becomes passive and ignores enemies, and State_Medic's approach stage only ends beside a patient it can see. If it can see the player but never reach them, it stays in that stage indefinitely, re-pathing in place. Combat can't break it out, and nothing times out. The original Raven DropPatient already notes //FIXME: what if they stay in this state?.

This matches #184: right after hangar1's lift, the script calls $elgar.enableHeal() on Tech Li just before giving him his exit tether. A player with low armor (10 in the report's screenshot) becomes his patient.

Fix (SP and MP, identical)

• ai_medicApproachTimeout (default 15 s, 0 disables): an approach that sets no new best distance to the patient for that long is abandoned. The medic leaves through the existing bail-out path (StopMove, DropPatient, post-heal script) and logs a warning.
• Progress means closing the distance, not moving. A medic circling beneath a patient it can't reach moves constantly without getting closer.
• Cooldown: automatic heals are held back for the same period, so the medic doesn't grab the patient again a second later. A heal the player asks for (TalkTo) isn't held back.
• No save-format change: the watchdog fields are transient and reset on restore.

Testing

Runtime on Linux x64 (Wayland, OpenGL, retail assets), hangar1:

Scenario	0.14.000	This branch
Tech Li, heal enabled, armor 10, player he can see but not reach (in noclip about 90 units above the floor)	Circles beneath the player in State_Medic for the whole 81 s run	At 15 s: medic 'elgar' made no progress toward 'player1' for 15 seconds (distance 92); giving up on the heal for now, then State_Combat → State_Passive
Post-lift sequence, armor 10, player across the platform	Heals in about 4 s, then walks to his tether	Same; no watchdog warning

• tools/tests/ai_medic_approach_watchdog.py (new, wired into both workflows and openq4_validate.py) checks:

  • SP/MP parity;
  • that the watchdog runs first in the approach stage;
  • the bail-out path;
  • that only automatic heals are held back;
  • that nothing new is serialized.

  It fails on main and passes here.

• validation_hardening.py passes.

Not verified

• The unreachable patient here is a synthetic stand-in (a noclip player); I haven't found the exact spot that stranded Tech Li in Hangar Perimeter: Tech Li stalls after elevator, blocking progression #184. I've asked the reporter for a save to confirm the cause.
• Only Linux x64 was run.

Side note: on current main, a startup +map … on the command line leaves the client at the main menu, while the same map command run from a config file loads normally. I tested through +exec instead; that's unrelated to this change.

🤖 Generated with Claude Code

[themuffinator]

Reviewed the current SP/MP diff and its integration with TakePatient, DropPatient and State_Medic. The timeout exits through the existing cleanup/post-heal path, the cooldown applies only to automatic patient acquisition, and the transient fields do not alter the save layout. I found no additional blocking source issue in this revision.

The reported Linux unreachable-patient and normal post-lift runs are useful scoped evidence. I have not independently run gameplay. Keep #184 open until its exact reported state or a supplied save confirms progression; the synthetic unreachable-patient test does not prove that report resolved.

CI run 39136341682 failed before platform validation at the unrelated clipboard test-double error. #190 is now merged and the subsequent atlas math include is repaired on main in 80060cc. I am updating this branch from main to obtain fresh checks. Merge remains pending completed CI; skipped platform jobs do not count as passes. Roadmap ownership: Q4-02.