LBALab/lba2-classic-community

Reported by EPmager against the 14 Aug build (99c99ab), Windows, retail GOG data. gfx_dither does soften the 16-step ramp banding on terrain shade elements, but on evenly lit surfaces it lays down an observable checkerboard, which blurs into interlaced horizontal lines while the auto camera moves. EPmager asks that it not be turned on by default until the checkerboarding is reduced. The default is already 0 (LIB386/pol_work/POLYGTEX.CPP:36, and docs/GFX_OPTIONS.md documents all three gfx options as off), and has been since the option landed in 817dda1. Recording the ask so it stays that way. Why the pattern is visible GTEX_ShadeRow (LIB386/pol_work/POLYGTEX.CPP:45) promotes a pixel by one whole shade row when the Gouraud sub-row fraction beats a 4x4 Bayer threshold: if (level < 15 && (decZ_gour & 0xFF) > GTEX_Bayer4[((py & 3) << 2) | (px & 3)]) level++; Two properties follow from that, and they are the two symptoms in the report: The amplitude is a full shade row. The dither's job

Reported by EPmager against the 14 Aug build (99c99ab), Windows, retail GOG data. With LBA2_ISO_DIV set, the Display submenu's Resolution row shows the divided interior render size while the player is standing in an interior, so it reads as though their resolution has changed. Someone opening the menu to check what their full resolution is gets the scaled one instead. What happens Display_FormatCurrentLabel and Display_FindCurrentEntry both read ModeDesiredX / ModeDesiredY (SOURCES/RES_MENU.CPP:46 and SOURCES/RES_MENU.CPP:63). Those hold whatever the current scene is rendering at, and an interior with LBA2_ISO_DIV=2 renders at half the chosen size. Two visible effects, both confined to interiors: The Resolution: <value> row reads 640x360 for a player who chose 1280x720. The divided size is usually not a catalog entry either, so the label falls to the bare %ux%u branch and loses the (Classic) tag in the case where it applies. The resolution sub-page's green "current" marker lands on no

Follow-up to issue #176 and PR #251. Symptom In the cold-start attract reel (SHIFT+D from the main menu, idle attract, or the SlideDemo debug entry), cube 201 — the Desert Island "bat scene" — sometimes derails in interactive variable-dt play. The RNG-driven bat can collide with Twinsen and snag him off his authored track. The scripted SET_CUBE opcode that ends the scene then never fires, and the reel sits indefinitely on a stuck hero until the #251 watchdog fires. Observed on macOS during PR #251 development. Whether the derail is Mac-only or also reproduces on Linux / Windows interactive runs at certain CPU speeds is not currently established — the harness path (where reproduction would normally be pinned) plays cleanly post-#251 because the canonical RNG seed is deterministic. The remaining divergence is in the interactive variable-dt cadence path, which is exactly the path our regression tests don't cover. Reproduction reports from any platform welcome. If you've seen the att