Tetris for Windows (WEP1 / Win16)

Gameplay regression correction — 2026-09-19

The old test's first failure required an active blue main caption while the modal About DLL was open. The saved native Win98 reference has an inactive gray caption at exactly that stage: test/output/win16-v86-comparison/wep16_tetris/native.png, with capture/payload metadata in native.json (2026-08-22). Current local captures show the same activation sequence: gray during About, blue after dismissal. The test now requires both states, rather than treating gray as missing paint.

Microsoft documents the active/inactive nonclient distinction in WM_NCACTIVATE. The native capture, not the modern documentation alone, is the evidence for this particular Win98 modal state. This does not prove every About pixel is native-identical: the native capture uses 16-color VGA and a taskbar, and dialog background/text differences are outside this test correction.

The former piece assertions also assumed magenta followed by green. Fresh captures instead show cyan/yellow or gray/cyan sequences. Gray is a valid piece, so a saturation-only predicate is also wrong. The test now samples nonblack colors from the active piece inside the black board and from the Next preview (excluding its gray background), then follows those colors to the settled piece and successor. The checked board areas exclude gray borders.

Timing matters: with the previous 200 ms/batch clock, removing Down still allows gravity to settle a piece before the final screenshot. Use 20 ms/batch, start via F2 at 55/56, capture at 150, Down at 170/171, capture at 230, stop at 245. This leaves time for spawn/painting without letting gravity substitute for the hard drop.

The top-of-board pixel minimum changes from color-specific 250/400 to 100: an entering piece may expose only one tile (149 nonblack pixels), depending on shape/spawn phase. In exchange, the settled-piece check is strengthened from >100 to >500 matching pixels (four tiles contribute 596), with an additional exactly-empty bottom-before-drop check. The >600 changed-pixel board assertion and all About-logo/client-background checks remain.

An in-memory negative variant with Down removed fails specifically at the settled-piece check. Local logs: /private/tmp/wa-tetris-corrected2.log, /private/tmp/wa-tetris-no-down2.log. No runtime behavior changed. The complete WEP1 suite passes 8/8, with two additional standalone Tetris passes. Logs: /private/tmp/wa-tetris-wep1-suite.log and /private/tmp/wa-tetris-repeat.log (plus the focused corrected run above).

node test/test-win16-wep1-gameplay.js tetris

Initial erase versus synchronous UpdateWindow — 2026-09-20

Enabling synchronous Win16 UpdateWindow exposed a runtime ordering bug: USER.42 ShowWindow(SW_MAXIMIZE) posted WM_ERASEBKGND for main HWND 0x10001; USER.124 then painted the tiled client immediately; the queued erase later reached USER.107 DefWindowProc and covered the tiles with the gray class brush. The opening About logo survived, but its exposed parent was flat gray. Trace: /private/tmp/wa-update16-tetris-trace.log; before/after captures: wa-update16-tetris-before.png, wa-update16-tetris-after.png.

ShowWindow now completes its initial erase through a far continuation rather than posting it. Tetris's existing startup-background and hard-drop checks pass again, as does the full eight-game WEP1 suite. This was an emulator ordering regression, not a reason to relax the native-reference assertions.

Modal About caption and foreground (fixed 2026-10-10, e82eaec16)

test-win16-wep1-gameplay failed on main: "Tetris caption must remain visible but inactive behind modal About".

• The About box is a Win16 DialogBox (USER.87); $win16_dlg_run showed it but never activated it.
• Since aba07804c (2026-09-20), the renderer's foreground is only what was explicitly activated, no longer the top of the z-order. A top-level caption is drawn active iff host_foreground_window() is that window, so Tetris kept a blue caption under its own modal dialog.
• The dialog is now published as foreground once shown, and the foreground goes back to a visible top-level owner when it ends. $active_hwnd never left the owner, so the focus transaction at dialog end saw no transition and would not republish it.

Gameplay A/B on --tick-ms-per-batch=20 (About OK at batch 40, F2 at 55):

• Down at batch 170 hard-drops the piece and spawns the next one.
• Without input, the piece falls by gravity at ~20 px per guest second, about 2 rows/s at level 1. It is linear over 1.6 s and 5 s windows.

Evidence: scratch/runs/20261010-wep16_tetris-drop-ab.