Anno 1602 demo (Max Design / Sunflowers, 1998)
lib/apps.js id anno1602_demo (localhost-only), exe
test/binaries/win98-games-a-d/Anno1602-demo-SW/extracted/1602.exe with the
app-local Maxnet.dll, Maxsound.dll and Language.dll (smackw32.dll
auto-loads beside the exe; msvcrt.dll comes from the shared DLL set).
The download Anno1602Demo-SWonly.exe is a RAR self-extractor whose
contents are the game tree itself (470 entries, 37 MB). 7-Zip lists it but
has no codec for this RAR version ("Unsupported Method"); it was extracted with
node-unrar-js (createExtractorFromFile). Manifest:
node tools/gen-win98-games-a-d-manifests.js. No registry or CD check.
Software DirectDraw at 800x600 (the CLI canvas is 640x480; ctl clicks are in
capture coordinates and go through the presentation transform). Input is plain
window messages. It runs clean: no unimplemented API on the route below.
Route to gameplay (2026-10-06, ~140k batches at the default clock)
node test/run.js --app=anno1602_demo --quiet-api --control=8187 --frozen \
--max-seconds=600 --max-batches=100000000 | grep -a -v 'EIP='
- step 15000 -> main menu.
- click "A New Country" (365,268) -> assignment text; click Start Game (165,389).
- step 30000 -> "Enter your name" hall; Enter accepts "Anonymous"; click the red flag (135,260); step 40000 -> sea map, the player's ship "Laguna Verde" selected.
- Left-click the ship (258,247), then left-click open water (120,120): it sails there. Right-click is back/cancel (it deselects), not "move".
Evidence: scratch/runs/20261006T1340Z-anno1602_demo-gameplay-w6.
Black screen in the browser under lazy loading (fixed ef8bf49c0, 2026-10-11)
The game reads haeuser.cod, figuren.cod and every gaddata\*.gad one
byte at a time with fgetc (hundreds of thousands of calls at startup). In
the browser those files stream lazily. The first byte parked, and getc's retry
ran as fgetc(NULL): getc used its own FILE* slot as the byte buffer and the
park left it zeroed. Every file then looked empty, and the main-menu loop ran
over an all-zero primary (--trace=dx: nzBytes=0 on every present). The
CLI's provider is synchronous, so the CLI never parks and never showed it.
Whether a browser run hit it depended on chunk residency, so it came and went
between builds. Diagnosis tools that worked: web-input-probe --trace=dx
(present non-zero bytes per DIB) and --trace-api=fopen,fgetc,fclose (the
argument to the retried call). Evidence:
scratch/runs/20261011T1730Z-anno1602-lazy-fgetc-claude1863.