PathingBehavior stuck in recalc loop when terrain changes make path impossible

[RiaDev1]

Description

When the bot is following a path and the environment changes (blocks placed/destroyed by other players, mobs, or natural events), the path becomes invalid. PathingBehavior correctly cancels the invalid path and recalculates — but the new path may hit the SAME obstacle, creating a recalc loop where the bot continuously re-paths without making progress.

Root Cause

In PathingBehavior.tickPath():

if (current.failed() || current.finished()) {
    current = null;
    if (goal == null || goal.isInGoal(ctx.playerFeet())) {
        // ... goal reached
        return;
    }
    // ...
    synchronized (pathCalcLock) {
        if (inProgress != null) {
            queuePathEvent(PathEvent.PATH_FINISHED_NEXT_STILL_CALCULATING);
            return;
        }
        queuePathEvent(PathEvent.CALC_STARTED);
        findPathInNewThread(expectedSegmentStart, true, context);
    }
    return;
}

findPathInNewThread starts a fresh path calculation from expectedSegmentStart, but there's no cooldown or backoff between recalculations. If the pathfinder repeatedly produces the same broken path, the bot enters a tight loop of: current.failed() → recalculation → same path → failure.

Expected Behavior

1. Implement exponential backoff: wait longer between each consecutive path recalculation
2. Track the number of consecutive path failures and escalate (e.g. fall back to explore mode after N failures)
3. When path fails at the same position repeatedly, mark that position as temporarily avoided

Suggested Fix

Add a consecutivePathFailures counter and a backoff timer in PathingBehavior. After ~5 consecutive failures from the same position, switch to explore/wander mode instead of re-pathing.

Related

• shredder/src/main/java/baritone/behavior/PathingBehavior.java
• PathExecutor.failed() — triggers the recalc

[3ndetz]

The v0.41.0 tungsten search stall-cap should address the recalc loop when terrain makes a path impossible: the search now tracks real progress (emit / block-path advance, not bare re-roots) and gives up after 20s without any, rather than recalculating forever. Please re-test on the latest 1.21.11 release.