pasmo and z80dasm to assemble and disassemble, and Python 3 for the tools. Nothing else.
The cartridge is not distributed with this repository: you need your own copy, named timepilot.rom in the project root. It is exactly 16384 bytes with this sha256:
183e80262301b18d41762d64a2fc326f4a4bef17832109225637e184d54a70d9With any other dump the listing will not reassemble. make comprueba tells you in one line.
make # trace, generate the listing and check everything
make verify # assemble the listing and compare its sha256 with the cartridge
make sanity # what reassembly cannot catch
make test # the tests on the listing, which do not need the cartridge
make imagenes # redraw the reconstructed pictures
make web # the pictures and these pagesmake chains the first four. If all goes well, the line that matters is this one:
ensamblado : 16384 bytes 183e8026...4d54a70d9
original : 16384 bytes 183e8026...4d54a70d9
OK: reproducible byte a bytesrc/timepilot.asm | the commented listing, generated; never edited by hand |
src/timepilot.notes | the annotations: labels, comments, headers and data ranges, anchored to addresses |
src/timepilot.entries | the entry points the trace cannot deduce, each with its justification |
src/timepilot.nocode | the zones the tracer must not read as code |
tools/ | the tracer, the listing generator, the checks and the drawing tools |
tests/ | the tests on the listing and the notes |
medidas/ | the raw output of the openMSX measurements |
docs/ | this site |
work/ | what make produces along the way |
Every routine has an uppercase name and, where it matters, a comment saying what it does. Data blocks are labelled DATA_<use>, with the width of their structure and an explanation of what they are and how that is known. Addresses are the real ones of the cartridge in page 1: 0x4000-0x7FFF.
To change anything, edit src/timepilot.notes and run make again: the listing is regenerated and the checks say whether it still holds.
The tracer (tools/z80trace.py) follows the flow from the header's entry point and from what timepilot.entries declares. What cannot be deduced on its own here is eight dispatch tables: Time Pilot does not put the table right after the call, the way other cartridges do; it loads it with a ld hl,nn and jumps through a jp (hl) that lives somewhere else. Each one is declared, with the instruction that loads it alongside.
Whatever is not code is left as a gap, and every gap is closed by finding the instruction that reads it (tools/quien_apunta.py) and checking that the format matches what the consuming code does with it.
What cannot be read is drawn. tools/graficos.py uploads to a make-believe video memory exactly what the cartridge uploads, at the same addresses, and builds the screen from its own label lists: if the reading were wrong, it would come out as noise.
tools/densidad.py | measures, routine by routine, how much of the listing carries a comment, and lists the ones below the bar |
tools/valida_c.py | catches comments anchored mid-instruction, and duplicates |
tools/quien_toca_la_ram.py | for each RAM byte, which instructions name it and from where. This is what caught the errors in the RAM map |
tools/quien_apunta.py | who reads a hole in the data, which is how holes get closed |
tools/marca_konami.py | looks for Konami's hidden mark at the end of the ROM |
tools/graficos.py | rebuilds video memory and draws the screens |
tools/omsx_*.tcl | the measurements in openMSX, told in In the emulator |