glass dirty-skip: fold palette generation into the repaint token

Latent flaw in the c9e25e5 dirty-skip (now in master): the token hashes plane PIXELS + lamp state only, but the ColorMapper gauge family (armor rosette tints, the adpal/adpal2 damage flash) animates by writing CLUT entries with ZERO pixel churn -- so a glass window whose plane is pixel-static held stale colors while the D3D surround (which re-expands every frame) tracked live. New SVGA16::paletteGeneration, bumped by the palette writers: - BuildSecondaryColor: only on a REAL entry change (the flash alternates palettes every Execute writing identical RGB at zero damage -- an unconditional bump would repaint every palette-expanding window every pump for nothing); - BuildSecondaryPalette / BuildAuxiliaryPalette / BlankPalette (full rebuilds): unconditionally. GlassWindowToken folds it in for palette-expanding windows (monoTint < 0, or all windows under BT_GLASS_MFD_PAL); mono-tint windows don't read the palette and keep their pixel-only token. KB: glass-cockpit.md gains the repaint-pump section (records f3d27f5 + c9e25e5 + this). Companion branch sec-phantom-fix carries the idx-254 translation-table fix -- disjoint hunks, merges in either order. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-08-10 10:56:49 -05:00
6 changed files with 122 additions and 34 deletions
@@ -414,6 +414,14 @@ Verified live: bay fire → lamp 0xD (the LRM's select button) flashes 0x37 + en
 on detonation/purge. Diagnostics: `BT_LAMP_LOG` → `[techstat]`/`[galarm]`/`[lamp]`. Details +
 the four load-bearing fixes en route: [[open-questions]] + [[decomp-reference]] §GaugeAlarm.

+**These conditions are ROUTINE and SELF-CLEARING [T2, field-verified 2026-08-08].** A `SET` is an
+operating state, not a fault: every laser volley trips Overheating (cond 3) and clears it on
+cooldown (33× on one LLaser in a single match), and BadPower (cond 6) flickers whenever
+simultaneous draw browns the bus. Across a full match every subsystem's SET/CLEARED counts are
+balanced — nothing latches. So a lamp flashing after a respawn is the mech *operating*, not a
+failed reset; diagnose from the SET/CLEARED balance, never from a lone SET. Full census + the
+#137 post-mortem it settled: [[decomp-reference]] §TechStatus.
+
 ## ConfigMapGauge (the weapon panel's trigger-config joystick) — LIVE via LinkToEntity (2026-07-21)
 The per-weapon btjoy.pcc joystick image + 4 cm_* state lamps (off/other/only/both) showing,
 for each mappable fire button (Pinky/ThumbLow/Trigger/ThumbHigh), whether THIS panel's weapon
@@ -549,6 +557,38 @@ and every instrument is now live [T2]:**
  transcription color bug caught by a period reference screenshot, 2026-07-09; same for the
  bottom bowtie carets @4569-4570); pegs at 1200 with no target; the DISPLAYED range slides at
  **500 m/s** toward the true pick range (HudSimulation :5652 [T1]).
+  **The HudSimulation tuning constants, read off .rdata 2026-08-08 [T1]** (`section_dump.txt`
+  rows ` 4b7ec0  8be55dc3 0000403f 0000803f 0000c842` / ` 4b7ed0  00000000`) —
+  `_DAT_004b7ec4` = **0.75f**, `_DAT_004b7ec8` = **1.0f**, `_DAT_004b7ecc` = **100.0f**,
+  `_DAT_004b7ed0` = **0.0f**, `_DAT_004b7f90` = **0.0f**. ec4/ec8 are the fire-control **LOCK**
+  limits (own HUD host zone < 0.75 damage, targeted zone < 1.0 — so a shot-up cockpit drops to
+  "target held, no lock", and a dead zone can't be re-locked); ed0 is the shared **zero** in the
+  range-slide `Abs()` idiom (the 500 is an immediate `0x43fa0000`, NOT a global). `hud.cpp` had
+  carried all five as 0.0f/500.0f stand-ins under guessed names ("SegmentTempLimit … heat
+  threshold for HUD page visibility" was neither heat nor page visibility) — corrected in place,
+  with the live lock/slide implementation staying in mech4.cpp's targeting step, which had both
+  thresholds right all along.
+  **THE CARET CAN DIE FOR THE WHOLE SESSION — #147, fixed 2026-08-08 [T2 code, T3 field link]:**
+  `sShownRange` (mech4's targeting step) is a **function-level static** — one cell per process,
+  shared by every mech, carried across drops. **NaN is absorbing** in `step = trueRange −
+  sShownRange; sShownRange += step`, and neither the producer's clamps nor
+  `BTReticleRenderable::Draw`'s (`range < minRange` / `range > maxRange`) catch it — both
+  comparisons are **false for NaN**. A poisoned value therefore reaches `AddPoint`/`ConcatMatrix`,
+  and the caret + its bar become **degenerate geometry that stops rendering while the static tick
+  marks keep drawing** — the exact reported symptom ("no range finder on this drop", ticks there,
+  caret gone), sticky until relaunch. Fixed four ways: re-seed on mech change (not on respawn —
+  that reuses the entity, and the binary doesn't reset the readout either), a producer NaN trap, a
+  NaN-safe consumer clamp falling back to the 1200 peg, and — the reason no log could settle it —
+  **`BT_RANGE_LOG` now prints the caret's actual input** (`[range] caret input shown=`). It never
+  did before: `BT_RANGE_LOG` instrumented the PICK (#4) and `[target]`'s `range=` is a *separate*
+  locally-recomputed Sqrt in the weapon-range check, so grepping field logs for NaN found nothing
+  because the poisoned variable was never printed. Causal link to the field report is INFERENCE.
+  **GAP — the 100 m RANGE BIAS is NOT reconstructed [T1 read, unimplemented]:** HudSimulation
+  subtracts `_DAT_004b7ecc` (100.0f) from `RangeToTarget@0x1EC` **every frame while the timed
+  flag @0x22C is set**, accumulating @0x21C by `time_slice` until it reaches @0x1D8, then clearing
+  both. The port's targeting step (mech4.cpp) does the slide but never the bias, so whatever
+  in-game state sets @0x22C currently produces no range offset. Trigger for @0x22C not yet
+  identified — see [[open-questions]].
  **VERDICT (Gitea #4, 2026-07-20): the "range slides in/out crazily while walking" report is
  AUTHENTIC behavior, not a bug [T2 measured].** Per-frame `BT_RANGE_LOG` traces (mech4.cpp, with
  an independent Möller-Trumbore cross-check `BTGroundRayHitExact` in btvisgnd.cpp) on scripted
@@ -783,22 +823,6 @@ pattern across entries [2^bits..255] (high-index art degrades to its
 index-mod-2^bits colour IN-PLANE, can never leak).  The 1995 binary ships the
 SAME 64-entry fill and relied on 6-bit art discipline -- garbage is not a
 preservable behaviour, so the cycle-fill is a guarded PORT deviation.
-
-**REFINED 2026-08-10 [T1] -- the cycle-fill itself made phantoms.** The
-in-plane wrap mapped art index 254 onto plane slot 62 = the LIVE
-`colorMapperMultiArmor` right-armor damage slot, and idx-254 art EXISTS:
-every SMODE.PCC frame fills the INACTIVE control-mode box interiors with 254
-(active border/text idx 9 yellow, inactive borders idx 5 orange-red), and
-BTSEC1.PCX carries a stray 52x13 idx-254 bar at port (199-250, 101-113) --
-between the heading dial and the armor rosette, geometrically a scratch
-duplicate of the rosette quadrant bars.  Both lit up in the current
-right-armor colour (adpal ramp green/orange/red) on EVERY render path --
-playtester-reported as "red boxes around MID/ADV" + "a block between Armor
-and Heading".  On the shipped machine those regions rendered BLACK (the
-garbage entry's low plane bits were 0 -> in-plane index 0), so the fill now
-maps [2^bits..255] to `translationTable[0]` (plane BACKGROUND) -- authentic
-on-screen result, same no-leak guarantee.  NB the 2026-07-19 audit verified
-smode BEFORE the 07-25 cycle-fill landed, which is why the row said CORRECT.
 Post-fix: **0 leaks over 60s** on the same probe; sim3 3-pod regression clean.
 **The tripwires are DEFAULT-ON in every build** (BT_PLANE_AUDIT=0 opts out): the
 plane-leak trap, plus an out-of-bounds draw-start trap in `buildDestPointer`
@@ -203,6 +203,32 @@ appears in the cfg like any panel and honours `,noframe`. Verified: a drag wrote
 at 321,222. NB the plasma window blits directly every frame (`GetDC`+`StretchDIBits`), so unlike
 the panels it has no `WM_TIMER` focus-throttle to worry about.

+## Glass-panel repaint pump — perf + dirty-skip (2026-08-09/10) [T2 measured]
+
+The exploded per-display windows are pure CPU/GDI (`ExpandPlaneToBGRA` +
+`StretchDIBits`), repainted synchronously on the MAIN render thread by
+`BTGlassPanels_Tick` (~16 Hz). Three layered changes after a playtester reported
+~20 fps in panels mode vs ~130 fps in the surround:
+
+1. **HALFTONE → COLORONCOLOR** (`f3d27f5`, in master): the `HALFTONE` stretch
+   (GDI's per-output-pixel resample) × 7 windows per pump was the primary sink;
+   nearest is crisper for the low-res MFD pixels anyway. `BT_GLASS_SMOOTH=1` restores.
+2. **Per-window dirty-skip** (`c9e25e5`, in master): each window carries a change
+   token = FNV over `SVGA16::PlaneChecksum(mask)` (the shared pixelBuffer masked to
+   every port that can feed the window) + each button's RENDERED lamp brightness +
+   held/latch. The pump re-blits only windows whose token moved: ~31 pumps/2s →
+   4-15 window repaints vs 217-224 always-on. `BT_GLASS_DIRTY=1` logs the tally.
+3. **Palette generation in the token** (branch `glass-palette-token`): the
+   ColorMapper family (armor rosette tints, adpal/adpal2 damage flash) animates by
+   CLUT writes with ZERO pixel churn — invisible to a pixel checksum.
+   `SVGA16::paletteGeneration` is bumped by the palette writers
+   (`BuildSecondaryColor` only on a REAL entry change, since the flash alternates
+   palettes every Execute writing identical RGB at zero damage; full rebuilds
+   unconditionally) and folded into the token for palette-expanding windows
+   (`monoTint < 0`, or all under `BT_GLASS_MFD_PAL`). Without it the glass radar
+   held stale armor tints between pixel repaints while the D3D surround (which
+   re-expands every frame) tracked live.
+
 **Turning it OFF: `L4PLASMA=NONE` (also `OFF`/`0`, 2026-08-06) [T2].** `L4GREND` creates a
 marquee whenever `L4PLASMA` is set at all (`SCREEN` → the desktop window, anything else → a real
 `PlasmaDisplay` on that serial port), and the GLASS profile force-defaults it to `SCREEN` — so on
@@ -696,7 +696,7 @@ reconfigure/externalConfigure); parse-skip list EMPTY ([gskip]=0), all 50 attr b
 | 30 | sec: schematic CRITICAL view (cmCrit) | subsystem simulationState/damage | LIVE this audit: N-cycle shows the full subsystem list (GEN A-D, LOOP 1-6, HUD, SENSORS, GYRO, TORSO, weapons) | T2 | CORRECT |
 | 31 | sec: schematic HEAT view (cmHeat) | subsystem currentTemperature tint | #6 pixel-verified; re-cycled this audit (mask 0x450421→0x490421→0x510421) | T2 | CORRECT |
 | 32 | sec: view cycling (N / pod 0x15) | CycleDisplayMode → vtbl+0x4C @4d1ae4 | #6 resolution re-verified live ([mode] display notify 0/1/2) | T2 | CORRECT |
-| 33 | sec: CONTROL MODE lamp (BAS/MID/ADV) | ControlsMapper/**ControlMode** oneOfSeveralPixInt (row previously mislabeled DisplayMode — that is the sibling sdspmod) | attr wave incr.5; M-cycle verified #6. **2026-08-10:** this row's verification PREDATED the 07-25 #48 cycle-fill, which then lit the inactive box interiors (art idx 254 → live armor slot 62) — playtester-reported, re-fixed by mapping out-of-range indices to plane background (see context/gauges-hud.md §#48 REFINED); both render paths re-verified against the reference | T2 | CORRECT |
+| 33 | sec: CONTROL MODE lamp (BAS/MID/ADV) | ControlsMapper/DisplayMode oneOfSeveralPixInt | attr wave incr.5; M-cycle verified #6 | T2 | CORRECT |
 | 34 | sec: duck / searchlight button lamps | duckState (ctor-zeroed only) / Searchlight LightOn | duckState writer missing (P3 leftover) | T3 | DEFERRED-FEED (duck); CORRECT (light attr published) |
 | 35 | sec: messageBoard ticker | StatusMessagePool (NULL stub) + kill ticker strip 0 | 7fc4acb; kill ticker live 2026-07-12, other strips unsurveyed | T2/T3 | DEFERRED-FEED (partial) |
 | 36 | MFD preset paging (J/K/L, pod RIO banks) | SetPresetMode table @0051dbf0 (little-endian re-decode) | #9; re-verified live this audit: cycles visit EXACTLY the populated set (MFD1: 1,2,4; MFD2: 1-4; MFD3: 1,2) | T2 | CORRECT |
@@ -2104,7 +2104,19 @@ static unsigned long
 			if (svga == NULL) svga = static_cast<SVGA16*>(p->graphicsDisplay);
 		}
 		if (svga != NULL && combined != 0)
+		{
 			token ^= svga->PlaneChecksum(combined);
+			// PALETTE-ANIMATED content (2026-08-10): a palette-expanding window
+			// (the radar; MFDs under BT_GLASS_MFD_PAL) changes colour with ZERO
+			// pixel writes -- the ColorMapper family (armor rosette tints, the
+			// adpal/adpal2 damage flash) writes CLUT entries.  Fold the palette
+			// write generation in so those changes repaint; mono-tint windows
+			// don't read the palette and keep their pixel-only token.
+			static int sPalTok = -1;
+			if (sPalTok < 0) sPalTok = getenv("BT_GLASS_MFD_PAL") ? 1 : 0;
+			if (w->monoTint < 0 || sPalTok)
+				token = (token ^ svga->PaletteGeneration()) * 16777619UL;
+		}
 	}

 	// Lamps: each button's RENDERED brightness + held/latched, so a flash toggle or a
@@ -5445,6 +5445,7 @@ SVGA16::SVGA16(
   BuildWindows(init_width,init_height,windowed, secondaryIndex, aux1Index, aux2Index);
 	for (int _i = 0; _i < 10; _i++)	// DEV-COMPOSITE: lazily created on first surface draw
 		mDevSurfaceTex[_i] = NULL;
+	paletteGeneration = 0;			// GLASS dirty-skip palette tracking
 //STUBBED: VIDEO RB 1/15/07
 #	if defined(DEBUG)
 		Tell("SVGA16::SVGA16()\n");
@@ -7190,6 +7191,7 @@ void

 	svga_palette->paletteData.Valid = True;
 	svga_palette->modified = True;
+	((SVGA16 *) graphicsDisplay)->paletteGeneration++;	// glass dirty-skip (full rebuild)

 	Check_Fpu();
 }
@@ -7234,6 +7236,13 @@ void
 	SVGA16Palette
 		*svga_palette(&((SVGA16 *) graphicsDisplay)->palette[paletteID]);

+	// GLASS dirty-skip: track whether this write CHANGES anything -- the
+	// ColorMapper family calls this every Execute (the adpal/adpal2 flash
+	// alternates palettes even at zero damage, usually writing identical RGB),
+	// so bump the generation only on a real change or every glass window that
+	// palette-expands would repaint every pump for nothing.
+	int palette_changed = 0;
+
 	//-------------------------------------------
 	// If any of the ...TransparentZero modes are used,
 	// leave color zero undefined for this bit group by
@@ -7275,21 +7284,31 @@ void
 					{
 					case RedChannel:
 					case RedChannelTransparentZero:
+						if (destination_triplet->Red != source_triplet->Red)
+							palette_changed = 1;
 						destination_triplet->Red = source_triplet->Red;
 						break;

 					case GreenChannel:
 					case GreenChannelTransparentZero:
+						if (destination_triplet->Green != source_triplet->Green)
+							palette_changed = 1;
 						destination_triplet->Green = source_triplet->Green;
 						break;

 					case BlueChannel:
 					case BlueChannelTransparentZero:
+						if (destination_triplet->Blue != source_triplet->Blue)
+							palette_changed = 1;
 						destination_triplet->Blue = source_triplet->Blue;
 						break;

 					case AllChannels:
 					case AllChannelsTransparentZero:
+						if (destination_triplet->Red   != source_triplet->Red
+						 || destination_triplet->Green != source_triplet->Green
+						 || destination_triplet->Blue  != source_triplet->Blue)
+							palette_changed = 1;
 						*destination_triplet = *source_triplet;
 						break;
 					}
@@ -7310,6 +7329,8 @@ void

 	svga_palette->paletteData.Valid = True;
 	svga_palette->modified = True;
+	if (palette_changed)
+		((SVGA16 *) graphicsDisplay)->paletteGeneration++;	// glass dirty-skip

 	Check_Fpu();
 }
@@ -7413,6 +7434,7 @@ void

 	svga_palette->paletteData.Valid = True;
 	svga_palette->modified = True;
+	((SVGA16 *) graphicsDisplay)->paletteGeneration++;	// glass dirty-skip (full rebuild)

 #	if defined(TESTPALETTE)
 		std::cout << "L4GraphicsPort::BuildAuxiliaryPalette for port " <<
@@ -7523,6 +7545,7 @@ void

 	svga_palette->paletteData.Valid = True;
 	svga_palette->modified = True;
+	((SVGA16 *) graphicsDisplay)->paletteGeneration++;	// glass dirty-skip (full rebuild)

 	Check_Fpu();
 }
@@ -7557,28 +7580,17 @@ void
 	// LEFT AS HEAP GARBAGE and any pixmap pixel >= that count wrote the
 	// garbage's high bits into other displays' planes (the #48 stray
 	// blocks; convicted live by BT_PLANE_AUDIT -- the 480x640 radar
-	// background carries index 217).  Map the remainder to the plane
-	// BACKGROUND (translationTable[0]): high-index art renders as the
-	// port's background colour in-plane, and can never leak.
-	//
-	// REFINED 2026-08-10 (the sec-surface phantoms): the first #48 fill
-	// cycled the remainder IN-PLANE (index mod 2^bits), which mapped art
-	// index 254 onto plane slot 62 -- a LIVE colorMapperMultiArmor damage
-	// slot.  SMODE.PCC's inactive control-mode box interiors and a stray
-	// 52x13 idx-254 bar baked into BTSEC1.PCX (between the heading dial
-	// and the armor rosette) lit up in the current right-armor damage
-	// colour on every render path.  On the shipped machine those regions
-	// rendered BLACK (the heap garbage's low plane bits were 0 ->
-	// in-plane index 0), so background IS the authentic on-screen result.
-	// (The 1995 binary shipped the same 64-entry fill and relied on art
-	// discipline; garbage is not a preservable behaviour, so this stays a
-	// guarded PORT deviation.)
+	// background carries index 217).  Cycle the in-plane pattern across
+	// the remainder: high-index art degrades to its (index mod 2^bits)
+	// colour IN-PLANE, and can never leak.  (The 1995 binary shipped the
+	// same 64-entry fill and relied on art discipline; garbage is not a
+	// preservable behaviour, so this is a guarded PORT deviation.)
 	{
 		int filled = 1;
 		for (int b = bitMask & 0xFF; b != 0; b &= (b - 1))
 			filled <<= 1;
 		for (int i = filled; i < 256; ++i)
-			translationTable[i] = translationTable[0];
+			translationTable[i] = translationTable[i & (filled - 1)];
 	}
 	Check_Fpu();
 }
@@ -382,6 +382,15 @@ protected:
 	SVGA16Palette
 		palette[PaletteCount];

+	// GLASS dirty-skip (2026-08-10): bumped by the palette writers
+	// (L4GraphicsPort::BuildSecondaryColor on a REAL entry change; the full
+	// palette rebuilds unconditionally) so the glass repaint token can see
+	// palette-only animation -- the ColorMapper family (armor rosette tints,
+	// damage flash) changes COLORS with zero pixel writes, which the pixel
+	// checksum alone can never catch.
+	unsigned long
+		paletteGeneration;
+
 private:
 	int NUMGAUGEWINDOWS;
   HWND *gaugeWindows;
@@ -425,6 +434,11 @@ public:
 	// glass repaint pump re-blit ONLY the windows whose plane actually changed
 	// (idle MFDs / static panels skip; the sweeping radar keeps updating).
 	unsigned long PlaneChecksum(int mask) const;
+
+	// GLASS dirty-skip: the palette write generation (see paletteGeneration).
+	// Folded into the repaint token of palette-expanding windows so ColorMapper
+	// palette animation (armor tints / damage flash) repaints without pixel churn.
+	unsigned long PaletteGeneration() const { return paletteGeneration; }
 };

 //########################################################################