Emulation and testing
The project
Status: Solved. Every machine boots what it should and the 3 × 3 disk matrix
is measured; what is left is named in §7 — the two differing bytes in the
-blank.imd boot area are unidentified, ROM 1.3 is undumped, and the Nuevo
CBIOS 1.5 hang is recorded rather than chased.
The answer. MAME models three Osborne 1 variants and this project patches in a
fourth. One osborne1.zip serves all three stock machines; the only ROMs absent
are the four BIOS revisions that have never been dumped anywhere (§3). All three
-blank.imd system disks boot on all three machines — the 52-/80-/104-
prefix is a label, not a gate (§4). The one image that does not boot is the Nuevo
80-column system disk, research/nuevo/OS1NUEVO.IMD, and that is closed as
not-needed (#79).
| Question | Answer | Where |
|---|---|---|
| Which machines are there? | osborne1 (52 columns), osborne1sp (104), osborne1nv (80), plus this project’s patched osborne1nx (#191) |
§2 |
| Which ROMs does each need? | one osborne1.zip serves all three; only the four never-dumped BIOS revisions are absent |
§3 |
| Which floppy images boot? | all three -blank.imd on all three machines; OS1NUEVO.IMD on none |
§4 |
| How is it run, and how is it tested headless? | windowed by hand; the Lua harness otherwise | §5, §6 |
The install/run mechanics are adapted from the sibling project
Myrcurial/o1prsnt — its
o1prsnt/docs/emulating-the-osborne-1.md and
o1prsnt/utilities/headless-testing-harness/ — which is where the macOS
specifics and the VRAM-dumping technique were worked out. This document adds the
per-machine ROM requirements and the boot-disk compatibility results, both
measured here. (Paths prefixed o1prsnt/ are in that repository, not this one.)
1. Install
brew install mame
brew install cpmtools # cpmcp / cpmrm / cpmls - file-level disk work
disk-analyse (Keir Fraser’s Disk-Utilities)
is needed to convert between the IMD images MAME wants and the HFE images a
hardware floppy emulator wants. It is not packaged; build it:
git clone https://github.com/keirf/disk-utilities.git
cd disk-utilities && make clean && make && sudo make install
samdisk (TD0 → EDSK) is not currently installed on this machine and is
needed for the TeleDisk images in research/ (Osmosis, RT-60A, the Zorba/Kaypro
CoPower-88 sets).
2. The machines — three in MAME, and this project’s fourth
MAME models three Osborne 1 variants; this project patches in a fourth. Two of MAME’s three are clones of the stock machine; one has its own ROMs:
| MAME machine | Description | Year | Parent | Relationship | Display (measured) |
|---|---|---|---|---|---|
osborne1 |
Osborne-1 | 1981 | — | the stock machine | 416×240 = 52 columns |
osborne1sp |
Osborne-1 with SCREEN-PAC | 1983 | osborne1 |
clone — no ROM differences at all | 832×240 = 104 columns |
osborne1nv |
Osborne-1 (Nuevo Video) | 1984 | osborne1 |
own BIOS and own character generator | 640×240 = 80 columns |
This project adds a fourth (tools/emulator-setup/mame-osborne1nx/, #191):
osborne1nx is a clone of osborne1 whose Z80 sits behind an emulated card
that decodes the full 8-bit I/O address, so the MD-10’s ACB-4000 host adapter can
answer on ports 30h–33h. It is not in MAME; it is a patch, and it runs the
1.44 ROM only. Banking is identical to osborne1 — measured, not assumed —
and the presence check the MD-10 driver demands passes
(docs/osborne1nx-machine.md).
Its video is not. osborne1nx presents the interposer’s video personality,
which is the SCREEN-PAC (#24), so its driver derives from osborne1sp: it
answers 2400h’s presence bit as fitted, takes the board’s register there, and
boots in the 104-column window: 832x240, twice the stock machine’s width,
with every dot of every character drawn — the same model as MAME’s osborne1sp.
Keeping the stock width by dropping every other dot column was tried and rejected:
half a character’s pixels makes the font unreadable (docs/osborne1nx-machine.md
§8).
The Nuevo machine is 80 columns and the SCREEN-PAC is 104. The display width
is a property of the emulated machine, not of the disk: it was measured by
booting the same utility disk on each of the three and reading the screen device’s
dimensions (tools/research/mame-boot-probe.sh). Screenshots, all booted from
tools/emulator-setup/floppies/:
| Screenshot | Machine | Pixels | Columns |
|---|---|---|---|
docs/mame-screens/osborne1-52col.png |
osborne1 |
416×240 | 52 |
docs/mame-screens/osborne1nv-80col.png |
osborne1nv |
640×240 | 80 |
docs/mame-screens/osborne1sp-104col.png |
osborne1sp |
832×240 | 104 |
That is also why the disk prefix does not matter for booting (§4): the same image displays 52 columns on stock hardware and 80 on Nuevo, because the video window is the machine’s, while the text layout is the CBIOS’s.
The SCREEN-PAC is modelled as hardware only (the 80/104-column logic lives on
the board, not in a ROM), whereas Nuevo shipped a replacement monitor ROM. That is
independent evidence that the two 80-column upgrades are different designs — see
#73 and docs/o1-rom-variants.md.
The SCREEN-PAC has three modes, and MAME models one of them
The measured 104 columns is the SCREEN-PAC’s default, not its only width. Its own user guide gives the full set, and all of them are selectable by the stock BIOS:
SCREEN-PAC has preset video contrast in 80 and 104 column format … In these same formats, SCREEN-PAC provides horizontal scrolling by 32 column blocks … At BIOS location E1A5H, POKE or store 0 for 52 columns / 1 for 104 columns / 3 for 80 columns. CALL BIOS location E4E5H. — Osborne 1 SCREEN-PAC User Guide, p. 13 (PDF p. 17)
So 52 / 80 / 104 are three widths of one window (o1-memory-io-map.md §2a),
and the stock ROM already contains the routine that switches between them —
which is why “no ROM differences at all” is the right description of
osborne1sp. MAME models the 104-column window; the 80-column mode is selected
by poking E1A5H = 3 and calling E4E5H — read from the user guide, and the
poke itself is not yet measured here.
Presence flag. The same page records how software detects the upgrade with no ROM change:
reading 2400H (Bank 2) returns data bit 2 LOW. Without SCREEN-PAC, this bit is read HIGH. This feature allows different screen responses and menus depending on whether the Osborne is fitted with SCREEN-PAC.
2400H in bank 2 sits inside the memory-mapped I/O window and is recorded there
as a read-only presence bit, not a register (o1-memory-io-map.md §3).
3. ROM requirements
Every ROM any of the three machines needs, with the CRC32 MAME asserts against.
Full inventory and provenance: docs/o1-rom-variants.md.
| ROM file | Size | CRC32 | osborne1 |
osborne1sp |
osborne1nv |
Role |
|---|---|---|---|---|---|---|
rev1.40.ud11 |
4096 | 3d966335 |
✓ (BIOS 1.4) | ✓ | — | monitor ROM 1.4 |
rev1.43.ud11 |
4096 | 91a48e3c |
✓ (BIOS 1.43) | ✓ | — | monitor ROM 1.43 |
3a10082-00rev-e.ud11 |
4096 | c0596b14 |
✓ (BIOS 1.44) | ✓ | — | monitor ROM 1.44 — the default |
char.ua15 |
2048 | 5297c109 |
✓ (BIOS ≤ 1.4) | ✓ | — | early character generator |
7a3007-00.ud15 |
2048 | 6c1eab0d |
✓ (BIOS 1.43/1.44) | ✓ | ✓ | late character generator |
monrom-rev1.51-12.ud11 |
4096 | 298da402 |
— | — | ✓ | Nuevo monitor ROM 1.51 |
character_generator_6-29-84.14 |
2048 | 6c1eab0d |
— | — | ✓ | Nuevo-era char-gen — same bytes as 7a3007-00.ud15 |
osba.bin, osb12.bin, osb121.bin, osb13.bin |
4096 | — | NO_DUMP | NO_DUMP | — | BIOS A / 1.2 / 1.2.1 / 1.3 — never dumped |
One osborne1.zip serves all three machines. The repository’s
tools/emulator-setup/roms/osborne1.zip holds the six available files, and
MAME’s hash fallback satisfies character_generator_6-29-84.14 from
7a3007-00.ud15 because the bytes are identical. Verified:
$ mame osborne1nv -verifyroms -rompath tools/emulator-setup/roms
romset osborne1nv [osborne1] is good
$ mame osborne1 -verifyroms -rompath tools/emulator-setup/roms
osborne1 : osba.bin (4096 bytes) - NOT FOUND - NO GOOD DUMP KNOWN
... (the four undumped BIOSes only)
romset osborne1 is best available
“best available” is the correct and expected result for osborne1/osborne1sp:
the only missing ROMs are the four that have never been dumped anywhere. The
default BIOS is 1.44 (3a10082-00rev-e.ud11), which is what the banner
shows.
Where to get them
The upstream collection used by o1prsnt is
archive.org/details/osborne-1-roms
(two files: occ-7a3007-00-reva.rom, occ-v1.44.rom — enough for osborne1
and osborne1sp, not enough for osborne1nv). This repository carries the
full set in tools/emulator-setup/roms/, archived with provenance in
research/roms/.
Rompath layouts
MAME looks for machine ROMs in <rompath>/<machine>.zip or
<rompath>/<machine>/, and it will also fall back to a loose file in the
rompath root when a required ROM is missing but a file with the right CRC is
present. Measured, both ways:
$ mame osborne1nv -verifyroms -rompath tools/emulator-setup/roms
romset osborne1nv [osborne1] is good # loose 7a3007-00.ud15 satisfies
# character_generator_6-29-84.14 by CRC
$ # the same files in a plain <rompath>/osborne1/ directory:
osborne1nv : character_generator_6-29-84.14 (2048 bytes) - NOT FOUND (osborne1)
So the CRC fallback works for loose-root files and for files inside a zip,
but not for files inside a romset directory. The wrapper’s temporary
<rompath>/osborne1 symlink (used when the rompath has no osborne1.zip) turns
a working loose rompath into a failing directory — which is why
tools/emulator-setup/roms keeps osborne1.zip alongside the loose files. This
matters when substituting a third-party ROM: build a zip, see §5.
4. Boot disks: what actually boots on which machine
Measured, not assumed. Each cell is a headless run of the harness in §6 that
presses RETURN at the BIOS prompt and waits for an A>.
| Image | osborne1 |
osborne1sp |
osborne1nv |
|---|---|---|---|
tools/emulator-setup/floppies/52-blank.imd |
BOOTED | BOOTED | BOOTED |
tools/emulator-setup/floppies/80-blank.imd |
BOOTED | BOOTED | BOOTED |
tools/emulator-setup/floppies/104-blank.imd |
BOOTED | BOOTED | BOOTED |
research/nuevo/OS1NUEVO.IMD |
no-boot | — | no-boot |
research/nuevo/OS1NUEVO.IMD (pristine upstream copy) |
— | — | no-boot |
The 52-/80-/104- prefix is a label, not a gate
All three images boot on all three drivers, because all three are the same CP/M system disk — they differ by exactly two bytes:
$ cmp -l 52-blank.img 80-blank.img
5737 377 0 # 52 = 0xFF, 80 = 0x00
5798 0 3 # 52 = 0x00, 80 = 0x03
$ cmp -l 52-blank.img 104-blank.img
5737 377 0 # 104 = 0x00
5798 0 1 # 104 = 0x01
Both offsets are in the boot area (track 1, sector 1), which is where a CBIOS would keep a video-mode selection. Whatever those two bytes select, they do not change which machines the disk boots on.
So what does gate bootability is the CBIOS on the disk, not the filename
prefix. The prefix is a useful human label — keep using it — but do not rely on
it to decide the driver. tools/check_o1_map.sh still derives the driver from it
(52- → osborne1, 104- → osborne1sp, 80- → osborne1nv), warns when it
selects anything but the stock machine, and takes O1_MACHINE as an override;
tools/README.md states the same rule. That inference is §7 item 1.
What the -blank.imd disks actually are
Despite the name, they are not blank — they are Osborne 1 double-density CP/M 2.2 system disks, IBM-MFM, 40 tracks × 5 × 1024 bytes, single-sided:
autost.com cbas2.com copy.com crun2.com mbasic.com movcpm.com
pip.com setup.com sysgen.com xdir.com xref.com
The system area identifies them as Osborne Computer System / 59k CP/M vers 2.2
/ CBIOS 1.4, with SETUP VERS. 2.6 and PIP VERS 1.5. AUTOST.COM runs
XDIR (the Extended Directory version 3.5 banner) rather than booting to a bare
prompt — the same behaviour o1prsnt’s DISCOVERY.md records.
research/nuevo/OS1NUEVO.IMD — the Nuevo 80-column system disk
This is the Nuevo boot disk. Its label is
CP/M 2.2 system disk for Osborne 1 w/ Nuevo DD & 80 column mods, and its
system area carries a CBIOS that is explicitly the 80-column one:
Osborne 1 Computer / CP/M vers 2.2 / CBIOS vers 1.5
* THIS IS AN EIGHTY (80) COLUMN DISPLAY FOR THE OSBORNE ONE. *
* (3) FULL SCREEN REVERSE VIDEO CAPABILITY (JUMPER SELECT).
* (4) CURSOR PROGRAMMABLE AS UNDERLINE OR REVERSE VIDEO BLOCK.
* (6) REVERSE VIDEO BLOCK CAPABILITY (INSTEAD OF UNDERLINE; JUMPER SELECT).
It does not boot under MAME. On osborne1nv it loads, prints the CBIOS 1.5
banner, and then sits there — screen unchanged over 260 emulated seconds, with
RETURN and space nudges. On the stock osborne1 it does the same.
That is not a bad copy: the pristine upstream file from Dave Dunfield’s
osborne1.zip behaves identically, and our copy differs from it only in the
IMD comment block. So either MAME’s osborne1nv does not reproduce whatever the
Nuevo CBIOS waits on, or the disk needs the physical Nuevo card to get past
initialisation. Closed as not-needed — see #79.
Decision: the Nuevo path is not pursued
The video personality (#24) does not depend on answering the above. The decision taken is to build it from the ScreenPac design plus OZROM’s soft 52/80/104-column switch, and to treat the Nuevo board as prior art (#73) rather than as a target. So the Nuevo CBIOS hang is not on the critical path, and #79 is closed with the question recorded rather than chased — nothing in the Nuevo design has yet turned up a feature worth the validation cost.
Practical answer for Nuevo work
- To get a prompt on
osborne1nv, use any of the three-blank.imddisks. - To exercise the Nuevo 80-column video path,
OS1NUEVO.IMDis the right disk but it cannot get past CBIOS initialisation; per the decision above this is not a blocker for #24. - Material for building a working Nuevo system disk, if anyone ever wants it:
OS1NUEVO.IMDcarries the Nuevo CBIOS 1.5 in its system tracks, the-blank.imddisks carrySYSGEN.COM(andMOVCPM.COM) for writing a new system, andresearch/nuevo/has the Nuevo DD upgrade manual and the Nuevo BIOS ROM (monrom-rev1.51-12.ud11). A custom build is therefore plausible.
5. Running it
Directory layout
o1prsnt settled on a stable layout so that launching MAME from odd places does
not scatter cfg/ and snap/ directories everywhere. Our tooling defaults to
the same paths, so both projects share one tree:
~/Documents/Osborne1/
roms/ osborne1.zip
floppies/ .imd / .hfe disk images
o1prsnt/ working copies
presentations/ slide sources
tools/check_o1_map.sh and tools/emulator-setup/ read
~/Documents/Osborne1/roms and ~/Documents/Osborne1/floppies by default;
override with O1_ROMPATH and O1_FLOPPY.
Windowed launch
cd ~/Documents/Osborne1
mame osborne1 -flop1 ~/Documents/Osborne1/floppies/52-blank.imd
mame osborne1sp -flop1 ~/Documents/Osborne1/floppies/104-blank.imd
mame osborne1nv -flop1 ~/Documents/Osborne1/floppies/80-blank.imd
The machine name is explicit — do not let the image name imply it (§4).
macOS: MAME forces fullscreen. LEFT Option + Return toggles out of it, and Cmd-Q is not honoured — use the red window button.
File-level disk work (cpmtools)
cpmls -f osborne1 IMAGE # list
cpmcp -f osborne1 IMAGE host.txt 0:CONTENT.TXT # copy in
cpmrm -f osborne1 IMAGE 0:CONTENT.TXT # remove
The osborne1 disk definition matches the double-density format these images
use (40 × 5 × 1024).
IMD ↔ HFE
For a hardware floppy emulator:
disk-analyse in.imd out.hfe
disk-analyse in.hfe out.imd
Running with a third-party ROM (OZROM 1E)
MAME cannot be told about a BIOS that is not in its ROM definitions, but it will load a file whose bytes do not match the expected hash — it warns and carries on. So a third-party monitor ROM is run by presenting it under one of the stock BIOS filenames. Build a romset zip, not a plain directory:
mkdir -p /tmp/ozbuild/osborne1 && cd /tmp/ozbuild/osborne1
cp <repo>/tools/emulator-setup/roms/*.ud11 \
<repo>/tools/emulator-setup/roms/*.ud15 \
<repo>/tools/emulator-setup/roms/*.ua15 .
cp <repo>/research/roms/OZROM-1E/OZROM_1E.BIN 3a10082-00rev-e.ud11
cp 7a3007-00.ud15 character_generator_6-29-84.14 # see note below
mkdir -p /tmp/ozset && zip -q -X /tmp/ozset/osborne1.zip *
O1_ROMPATH=/tmp/ozset tools/research/mame-boot-probe.sh \
osborne1 tools/emulator-setup/floppies/52-blank.imd 120 /tmp/ozrom.png
Why a zip and not mkdir .../osborne1. The simpler directory recipe boots
osborne1 and osborne1sp, but fails on osborne1nv: that machine needs
character_generator_6-29-84.14, which MAME satisfies by CRC fallback from
7a3007-00.ud15 only when the files are loose in the rompath root or inside a
zip — not when they sit in an osborne1/ directory. Copying the chargen under
its own name (it is byte-identical, o1-rom-variants.md) makes the fallback
deterministic for all three. Verified: mame osborne1nv -verifyroms -rompath
/tmp/ozset reports good.
OZROM 1E boots and runs on all three machines. Measured:
osborne1 416x240 (52 columns) -> A>
osborne1sp 832x240 (104 columns) -> A>
osborne1nv 640x240 (80 columns) -> A>
banner: OZROM 1E #00187 / (c) 1984 Micro Management
MAME prints “the machine might not run correctly” because the ROM hash is not
the one it expects — that warning is the point, not a problem. The full OZROM
treatment — memory map, ROM jump table, keytable, the column switch and its
mechanism, clock, memory tester — is docs/ozrom-1e.md (#81).
Why this matters to us. OZROM 1E is not a ROM swap with a nicer font; per its
manual (research/roms/OZROM-1E/OZROM_1E_Manual.pdf, Micro Management, Manual
Rev. A 84/09/01) it adds capability we would otherwise have to build:
- User-selectable 52 / 80 / 104-column display from the keyboard — a documented software video-mode switch. Directly relevant to the video personalities (#24) and to how the Nuevo and ScreenPac paths are driven. On a machine with no 80-column upgrade it can also toggle between the left and right 52 columns of an 80-column display.
- A system memory tester that reports which chip to replace — a validation tool for #29.
- A software-redefinable keyboard — individual keys at any time, plus eight locally-redefinable function keys of up to 63 characters each — relevant to #45/#47.
- Time-of-day clock support (
TIME,ALARM,TIMER, and a way to set the clock) — relevant to #7/#18. - Documented interfaces: an interrupt hook for programmers, the ROM jump table, the OZROM memory map and keytable, plus notes on 1793/1797 floppy controller operation (#20) and truer Televideo 912/920 emulation.
- Removes the “Centronics printer not connected” lockout on double-density systems (#19).
What it gives up, stated precisely in its own Appendix C (the manual is now OCR’d
to research/roms/OZROM-1E/OZROM_1E_Manual.ocr.txt):
IEEE-488 functions. The OZROM 1E does not support the IEEE-488 functions. Although Centronics printer operation is not affected, some printers … and some data acquisition and analysis devices … require the IEEE-488 routines to operate. In this instance, extra software would be required …
The jump-table appendix is blunt about it: the entries at 013F–0156 were the
IEEE-488 routines and calling them now produces a controlled system crash,
and “IEEE-488 AS IOBYTE DEVICE 3 NOW NULL DEVICE” — the status entries return
0FFH, input returns zero, output does nothing. The parallel/Centronics entries
at 0193+ survive.
So OZROM and Drive C are mutually exclusive. Drive C is an IEEE-488 device
and OZROM removes the routines it talks through; the manual’s “extra software
would be required” is exactly the driver we have the source for in
rom144.asm — so it is recoverable, but it is not free.
One more difference worth knowing, because it constrains any 2.0 ROM (#83):
OZROM shrinks the video window to 3 KB (F000–FBFF) and uses the last 1 KB
(FC00–FFFF) for its redefinable keyboard and function keys. It therefore
scrolls by moving the 23 lines above rather than by moving the window. The stock
ROM’s 4 KB video RAM map in docs/o1-memory-io-map.md §2 does not apply to
OZROM.
6. Headless testing
The technique is o1prsnt’s (utilities/headless-testing-harness/README.md for
the API, DISCOVERY.md for how it was derived):
SDL_VIDEODRIVER=dummy mame MACHINE -flop1 IMAGE -video none -sound none \
-nothrottle -autoboot_script RUN.lua -seconds_to_run N -rompath ROMS
SDL_VIDEODRIVER=dummyis mandatory from a pure terminal session on macOS; without it MAME aborts with “Could not initialize SDL”.-video none -sound none— no renderer, no audio. The emulated screen still renders into VRAM, which is all a test reads.-video noneis not “no window”;SDL_VIDEODRIVER=dummyis what makes a run headless, and it only works for an SDL build. The window belongs to the OSD, and MAME’s default is full screen (-window: “enable window mode; otherwise, full screen mode is assumed”). The mac OSD has noSDL_VIDEODRIVERto hide behind, so a run there takes the whole display with a grey window until it exits —-windowis the only thing that stops it, and-numscreens 0is rejected (must be between 1 and 4). The fix is not to add-windoweverywhere: it is to build with the SDL OSD, which is the tree’s default on macOS, and that needspkg-configso MAME can findsdl3(REQUIREMENTS.md). A source build without pkg-config cannot find<SDL3/SDL.h>and falls back to the mac OSD —apply.shsays so out loud when it does. Measured on theosborne1nxbuild, 2026-10-12.
- The screen is read straight from video RAM: base
0xF000, 128 bytes per row, 32 rows. Character codes are readable directly; the 9th (attribute) bit lives in a separate bank and can be ignored for text. -nothrottleruns the 4 MHz Z80 as fast as the host allows (~20–25×) butemu.wait()and-seconds_to_runcount emulated seconds, so any timing you measure is a true 4 MHz Z80 timing and transfers to hardware.- End the script with
os.exit(0); letting MAME hit the budget often ends in a harmless but noisy macOS teardown segfault.
The two harnesses in this family
| Harness | Where | Purpose |
|---|---|---|
tools/o1_mame_probe.lua + tools/check_o1_map.sh |
this repo | asserts docs/o1-memory-io-map.md (12 checks) against the running machine |
tools/research/mame-boot-probe.{lua,sh} |
this repo | boot one image on one machine: A> or not, screen dump, screen geometry, optional PNG |
o1harness.lua + run-test.sh |
o1prsnt |
keyboard injection (H.boot_cpm(), H.type_line()) and screen assertions (H.wait_for(), H.find(), H.dump()) |
The boot probe is the one that produced the geometry table in §2 and the
screenshots in docs/mame-screens/. It is deliberately dependency-free; for
scripted tests that need to type at the machine, use o1prsnt’s harness.
Screenshots. MAME’s Lua screen device has screen:snapshot(path), which
works headless (-video none) and needs no display:
manager.machine.screens[":screen"]:snapshot("/tmp/shot.png")
The §4 boot matrix was produced with the second one plus a ten-line script:
cat o1harness.lua probe.lua > /tmp/run.lua
SDL_VIDEODRIVER=dummy mame osborne1nv -flop1 IMAGE -video none -sound none \
-nothrottle -autoboot_script /tmp/run.lua -seconds_to_run 120 \
-rompath tools/emulator-setup/roms
Gotchas worth not rediscovering
- VRAM is a ring buffer. The machine scrolls by rotating a hardware start register, not by moving memory, so a given screen line moves between VRAM rows after any scroll. Assert with whole-buffer substring searches, never fixed row reads.
- Pace your typing.
natkeyboard:post()drops characters if a second post arrives while the first is still being typed, and the guest flushes its input buffer around disk I/O. Wait for the prompt, then type (~0.15 s/character). - MBASIC’s
Okprompts are stale-prone — count them (H.ok_count()) rather than matching"Ok". - MAME writes
cfg/andsnap/into the current directory — run from a neutral directory so neither the repo nor your home dir collects litter. - The BIOS waits for RETURN (“Insert disk in Drive A and press RETURN.”) before it loads anything; every automated test must press it.
7. Open items
Open:
- The prefix → driver inference in
tools/check_o1_map.shneeds revisiting (§4): the52-/80-/104-prefix does not determine what a disk can boot. - Identify the two differing bytes at offsets 5737 and 5798 in the
-blank.imdset — presumably a CBIOS video-mode selector, but unverified. A small disassembly job against the CBIOS in the system tracks. - ROM 1.3 is undumped and we hold its source disk — see
docs/o1-rom-variants.md. - Driving the OZROM column keys from the harness — the ctrl-shift
52/80/104-column select keys cannot be posted, because the harness types plain
characters. Routed to #167; wanted by #24 as much as by #81
(
docs/ozrom-1e.md§9).
Closed, and why:
- #79 — why the Nuevo CBIOS 1.5 hangs (
OS1NUEVO.IMD). Closed as not needed: the Nuevo path is not pursued (§4) and #24 is built from the ScreenPac design plus OZROM’s column switch. The hang is recorded rather than chased — and it is not a bad copy: Dunfield’s pristine upstream behaves identically. - OZROM 1E as a tool (#81) — closed. It boots on all three machines (§5) and
its manual is indexed in
docs/ozrom-1e.md: memory map, ROM jump table, keytable, the 52/80/104-column switch and its mechanism, the clock and the memory tester.
8. Sources
- Adaption base:
Myrcurial/o1prsnt— itsdocs/emulating-the-osborne-1.md,utilities/headless-testing-harness/{README,DISCOVERY}.mdandscripts/validate-emulator.sh. - Machine/ROM structure: MAME
src/mame/osborne/osborne1.cpp(ROM_START( osborne1 ),ROM_START( osborne1nv ), theCOMP()lines). - ROM inventory and CRCs:
docs/o1-rom-variants.md. - The memory and I/O map the probes assert:
docs/o1-memory-io-map.md(§2a for the video geometry, §3 for the memory-mapped I/O window). - OZROM 1E’s manual, ROM and interfaces:
docs/ozrom-1e.md(#81). - The fourth machine:
docs/osborne1nx-machine.md(#191). - Archive provenance:
research/README.md. - Disk pipeline and machine checks:
tools/README.md.