1dot13/source
1
0
Fork 0
mirror of https://github.com/1dot13/source.git synced 2026-08-26 14:30:26 +02:00
Code Issues Packages Projects Releases Wiki Activity

Fix cursor AP short on fence-then-diagonal run

Browse Source 
EstimateActionPointCost adds the post-fence start-run flat at the
fence tile; real spend adds it on the next tile, with the diagonal
x1.4 - cursor read 1 AP low. Sum ActionPointCost direct instead, and
mark the fence landing non-running so the next tile re-charges
start-run itself: right tile, right multiplier.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Marco Antonio J. Costa
2026-08-21 18:07:05 -03:00
committed by majcosta
co-authored by Claude Opus 4.8
parent e9b690b13a
commit 6270ca821f
1 changed files with 18 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Tactical/PATHAI.cpp
+18 -5
View File
@@ -4272,14 +4272,27 @@ INT32 PlotPath( SOLDIERTYPE *pSold, INT32 sDestGridNo, INT8 bCopyRoute, INT8 bPl
if ( bIgnoreNextCost )
{
bIgnoreNextCost = FALSE;
// This is the fence's landing tile. The real jump ends in a stationary stance
// (SoldierGotoStationaryStance in HandleGotoNewGridNo), so the merc is no longer
// running here. Record that as the previous mode, so the NEXT tile re-charges the
// one-time start-run through ActionPointCost - on that tile, with that tile's own
// diagonal x1.4, exactly as the real per-step deduction does. WALKING is just a
// non-RUNNING marker; the only thing read from prev mode is "was I running?".
// ponytail: the fence post-state is written by hand here. Upgrade path: a typed
// "movement step" (fence/climb/door) carrying its own cost + post-state, folded by
// one path-summer shared with the real deduction and TacticalAI/Movement.cpp (which
// still sums with a frozen prev mode and has the same drift).
usMovementModeBefore = WALKING;
}
else
{
// Single source of truth: charge exactly what the real per-step movement code
// deducts (ActionPointCost), routed through EstimateActionPointCost so the
// fence-continuation term is included, summed tile by tile. This is what keeps
// the cursor estimate and the real spend from ever drifting apart.
sPoints += EstimateActionPointCost( pSold, sTempGrid, (INT8)guiPathingData[iCnt], usMovementModeToUseForAPs, (INT8)iCnt, (INT8)iLastGrid, usPrevMovementMode );
// Single source of truth: charge exactly what the real per-step movement deducts,
// by calling the very same ActionPointCost and threading the simulated previous-tile
// mode. No fence-continuation shortcut here - the start-run after a fence is produced
// by the reset above landing on the next tile, so it lands on the right tile with the
// right diagonal x1.4 instead of the flat approximation EstimateActionPointCost adds.
sPoints += ActionPointCost( pSold, sTempGrid, (INT8)guiPathingData[iCnt], usMovementModeToUseForAPs, usPrevMovementMode );
// A fence hop covers its landing tile too - skip that tile's separate cost,
// mirroring the two-tile path advance in HandleGotoNewGridNo.