Enigma 1.01 (SDL 1.2 + embedded Lua)
App summary. The dated enigma-*-2026102x.md files beside this one hold the
full investigations and their evidence. Enigma is not registered in
lib/apps.js. Every run below is a diagnostic route on a historical module,
not a current-main or gameplay claim.
Binaries and route
- The shipped Windows build packs all ten PE modules with UPX. Authentic
upx -doutput recovers 849 imports over 40 module edges. The packed tables show only 81. Seeenigma-authentic-static-unpack-20261024.md. The working route runs those unpacked diagnostic modules, never the packed originals. - Asset tree: 2460 files. Mount it with one
--vfs-tree. Per-file comma mounts are wrong for this tree. - Use the native providers, listed with
--dlls, in dependency order:msvcrt, zlib1, libpng12-0, SDL, SDL_image, SDL_mixer, libfreetype-6, SDL_ttf. With--dlls=msvcrt.dllalone, SDL is never preloaded. The first SDL import then hits the built-in stub (SDL_getenv("SDL_STDIO_REDIRECT"), EIP0x5de8e0, IAT0x9ef7bc). That is a configuration error, not a missing emulator API.--dllsreplaces automaticresolveDllGraph. jpeg.dll,mikmod.dllandloadall.dllare optional or conditional. None is proven necessary.- Lua C-module loading is
package.loadliband the cpath searchers. ItsLoadLibraryAcall is at0x5ab960, in the function at0x5ab8f0. The default cpath is.\?.dll;!\?.dll;!\loadall.dll. Seeenigma-register-load-cause-20261024.md.
Startup progression (Oct 24–26)
(Recovered from unmerged branches task/enigma-*-oct26, Oct 2026.)
- With the eight providers, startup passes SDL init. It creates the
SDL_appwindow and readscompat.lua,startup.lua,enigma_conf.luaand model scripts. No crash, unimplemented API or Lua error appeared. data/enigma_conf.luais a zero-byte file. A staging step that skips empty files breaks the tree.- Filesystem logging is very verbose. A 131072-byte output guard hit while the
game was still reading
data/gfx32/*.png. That was a harness limit, not a guest fault. - At batch 30000, a lossless 640x480 frame showed the title, logo, marble background and "Loading Enigma levels". That run reached batch 97160 (4.19M API calls) in 55 s with no trap.
- The Main Menu (Start Game, Level Pack, Options, Credits, Quit) appears by about batch 120000–130000. A single frozen 100000-batch control step took longer than a 45 s response wait. Small steps of 256 batches or fewer were reliable. At about 20.5k batches the game was still loading assets.
Start Game click does not transition (open)
- A click at screen (344,212), batch 120833, was delivered by the host. The
trace shows
check_inputreturning WM_NCHITTEST (0x84), WM_LBUTTONDOWN (0x201) and WM_LBUTTONUP (0x202) to HWND0x10001. Down/up lParam was0x00a90141, client (321,169). - The screen-to-client offset is (23,43). That matches the decorated SDL window
(style
0xca0000). The client point still falls inside the Start Game item, so no coordinate bug is shown. - Five seconds after the click, the frame was byte-identical to the menu. No gameplay was reached. Cause unknown.
- Hypothesis, unproven:
test/run.jssetsrenderer._buttonHoldFloorMs = -1, so an ordinaryclickdelivers down and up in the same batch. SDL apps that sample mouse state each frame (SDL_GetMouseState) rather than the event queue can miss such a press. Next steps:- Use separate
mousedownandmouseupevents with a gap of a few batches. - Add
--trace-input-dispatch --trace-mouse-stateto see whether SDL dequeues and dispatches the message.
- Use separate