Skip to content

Releases

HavenDOS 0.9.7

HavenDOS 0.9.7 Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 20 Nov 13:53
0c926bc

Which file do I want?

File For
HavenDOS-0.9.7.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.7-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.7-source.zip Building it yourself -- make && make test

HavenDOS v0.9.7 Beta "Mnemosyne" — the old builds, the new numbers, and the mouse everywhere

The 0.9.7 line, released 20 November 2026. Kyan, after testing 0.9.6.7:
"just gimme the emulator lol, also I like it so that u release 0.9.7
with all planned stuff okay"
. So this is everything THS-RM-1005 REV C
planned for 0.9.7:

  • Retro PC (exception E-1): the studio's old builds, running inside
    HavenDOS.
  • The new version numbers in the updater, and update channels.
  • The editor: select text with the mouse, and one clipboard for every
    app (exception E-2).
  • Files: a right-click menu (exception E-3).
  • The taskbar: pinned apps, and smaller desktop icons (exception E-4).
  • The mouse wheel on real mice, a fix from his 0.9.6.7 test.

Mnemosyne is memory: the release that keeps the oldest builds alive, and
teaches every install the numbers it will be known by.

This is the first GitHub pre-release. From 0.9.7, every release
below Stable is marked pre-release (THS-RM-1004), so GitHub's "latest"
stays on 0.9.6.7 until HavenDOS 1 is Stable on 1 January. Get 0.9.7 from
the releases list.

Fixed: the mouse wheel on a real mouse

Kyan, on the HP: "now the scroll wheel on mouse still no".

0.9.6.3 read the wheel from a fourth byte that QEMU's mouse sends in
boot protocol. The boot protocol does not promise it, and most real mice
send three bytes there, keeping the wheel for their own report format.

  • A USB mouse is now asked for its report descriptor. The trackpad's
    parser (usb/hid_rd.c) reads it, and if X, Y and a button resolve,
    the mouse stays in report protocol, which carries the wheel.
  • Big moves are split across packets, so none of the movement is
    clipped.
  • Anything else falls back to boot protocol, exactly as before: no
    descriptor, a refusal, an absolute pointer (a tablet), or no X/Y.
  • dmesg says which: "USB mouse 046d:c077 interface 0: report
    protocol, wheel".

The new version numbers (THS-RM-1004, RM-1005 section 7)

HavenDOS 1 will call itself, the way Windows does:

HavenDOS 1, version 27H1 (OS Build 27100.1000)

0.9.7 teaches every install those numbers before they arrive.

  • include/version.h gains PRODUCT, RELEASE, BUILD, REVISION,
    CHANNEL_ID and LEVEL. PRODUCT 0 is the 0.9.x line, which still shows
    itself as it always has.
  • One version string (kernel/verinfo.c) for ver, the System app,
    Settings' About page, neofetch and the boot splash. On a channel other
    than Stable, its name goes inside the brackets: "(OS Build
    27100.1000, Release Preview)".
  • The update file (update.inf) now carries channel= and level=,
    and from HavenDOS 1 also product, release, codename, build and
    revision. Updaters from before 0.9.7 ignore the new keys, and
    version= is still what gets compared. The comparison is proved, not
    assumed: 0.9.9.10 < 27100.0999 < 27100.1000 < 27100.1001 < 27101.1000.
  • Channels. A system follows one: Beta, Public Beta, Release Preview
    or Stable.
    • update refuses an update from a less stable channel, by name,
      before writing anything, and says which command takes it anyway.
    • update channel shows which channel the system follows, and
      update channel stable changes it. The choice is a file beside the
      kernel (/boot/channel.inf), so promoting a build never changes the
      kernel's bytes.
    • With no file, a system follows its own channel: a Beta follows Beta,
      and HavenDOS 1 on Stable follows Stable. (Defaulting a 0.9.x Beta to
      Stable would refuse the next 0.9.x, since every one of them is a
      Beta.)
    • A manifest from before 0.9.7 names no channel, and is taken as it
      always was.
  • Release names come from one script, tools/release_names.py, used
    by the workflow: v0.9.7 and HavenDOS-0.9.7.iso now; 27100.1000
    and HavenDOS-1-27H1.iso from HavenDOS 1. Everything below Stable is
    a pre-release.

The editor: select with the mouse, one clipboard (exception E-2)

  • The editor reads the mouse on every pass. Before, it only looked
    after a key, so on the HP the pointer never showed there; that was one
    of the "cursor doesn't appear everywhere" places.
  • Click puts the cursor where you point, drag selects, and
    Shift+click extends the selection. The wheel scrolls three
    lines a notch.
  • Ctrl+C, Ctrl+X, Ctrl+V and Ctrl+A as before, but now through one
    clipboard every app shares
    (shell/clipboard.c). Copy in the editor
    and Ctrl+V at the Shell's prompt pastes it (its first line).
  • The right-click menu (Copy, Cut, Paste, Select All) now works with
    the mouse.

Files: a right-click menu (exception E-3)

Right-click a file or folder, or press F2, for Open, Rename, Copy,
Move, Delete and New folder. Right-click empty space for New folder.

  • Delete asks first. A folder must be empty.
  • Move renames into another folder where the file system can (FAT32).
    Where it can't (FAT16), it copies, checks the copy, and only then
    deletes the original.
  • Every action is judged by what happened on disk, not by a return
    code (the FAT drivers disagree on those), and every failure is named.
  • The menu works with the mouse (hover, click, click away) and with
    the keys.

The taskbar: pins, and smaller icons (exception E-4)

  • Pinned apps sit beside Start, on the desktop's bar and on the bar
    under every app. There are up to eight, and Files is pinned by
    default. Right-click an app and choose Pin to taskbar or Unpin
    from taskbar
    . They're kept in TASKBAR.CFG.
  • Small desktop icons: one row per app, the mark and the name side by
    side, about four times as many on the screen. Settings > Appearance >
    Desktop icons, or right-click the desktop.

New: Retro PC — the old builds, running inside HavenDOS

Kyan, 17 November, sending kernel.img: "this is the first ever build
i still have i think that survived everything maybe add is as an
emulator including in HavenDOS."

By his decision this is built and lands in 0.9.7, as exception E-1 in
THS-RM-1005.

Open Retro PC from the Start menu, or type retro in the Shell.

  • Built in: K-DOS Beta 1.1b, the oldest build that survived. Log in
    as admin, password 1234. Its command bar, help, time, calc,
    save and load behave as they did on a real PC, checked screen for
    screen against QEMU's own PC.
  • Your own images: put a .IMG floppy image (up to 2.88 MB) in
    RETRO and it appears in the list. retro FILE boots one straight
    from the Shell.
  • Each old PC gets its own hard disk: a 504 KB pretend disk, kept as
    a file beside it (RETRO/KDOS11B.HDD for K-DOS). It is saved when you
    leave. It is only ever that file: the old PC has no way to reach a
    real disk.
  • Leaving: F4 or the window's X takes you back to the list, and Q
    closes Retro PC. ESC goes to the old PC, because old programs used it.

What the old PC is

  • An 80186 with the 386's 16-bit additions. K-DOS was assembled for
    a 386 and uses them: the long conditional jumps, SETcc, MOVZX/MOVSX,
    the bit instructions, SHLD/SHRD, FS and GS.
  • A BIOS that does the screen (text modes), the keyboard, the disks,
    the clock and the timer, and boots from the floppy with INT 19h.
  • An 8259 interrupt controller, an 8253 timer and an 8042 keyboard
    port
    , so programs that hook the timer or the keyboard interrupt work.

When an old program asks for something it does not have

It says so on screen and in dmesg (lines start retro:), never
silently:

  • 32-bit instructions (the 66h and 67h prefixes), CPUID, and the
    protected-mode instructions stop the old PC. A box names the
    instruction, its bytes and its address. R restarts the PC.
  • A graphics mode, a call to DOS, or a BIOS service that is not
    provided
    shows a notice saying which one, and the program carries on
    if it can.
  • A disk with no boot signature is refused with that reason.

How it was checked

  • test_emu86, 86 checks: the 186 and 386 additions, what is refused,
    interrupt delivery while one is in service, the BIOS services, and
    K-DOS logging in and running its commands.
  • The CPU was held to the SingleStepTests 8088 suite, real-chip
    captures, and matches all 2,522,158 cases it can be held to. The few
    where an 8088 and a 186 differ are skipped and listed.
  • check_emu86_suite plants 28 defects in the emulator, one at a time,
    and the suite catches every one. It correctly misses the one harmless
    change.
  • harness/retro_boot.py boots the HavenDOS ISO in QEMU, opens Retro
    PC, logs into K-DOS and runs its commands: 9/9.

Found on the way, in K-DOS itself

These are K-DOS Beta 1.1b's own behaviour. The emulator reproduces them
on purpose, and QEMU's PC does the same:

  • save writes to the disk head in DH, which is only right by chance.
    After time, DH holds the seconds, so the write goes nowhere while
    K-DOS still prints [OK].
  • taskman garbles the screen, and the next key returns to the login.

Also in this change

  • A USB keyboard's F1–F4 now reach apps, as they already did on the
    PS/2 keyboard.
  • The window's X can be read by an app that wants it to mean
    something other than ESC (chrome_take_close()). Retro PC is the
    first to use it.

Tests

  • New suites:
    • test_verinfo (43): the display string, channels however they are
      typed, and the subscription rule.
    • test_clipboard (11).
    • test_emu86 (86, Retro PC).
  • Grown suites:
    • test_vercmp 39: HavenDOS 1's o...
Read more

HavenDOS 0.9.6.7

Choose a tag to compare

@github-actions github-actions released this 17 Nov 15:27
cb5f818

Which file do I want?

File For
HavenDOS-0.9.6.7.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.7-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.7-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.7 Beta "Daedalus" — the boot no longer waits on the trackpad

The seventh maintenance update of 0.9.6, from Kyan's photo of 0.9.6.6 on
the HP, 17 November. Fixes only.

What the photo shows

The boot splash stopped with:

I2C 00:1F.5 at FE010000: not a DesignWare core (type 00000000)
press any key for the boot log

No key did anything.

That line is the last of the I2C step, so the boot had got past
assigning the trackpad's controllers their memory window, which 0.9.6.6
added. Since 0.8.8.4 they had never come up, so the search for the
trackpad had never run on the HP.
0.9.6.6 made it run for the first
time, and it held the boot.

  • Each I2C wait was 2,000,000 register reads. In QEMU that is
    nothing. On the HP it is one to two seconds.
  • The search tries 112 addresses on each controller. If the
    controller never moves the bus (for example, if its clock is not
    running), every address waits its full time. That is many minutes.
  • It ran before USB. On the HP the keyboard is USB, so "press any
    key" could not be answered until the search was over.

Fixed

  • The keyboard comes first. I2C and the trackpad search now run after
    USB and the USB keyboard and mouse.

  • A wait has a clock. An I2C wait now ends after 50 ms of the timer,
    a hundred times the longest transfer it makes, as well as after the
    spin count.

  • A controller that times out is left alone. A NAK is an answer
    ("nobody here") and the search goes on. A timeout means the controller
    is not moving the bus at all, so the first one marks it stuck and every
    later transfer is refused at once. The log says so once:

    I2C 00:15.0: a transfer timed out -- the controller is not moving the
    bus, so it is left alone this boot
    

    i2c lists it as "timed out at boot -- left alone".

  • noi2c leaves the whole step out: no controllers, no search. Press
    e on a boot entry in the boot menu and add it to the multiboot line.

  • The boot log says when the step starts: "I2C: looking for the
    trackpad's controllers", then "I2C: searching 2 controllers for the
    trackpad".

Getting 0.9.6.7 onto an installed 0.9.6.6 that will not boot

  1. In Windows: run Update-HavenDOS.cmd from this bundle as usual.
    It puts the new kernel on the EFI partition.
  2. Boot this release's ISO from a USB stick and choose "Update
    installed system from the ESP". That runs 0.9.6.7 itself, with the
    installed partition mounted.
  3. Run update.

Then boot normally. If the trackpad's controllers time out, the log now
says so and the boot carries on.

What to photograph

dmesg | grep I2C after it boots. The lines say:

  • where each controller's window went;
  • whether each one is a DesignWare core;
  • whether its first transfer worked, timed out, or found the trackpad.

Tests

  • test_i2c 62, up from 55. A simulated controller that never moves
    the bus:
    • a probe on it times out within a bounded number of register reads
      while the timer runs, and within the spin count when it doesn't;
    • the bus is then stuck and not ready, and the next transfer is refused
      with no register read at all;
    • a whole scan stops at the first timeout;
    • a NAK leaves the bus alone.
  • check_i2c_suite: 18 planted defects, all caught, four of them new.
    A fifth, a stop guard in the scan, was MISSED. That was correct, since
    a stuck bus already refuses every probe without reading anything, so
    the guard was removed rather than kept untested.
  • 7479/7479 checks across 72 suites.
  • QEMU: fixes_0966 9/9 and hid_boot 3/3. QEMU has no DesignWare
    controller, so the I2C path itself only runs on the HP.

HavenDOS 0.9.6.6

Choose a tag to compare

@github-actions github-actions released this 17 Nov 14:18
f28f9ad

Which file do I want?

File For
HavenDOS-0.9.6.6.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.6-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.6-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.6 Beta "Daedalus" — what the HP showed in 0.9.6.5

The sixth maintenance update of 0.9.6, from Kyan's test of 0.9.6.5 on
the HP on 17 November. Its message:

  • the window system "is cool, it starts to look like a real OS now";
  • Alt+Tab and the mouse work;
  • the monitor works, but on the unscaled fallback;
  • the Shell's buttons don't work;
  • full screen should fill the monitor;
  • the trackpad still doesn't work ("still the memory window error");
  • a mouse, iPhone or stick unplugged and plugged back in doesn't come back;
  • the resolution is only applied once the Shell has been opened;
  • long lists like the Start menu's All Apps don't scroll.

These are fixes. Full screen filling the screen was his choice of what
full screen should mean, asked on the same day. The features he asked
for alongside them come before HavenDOS 1 as written exceptions E-2 to E-5
in THS-RM-1005 REV C:

  • mouse selection and copy/paste in the editor;
  • a right-click menu in Files;
  • pins on the taskbar and smaller icons;
  • real audio through the speakers.

Fixed

The Shell's X did nothing

The window's X sends the app an ESC, which closes almost every app. The
Shell's ESC clears the line it is typing, so its X did nothing at all.
The title bar also said "ESC closes" above it.

  • The Shell's X now runs exit. The window remembers that the ESC
    came from the X (chrome_take_close()), and the Shell asks.
  • The title bar says "exit closes" in the Shell.
  • A real ESC on the keyboard can never be taken for a click: the
    remembered X is forgotten as soon as any key arrives.

Full screen fills the screen

An 80×25 app (most apps, every game) used to stay an 80×25 box in the
middle of the screen when made full screen, at the size it already had.
Kyan: "full screen should render the app on full monitor".

  • Now the whole console becomes 80×25 with the biggest characters
    that fit,
    keeping their shape. On the HP's 1024×768 they are 12×24
    instead of 8×16, so the app uses 960×600 of the screen instead of
    640×400. What the app had drawn moves with it.
  • F11 or Alt+Enter brings the window back at the grid you chose,
    with its title bar and the taskbar.
  • It goes back by itself when the app closes, and before Alt+Tab or
    a taskbar click shows another app. Every other app is laid out for
    your grid.
  • Apps that use any size (Settings, the Shell) are unchanged: their
    full screen was already the whole screen.

A mouse plugged back in never moved again

When something is plugged in or pulled out, HavenDOS walks the whole USB
bus again. That rebuilds its table of devices, but keyboards and
mice were never bound again afterwards.
Their driver kept pointers into
the old table. So:

  • a mouse unplugged and plugged back in was found ("device arrived") but
    never moved;
  • plugging in a stick or the iPhone could leave the keyboard or mouse
    driver polling whatever device had taken its place.

The rescan now binds keyboards and mice again, right after the bus walk
and before the sticks and the network adapter. QEMU shows it: a mouse
pulled out and plugged back in gets a second "USB mouse" line, and a
stick plugged in after that comes back as a disk with the mouse still
bound. Without the fix the same test shows one "USB mouse" line and the
mouse is dead.

The trackpad: its controller had no memory window

On every boot dmesg said "I2C 00:15.0: BAR0 is unassigned -- the
firmware gave it no memory window". The HP's firmware leaves the
trackpad's two I2C controllers with no address to talk to them through.
Windows and Linux assign one themselves; HavenDOS never had.

drivers/pci_assign.c assigns one. Where it goes, and why it is
safe:

  • Above 4 GB, right after a device the firmware placed there itself.
    On the HP every chipset device sits at 0x60_xxxxxxxx (the USB
    controller at 0x60_02120000). An address right after one of them is
    inside the range the firmware already makes the chipset answer.
  • Never below 4 GB, where the interrupt controllers, the timer and
    the flash chip live without being listed anywhere this code can see.
  • No other device's registers are touched. Measuring another
    device's window would mean writing to it while it runs (the screen,
    the keyboard's USB controller). It isn't needed: a window is always
    aligned to its own size, so where it starts is enough to bound how
    big it can be.
  • Written with the controller's memory decoding off, then read back.
    If the address doesn't stick, the controller is put back to zero and
    left off, and the reason is logged.

dmesg now says either "I2C 00:15.0: BAR0 assigned at ..." followed by
what the controller is, or exactly why no window could be given. This
is the step the trackpad has been waiting on since 0.8.8.4. Whether the
trackpad then answers is what the HP will show.

Long lists didn't scroll with the wheel

The desktop and Files read the mouse themselves, so 0.9.6.3's wheel
never reached them.

  • The Start menu: the wheel, or two fingers on a trackpad, moves
    through the open category three apps a notch. All Apps is longer
    than the screen.
  • Files: the wheel moves through the list.

The resolution was only applied once the Shell opened

A resolution you saved was applied only at the Shell's prompt. That rule
dates from before apps were windows, when a big grid would have
stranded 80×25 apps in a corner. Since 0.9.6.5 they are boxed and
centred, so the saved grid is now applied at boot, before the
login screen. It applies on the framebuffer console only, and only for a
grid that holds 80×25 (a smaller one is still replaced, as since
0.8.8.5). dmesg says "Console: 128x48, as saved". This needs an
installed system to see: a live ISO keeps no settings.

Not fixed yet, and what would help

  • The monitor on the unscaled fallback. 0.9.6.4 made the fallback
    say why it was taken. A photo of dmesg | grep monitor shows the
    step where the scaled picture stopped.
  • "The cursor doesn't appear everywhere." Which app or screen? One
    cause is known: the text editor only looks at the mouse after a key.
    That is rebuilt with E-2 (mouse selection) in 0.9.7.
  • The iPhone after a replug. The network adapter now comes back, and
    DHCP starts again. Whether the phone shares its connection again
    without typing pair is for the HP to say.

Tests

  • test_pci_assign, new, 44 checks. It runs on a simulated bus laid
    out like the HP's, from its own lspci, with the true size of every
    device's window. The planner never sees those sizes; the checks
    compare against them.
    • Both I2C controllers get windows that touch nobody, above 4 GB,
      inside the chipset's cluster.
    • Every write is made with decoding off, and no other device's
      register is written.
    • Each refusal is named.
  • check_pci_assign_suite, new: 12 planted defects, all caught. Its
    first run caught 10; the two that got through led to two new cases in
    the suite.
  • test_kpoll 159: the rescan order now includes binding keyboards
    and mice. check_hotplug_suite has 15 defects, all caught, two of
    them new.
  • test_console 337: 80×25 filling the HP's 1024×768 at 12×24,
    pixel for pixel, and back.
  • test_window 34: full screen grows and always comes back (F11,
    another app in front, the app closing), with the app's cells moved
    both ways.
  • test_i2c 55: an unassigned window is asked for once; a refusal
    means no bus.
  • 7472/7472 checks across 72 suites.
  • QEMU:
    • fixes_0966, new, 9/9: the Shell's X, full screen filling, a mouse
      and a stick unplugged and plugged back in. Against a build without
      the rescan fix, the mouse checks fail.
    • window_apps 11/11 and mouse_apps 14/14.

HavenDOS 0.9.6.5

Choose a tag to compare

@github-actions github-actions released this 16 Nov 16:39
5e1cca7

Which file do I want?

File For
HavenDOS-0.9.6.5.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.5-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.5-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.5 Beta "Daedalus" — apps are windows, the taskbar stays

Kyan, 16 November, after 0.9.6.4 got the monitor working: "settings no
mouse and I would like that the taskbar is visible at all times, with
the X, full screen button and minimize button at the window's top bar."

By his decision this ships now, in 0.9.6.5, bending the 0.9.x "fixes
only" rule again. Windows proper are still 29H1's.

New: every app is a window

Every app opened from the desktop, the Start menu or open now has:

  • a title bar on top, with the app's name and three buttons:
    • _ minimize: the desktop comes forward. The app stays open,
      with its chip on the taskbar.
    • [ ] full screen: the title bar and taskbar go and the app has
      the whole screen. F11 or Alt+Enter brings them back; Alt+Enter
      is there for laptops whose F11 is a media key. An 80x25 app keeps
      its title bar, centred on the whole screen.
    • X close: does what ESC does, so an app closes, or asks to save
      first, the way it always has. On a sub-screen, one click goes back a
      level.
  • the taskbar on the bottom row, always:
    • Start opens the desktop with its start menu;
    • a chip for every open app brings that app forward;
    • the clock.

Apps that fit any size (Settings, the Shell) get the whole screen
between the title bar and the taskbar: 128×46 on the HP. 80×25 apps
keep their box, now placed between the two.

On a screen with no room for any of this it is skipped, so apps run as
before. That means 80×25, which is QEMU's default, or anything under 27
rows. It is also skipped without tasks (safe mode, notasks).

How it works:

  • The pointer now lives on the whole screen, not just inside the app's
    box, so it can reach the title bar and the taskbar.
  • An app still gets coordinates inside its own box, held at the edge,
    so no app sees a column it doesn't have.
  • A click outside the box belongs to the window, not the app.

Fixed: the mouse in Settings

Settings opened from the desktop's right-click menu, or from the
Accounts shortcut, ran inside the desktop's own task. The desktop reads
the mouse itself, so the mouse layer is switched off there. Settings now
turns the mouse on for itself however it is opened, and so does every
app the desktop runs.

Tests

  • New test_window, 25 checks: the layout on 128×48 and on grids
    with no room, and what every click hits. check_window_suite 9/9.
  • test_kbdinit 49: the pointer outside an app's box, held at its
    edge for the app, with the click kept for the window.
  • test_uimouse 53.
  • QEMU, new harness/window_apps.py, 11/11 three runs in a row,
    with a USB mouse at 128×50:
    • the title bar and taskbar;
    • the wheel in Settings;
    • full screen and Alt+Enter;
    • minimize, with the app still open;
    • opening it again brings back the same window;
    • X closes it.
  • mouse_apps 14/14, appguard_boot 19/19, resolution_tasks 8/8,
    hid_boot 3/3, gfx_boot 15/15.
  • desktop_layout: the boxed app's title bar still passes. Its two
    drawer checks fail on 0.9.6.2 too.
  • 7401/7401 across 71 suites.

On the HP

  1. Open Settings from the Start menu, and also from the desktop's
    right-click menu. Click rows, scroll with two fingers, and try X.
  2. Open a few apps. Try minimize, then their chips on the taskbar, then
    Start.
  3. Try full screen, then Alt+Enter to get back. Also try F11, which
    may be a media key on the HP.

HavenDOS 0.9.6.4

Choose a tag to compare

@github-actions github-actions released this 16 Nov 15:20
297f5d0

Which file do I want?

File For
HavenDOS-0.9.6.4.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.4-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.4-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.4 Beta "Daedalus" — the monitor never gets stuck

The fourth maintenance update of 0.9.6, from Kyan's photo of
16 November. Fixes only.

Fixed

gfx on was refused as "already lit" while the monitor was dark

What the photo shows:

  • the warning "Monitor not lit -- gfx on says why";
  • then gfx on refused with "pipe B is already lit -- gfx off first",
    before and after a gfx off.

Yet all three registers the refusal printed read off:

  • TRANSCONF B 00000000;
  • DDI_FUNC B 00030000 (only the sync polarities);
  • DDI_BUF B 00000080 (idle).

What happened:

  1. 0.9.6.3's watcher tried to light the monitor scaled, and the attempt
    stopped part-way, after the PLL, the clock and the scaler but before
    the pipe started.
  2. That left pipe B half set up. The refusal counted the scaler as
    "lit" but never printed it.
  3. Each later attempt was refused, and the watcher gave up after three.
  4. gfx off only cleared parts that were enabled, so the leftovers
    stayed and gfx on stayed refused.

In 0.9.6.4:

  • gfx on, the watcher and Settings clear a half-set-up pipe B by
    themselves first
    , saying so ("Pipe B was left half set up by an
    attempt that stopped: clearing it first").
  • If a scaled light stops, the monitor is lit unscaled instead, the
    way 0.9.6.2 lit it on the HP. gfx on prints the step that stopped
    scaling, with the register, the value written and the value read back.
    The watcher puts the same line in dmesg and shows a notification
    saying it is unscaled. The monitor shows a picture either way.
  • The refusal names its reason. It also prints PLANE_CTL B and both
    of pipe B's scalers, lists what is on, and says when pipe B isn't
    actually running.
  • gfx off clears everything of pipe B: any scaler that is set, not
    just one that is enabled, with its window; and DDI_FUNC's leftover
    bits.
  • The scaler's read-back checks only what matters: the enable bit,
    and the window's position and size. i915 reads none of these back,
    and the phases' spare bits are the hardware's to report however it
    likes. That may be what stopped it. If it was something else, the
    step is now on the screen.
  • gfx b also shows pipe B's scaler, its window size, and pipe A's
    scaler for comparison.

How to test

  1. Update and reboot with the LG plugged in.
  2. Expected: within a few seconds the monitor lights.
    • Scaled: the notification says "Monitor on: fitted, shape kept".
      Photograph the monitor.
    • Unscaled: the notification says so. Run dmesg and photograph
      the "monitor: scaling stopped at ..." line. That line says exactly
      which register the scaler stopped at.
  3. Then gfx off, then gfx on. Photograph the end of its output.

Tests

  • test_xemode 285 (floor 285). It replays the HP's state: a scaled
    light stopped at the scaler, refused as busy, the refusal naming
    "scaler 1", gfx off clearing the scaler and DDI_FUNC 00030000, then
    lighting again.
  • check_xemode_suite 47/47, with four new injections.
  • 7367/7367 checks across 70 suites.
  • QEMU: mouse_apps 14/14 and gfx_boot 15/15. QEMU has no Iris Xe,
    so the fallback itself only runs on the HP.

HavenDOS 0.9.6.3

Choose a tag to compare

@github-actions github-actions released this 16 Nov 12:14
8a52204

Which file do I want?

File For
HavenDOS-0.9.6.3.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.3-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.3-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.3 Beta "Daedalus" — the monitor fills, and the mouse works everywhere

The third update of 0.9.6. Kyan asked for this on 16 November, after
0.9.6.2 got the HDMI monitor working. It bends the 0.9.x "fixes only"
rule once, by his decision: it adds mouse support in every app, and the
monitor picture now scales to fill the screen.

The monitor

The picture fills the monitor

The firmware gives HavenDOS a 1024×768 screen. The laptop panel
stretches it with a hardware scaler. In 0.9.6.2 the monitor got the
same 1024×768 unscaled, as a box in the middle of 1920×1080.

Now pipe B uses its own scaler, the same way Linux does
(skl_pfit_enable):

  • The mode is the monitor's own best (1920×1080 on the LG), not
    the smallest mode that fits the box.

  • The picture is fitted to it, in one of three ways:

    setting on the LG where to set it
    fitted (the default) 1440×1080, the 4:3 shape kept, thin black bars left and right Settings
    stretched 1920×1080 edge to edge, like the laptop panel Settings
    centred 0.9.6.2's unscaled box gfx fit centre only
  • Settings > Display > HDMI monitor > Picture switches between
    fitted and stretched. If the monitor is lit, the change shows straight
    away. gfx fit keep|stretch|centre does the same from the shell.
    The choice is saved.

  • Pipe A's scaler is never written. Only pipe B's scaler registers
    (0x68900–0x68AFF) are on the allow-list. The fatal-error mask is set
    for the scaler's start, as Linux does on Alder Lake-P, then restored
    to exactly what it was.

The monitor lights by itself

  • Plug in an HDMI monitor, at boot or later, and it lights on its
    own, with no gfx on and no question.
  • Unplug it and pipe B switches off.
  • How it watches: a background job checks the HDMI hot-plug line
    once a second, and only believes a change once two reads in a row
    agree.
  • If the monitor won't light, it tries three times, a second apart,
    then stops until the cable is unplugged and plugged back in. You get a
    notification and a dmesg line saying why.
  • Settings > Display > HDMI monitor > Light it by itself turns this
    off, as does gfx auto off.

The mouse in every app

Until now only the desktop, Mines, Chat, the file manager, Notes and
the login screen used the mouse. Settings and every other app were
keyboard-only.

The new mouse layer (shell/uimouse.c) gives every app the mouse
without rewriting it, by turning what the mouse does into the keys the
app already reads:

you do the app gets
click a row in a list the arrow presses that select it; a click on the row that is already selected is Enter, so a double-click opens it
click a key hint on the bottom two rows ("Enter apply", "ESC close", "Q quit", "[Space] pause") that key
scroll wheel, or two fingers on the trackpad Up / Down, one per notch
right-click ESC: back, or close
  • Settings is fully clickable:
    • the sections in the sidebar;
    • every row on every page;
    • the Y / N in a question;
    • the key hints at the bottom.
  • Apps built with the UI kit get clickable rows.
  • Every other app gets the key hints, the wheel and right-click,
    because those need nothing from the app. The hints are read off the
    screen at the moment you click.
  • The pointer shows in every app as an inverted cell, so it is
    visible on any colour, including a selection bar.
  • Where it holds back:
    • games (Snake, Tetris, Pong, 2048, Breakout) keep right-click
      and the hints but not the wheel, so a brushed trackpad doesn't
      steer a snake;
    • the shell prompt keeps clicks on hints, but the wheel and
      right-click do nothing there;
    • apps that already handle the mouse themselves (the desktop,
      Notes, Files, Mines, Chat, the login screen) are left exactly as
      they were.
  • The scroll wheel is new in the drivers:
    • a USB mouse's fourth byte, which most mice send even in the boot
      protocol;
    • the trackpad's wheel field, if it has one;
    • otherwise a two-finger drag on the trackpad, with one notch per
      1/24 of the pad's height. While scrolling, the pointer doesn't move.

Tests

  • New suite test_uimouse, 51 checks:
    • hints, rows, buttons and prefixes;
    • the hook end to end;
    • the pointer never leaving a cell behind;
    • per-task tables.
    • check_uimouse_suite: 14/14 injections caught.
  • test_xemode 265 → 274:
    • fit windows and phases (i915's formula);
    • the scaled plan on the model engine, with pipe A untouched;
    • gfx off clearing the scaler;
    • the allow-list;
    • the watcher's decisions.
    • check_xemode_suite 43/43, with 10 new injections.
  • Wheel input:
    • test_i2c_hid 90: two-finger scroll;
    • test_usbhid 39: the wheel byte;
    • check_usbhid_suite 8/8.
  • QEMU, harness/mouse_apps.py 14/14, with a USB mouse on xHCI:
    • click a section;
    • click a row twice to toggle it;
    • the wheel, checked on the screenshot's highlight;
    • right-click back;
    • click "ESC close";
    • gfx fit, gfx auto, and the watcher in kpoll.
  • harness/qemu.py's mouse_to is fixed. It still assumed 4 counts
    per row, where the driver has used 8 on QEMU's screen since 0.8.9.6.
    desktop_layout's bell click passes again. Two layout checks there
    fail on 0.9.6.2 too, and are unrelated.
  • 7356/7356 checks across 70 suites.

Not yet proven on the HP

  • the scaler on the LG: does the picture fill it, and is it sharp
    enough?
  • the monitor lighting by itself, on plug-in and at boot;
  • two-finger scrolling on the HP's trackpad (the suite uses a model
    pad);
  • the wheel on your USB mouse.

QEMU has none of these.

HavenDOS 0.9.6.2

Choose a tag to compare

@github-actions github-actions released this 16 Nov 10:49
f8db7a3

Which file do I want?

File For
HavenDOS-0.9.6.2.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.2-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.2-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.2 Beta "Daedalus" — the monitor's buffer

The second maintenance update of 0.9.6, made from Kyan's gfx try photos
of 16 November. It contains fixes only (THS-RM-1004 section 12).

Fixed

The HDMI monitor showed one colour edge to edge, never the picture

gfx try on the HP settled it. Each of the four trials painted the
monitor in its own colour (green, red, white, blue-green) and none showed
a picture. The underrun rate hardly moved:

before deep watermark 64×64 plane both DBOX credits
underrun / 20000 1504 1346 1342 1485 1497

A plane one sixteenth the size starved just as badly, so the cause was
never bandwidth. The plane was reading a buffer that does not exist.

Alder Lake-P has four display-buffer slices of 1024 blocks:

  • pipe A (the panel) uses slices 1–2;
  • pipe B (the HDMI socket) uses slices 3–4.

On this display generation each pipe's PLANE_BUF_CFG is counted from
the start of its own half. Linux subtracts mbus_ddb_offset(), which
is 2048 for slices 3–4. 0.9.6.1 wrote pipe B's buffer as blocks
2048–4031, counted from slice 1. Pipe B read that as 2048 blocks past
the start of its half, which is past the end of the 2048 blocks it has.
Every fetch came back empty, whatever the plane's size or watermark.

gfx on now writes 0–1983 (PLANE_BUF_CFG 07BF0000): the same
1984 blocks, counted the way the hardware counts them. 64 blocks are
still left free for a cursor.

Changed

  • gfx try has a fifth trial, YELLOW, which is the fix itself. It
    moves the buffer to 0–1983 on the running pipe. Lighting the monitor
    again isn't needed. Trial 0 still puts everything back afterwards,
    the buffer included.
  • gfx b names the bug when it sees it. If pipe B's buffer reaches
    past block 2047, it prints four red lines saying why the plane
    underruns and what fixes it.
  • gfx on measures the underrun rate over 20,000 reads after it
    lights the monitor, instead of reading the register once. It says
    either "the plane keeps up" or "the plane is starving".

How to test

  1. Update and reboot. Plug the monitor into the HDMI socket.
  2. Run gfx on.
    • Expected: the HavenDOS screen in the middle of the monitor, and
      underrun in 0 of 20000 reads.
    • Photograph the monitor and the last lines on the laptop.
  3. If the monitor is still a flat colour, run gfx b and gfx try, and
    photograph them as before. Trial 5 is yellow.

A monitor that was lit by 0.9.6.1 in the same boot keeps the old buffer
until gfx off. gfx try trial 5 fixes it without that.

Tests

  • test_xemode 236 (floor 226 → 236): the buffer value, trial 5,
    xm_buf_outside(), and gfx b's new lines;
  • check_xemode_suite 33/33: three new injections, one of them
    0.9.6.1's absolute numbers put back;
  • 7258/7258 checks across 69 suites.

HavenDOS 0.9.6.1

Choose a tag to compare

@github-actions github-actions released this 16 Nov 10:18
0306d68

Which file do I want?

File For
HavenDOS-0.9.6.1.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.1-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.1-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.1 Beta "Daedalus" — the first fixes from the HP

The first maintenance update of 0.9.6, made from Kyan's photos of 16
November. It has fixes only, as the 0.9.x rules in THS-RM-1004 section 12
require. The one new command, gfx try, is a diagnostic for the external
monitor.

Fixed

resolution refused on the desktop, every time

0.9.6 refused to change the grid whenever any other app was open, so that
open apps would not come back scrambled. On the desktop that meant always,
because the desktop and the Shell it opens are already two apps. The shell
also never got back to the resolution you chose for it.

Now the grid can change whatever is open:

  • Other apps are re-laid out when you switch back to them. Every cell
    keeps its row and column at the new width. What no longer fits is
    dropped, and the new edge is blank.
  • An app made for 80×25 is centred again in the new grid, and what it
    drew moves with it.
  • The cursor stays inside the screen.
  • Nothing is asked of the app: it sees its own screen in the same place,
    and lays itself out for the new grid the next time it draws.

The re-layout is ctx_reflow() in drivers/vga.c, run from
vga_ctx_load() whenever a console made for one grid is loaded at another.

The "Welcome / Create new account" screen came back on every boot

It was meant only for a machine nobody had made an account on. It checked a
flag held in memory, and memory is empty at every power-on, so the check was
always true, even after setup had made your account. It now shows only when
no account exists besides the three built in (admin, bootos and guest).

BOOT Chat is out of the menus

Its server is shut down until further notice. chat on its own now says
so, and names what replaces it in 27H2: IRC on Libera.Chat, Haven News, and
LAN chat. chat user@host still connects to a server you run yourself.

gfx on said "no underrun" when there was one

It read the interrupt register that latches events, but an underrun that is
masked there never latches. It now says one read proves nothing, and points
to gfx b.

For the external monitor: gfx b and gfx try

On the HP (16 November), gfx color blue turned the LG blue edge to
edge.
So the HDMI link, the timing and pipe B all work; what is missing is
the picture layer (the plane). gfx b showed:

  • the plane is on (PLANE_CTL 94000008) and armed (SURFLIVE 0, the same as
    PLANE_SURF);
  • PIPE_ISR bit 31 set: pipe B's FIFO underran. The plane is trying to
    fetch the screen and can't keep up, so nothing it fetches is shown.

What 0.9.6.1 adds to find out why:

  • gfx b now samples bit 31 20,000 times and prints how many reads saw
    it set. That's a rate rather than one lucky read.

  • gfx try runs four trials. Each changes one thing on pipe B's plane,
    measures the underrun rate, and paints the background its own colour, so
    a photo of the monitor shows which trial it was:

    Trial Background Change
    1 green deep watermark: WM0 gets the whole buffer, every other level off
    2 red a 64×64 plane, a sixteenth of the bandwidth
    3 white both together
    4 blue-green the MBUS DBOX B credits 12 → 8, as i915 sets them on Alder Lake-P

    Everything is put back at the end, or straight away at Q. Only pipe B's
    own registers are written, through the same allow-list as gfx on.

Tests

  • make test: 7248/7248 across 69 suites.
    • test_console 309 → 326: re-layout at 80×25 → 90×30 → 80×25, a
      centred viewport, and a console larger than the new grid.
    • test_xemode 209 → 226: every trial against the model of the HP's
      engine. PLANE_SURF is written last; trial 0 puts back every register;
      nothing outside pipe B is written; the underrun rate counts bit 31 only.
  • check_console_suite.py: 38/38 caught, including three new
    injections for the re-layout.
  • New harness/resolution_tasks.py: 8/8. In Shell Mode it opens Notes
    and writes a line, then sets resolution 90x30 with Notes open. Notes
    comes back with its line moved as one piece, and is still intact at
    80×25. The 0.9.6 ISO fails the same harness, 5/7.
  • appguard_boot.py 19/19, which includes the login screen.

Not yet proved on the HP

  • gfx try: that's what it's for.
  • The resolution change on the HP's 128×48 framebuffer grid.
  • Alt+Tab with the USB keyboard, the Welcome fix, and migrate, all still
    waiting for photos.

HavenDOS 0.9.6

Choose a tag to compare

@github-actions github-actions released this 15 Nov 21:30
f073926

Which file do I want?

File For
HavenDOS-0.9.6.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6 Beta "Daedalus": several apps at once

15 November 2026.

Until this release HavenDOS ran one program at a time. The desktop
called an app, and the app had the machine until it returned. Kyan
chose both halves of the old 0.8.x promise:

  • tasks that switch when an app waits for a key or yields, with
    background jobs;
  • switching between open apps with Alt+Tab, each keeping its state,
    with background tasks still running.

Both are in this release. So is migrate, which moves a FAT16 install
to a new FAT32 partition on the same disk, from inside HavenDOS (below).

What you see

  • Alt+Tab. Hold Alt and press Tab: a panel lists what is open,
    most recently used first. More Tabs move down, Shift+Tab moves up,
    and letting go of Alt switches. A quick Alt+Tab flips between the
    last two. Enter switches too, Esc cancels, and X closes the
    highlighted app.
  • Open apps on the bottom bar. One chip per open app, between your
    name and the alerts: grey when it is paused, green when it is running
    behind. Click one to bring it forward.
  • Apps keep their state. An app you switch away from stops where
    it was and comes back exactly as you left it, screen and all. Games
    pause. Editors keep what you typed.
  • Two apps keep running behind.
    • Stopwatch: a timer still rings while you do something else.
    • Music Player: the song plays on while you switch away. (The
      player had never played a note before this; see below.)
  • One of each. Opening an app that is already open brings it
    forward instead of starting a second copy, so two Notes can't save
    over each other. The Shell is the exception: open as many as you like.
  • Jobs. In the shell, ping 8.8.8.8 & runs the command behind, in
    its own task with its own screen. Alt+Tab or fg <id> shows it. When
    it ends, the alerts say so; any key then closes it.
  • ps lists the desktop, each open app and each job: id, name,
    state, how long it has been open, how much of its stack it has used,
    and how often it has run. jobs is the same command.
  • fg <id or name> brings one forward. kill <id or name>
    closes one.
  • System > Overview > Tasks counts what is open and what is running
    behind, instead of "one app at a time".
  • The user guide has a page on it ("Several apps at once") and the
    shell page has jobs.

How it works

The design is in include/task.h. The short version:

  • Every app is a task: its own 96 KB stack, its own console and its
    own crash guard. Task 0 is the context the kernel booted on (the
    desktop, or the shell in Shell Mode).
  • One task is in front. It owns the screen, the keyboard and the
    mouse buttons. Each task behind is either paused (the default: it
    stops at its next keyboard read) or running (it carries on and
    draws into its own console, which nobody sees until it comes forward).
  • Cooperative, not preemptive. Every layer in this tree keeps
    module-level buffers and none of them is reentrant, so switching from
    the timer interrupt would corrupt transfers in flight. A switch
    happens only where kernel_poll() is already allowed to run --
    keyboard_getchar(), kpoll_sleep_ms() and the desktop's idle wait
    -- and is refused for the same reasons: a driver holds a shared buffer
    (kpoll_depth()), the bottom half is running, or the switch is
    locked. A switch is therefore exactly as safe as the polling that
    already happened at the same instruction.
  • sleep_ms() does not switch. It is the driver-safe wait. An app
    that keeps running behind waits with kpoll_sleep_ms() instead, or it
    would hold the app in front up for every wait.

New files:

  • include/task.h, kernel/task.c: the table, the switch (eight
    instructions of assembly), who is in front, and the refusals.
  • shell/taskswitch.c: the Alt+Tab panel, ps, fg, kill and jobs.

Changed:

  • drivers/vga.c: a console per task. The shadow buffer is a
    pointer now; vga_ctx_t holds the buffer, cursor, colour, viewport,
    frame depth and clip count. Only the console in front paints. Bringing
    one forward repaints from its own buffer, so an app reappears without
    drawing anything. A console that is not on the screen cannot change
    the mode (VGA_MODE_EHIDDEN).
  • kernel/appguard.c: a crash guard per task, swapped on every
    switch, so a fault lands in the frame of the task that faulted. An app
    opened from the shell (open notes) used to run unguarded, because
    the shell's guard was already armed; it has its own guard now.
  • drivers/keyboard.c, usb/hid.c: Alt is tracked on both
    keyboards (PS/2 left and right, the USB modifier byte), and Alt+Tab
    produces KEY_TASKSW, which never reaches an app. The keyboard asks
    the task layer before every read, through a hook.
  • drivers/mouse.c: buttons and clicks belong to the task in front. A
    task behind sees where the pointer is but never a press, and never
    consumes one.
  • kernel/kpoll.c: kpoll_sleep_ms() gives other tasks a turn between
    slices; kpoll_in_poll() is new.
  • shell/desktop.c: desktop_run_app() starts a task and waits to be
    in front again. The old body is app_run_inline(), still used when no
    task can be started (no slot, no memory, safe mode, notasks). The
    registry gained a tasking field: TK_PAUSE, TK_RUN, TK_SOLO.
  • kernel/main.c: tasks start after the heap and the console.
    notasks on the boot line turns them off, and safe mode never
    starts them; either way apps run one at a time, as in 0.9.5.

Security: the lock and login screens

Writing this turned up three holes that task switching would have
opened. All three are closed:

  • Alt+Tab on the lock screen (lock, or Lock in the start menu)
    would have switched straight out of it to whatever was open behind.
    The front is now pinned while a lock or login screen is up:
    nothing can be brought forward, and the switcher says "The screen is
    locked". Apps behind keep running, so a song plays on under the lock.
  • Switch user left the previous user's apps open, so the next person
    was one Alt+Tab from somebody else's shell, signed in as them. When a
    different user signs in, the previous user's apps are now closed,
    and an alert says so. The same user signing back in finds everything
    as it was.
  • The installer holds a switch lock (TK_SOLO): nothing else runs
    and Alt+Tab is refused while it is open.

Fixed along the way

  • The Music Player had never played a note. "Any key stops it" was
    written as keyboard_getchar()!=0, and "no key" is -1, so the loop
    stopped before its first note on every machine. The screen said "Now
    playing" over silence. It now plays, with each note timed by the clock
    instead of a busy loop calibrated in CPU cycles.
  • tools/make_appreg_inc.py left out every app that set fits
    (Shell, System, Minefield, Settings): its pattern ended at the
    category. test_menugeom was checking 43 apps out of 50. It reads
    all 50 now, and fails the build if a registry row can't be parsed. The
    menus still fit.
  • The boot harnesses had not booted since 0.9.5.1.
    appguard_boot.py, helios_boot.py, lockdown_boot.py and
    layout_audit.py waited for "System ready" on screen, and 0.9.5.1
    clears the screen before the shell's welcome. They wait for the
    prompt as well now. appguard_boot.py also waits out the crash
    notice's 600 ms key hold before pressing Esc; pressing it sooner was
    discarded about one time in three, by design.
  • ping yields between echoes, so ping ... & doesn't hold the
    machine for all four.
  • resolution refuses with other apps open. Each open app's console
    is laid out for the current grid.

Also in 0.9.6: moving an install from FAT16 to FAT32 (migrate)

Every install made before 0.9.3 is FAT16, and FAT16 stops at 2 GB.
0.9.3 made new installs FAT32, but an existing install could only get
there by being wiped and reinstalled. Now it can move itself, on the
same disk, from inside HavenDOS.

  1. Make room. In Windows' Disk Management, shrink a volume and
    either leave the space unallocated, or make a partition and pick
    "Do not format this volume". 512 MB is the least it will take.
  2. migrate lists where the install can go: free space, or empty
    partitions. It also lists what it is leaving alone and why ("holds
    NTFS", "Windows recovery", "Microsoft reserved").
  3. migrate 1 (or whichever number) says exactly what it will do,
    then waits for you to type a word: MOVE for free space, ERASE for an
    existing partition. Then it:
    • makes the partition (or marks the empty one as HavenDOS's);
    • formats it FAT32 and copies this system's kernel onto it;
    • copies everything from the FAT16 install except /boot: users,
      passwords, settings, the iPhone pairing, your home folder;
    • adds it to the boot menu as the first HavenDOS entry.
      Windows stays the default. The old install stays too, as the third
      entry.
  4. Reboot, pick the first HavenDOS entry, log in, and check your
    files are there.
  5. migrate finish, from the new install, after typing REMOVE:
    • takes the old install out of the menu;
    • erases its boot sector;
    • deletes its partition, which leaves that space unallocated for
      Windows.

What keeps this safe

  • It never takes a partition with a filesystem on it, whatever its
    type says. It reads the first sectors and refuses NTFS, exFAT, FAT,
    BitLocker, ReFS, ext2/3/4, XFS, btrfs, LUKS, Linux ...
Read more

HavenDOS 0.9.5.1

Choose a tag to compare

@github-actions github-actions released this 15 Nov 10:38
c8552dd

Which file do I want?

File For
HavenDOS-0.9.5.1.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.5.1-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.5.1-source.zip Building it yourself -- make && make test

HavenDOS v0.9.5.1 Beta "Hephaestus": the modern look, everywhere

15 November 2026.

Kyan liked the look of the new Settings and System apps in 0.9.5 and
asked for it on every app and every screen, including login, setup and
the installer. He also asked for new games and apps.

This release does both. Multitasking is still next, in 0.9.6.

The look, written down

The rules are in a comment block at the top of shell/ui.c, and the
helpers that enforce them are new in the same file:

  • No dashed boxes. A panel is a filled title band over a plain body
    (ui_kit_panel). ui_draw_window() now just calls it, so every old
    caller got the new look at once.
  • One top bar. The app's name is on the left, "ESC close" and the
    clock on the right (ui_draw_taskbar).
  • One selection colour. ui_sel_colors() gives the theme accent as
    a background, with black or white text chosen for contrast. Nothing
    uses a bright background any more; on VGA text those blink.
  • Keys go in a footer. A dark grey strip on the bottom row
    (ui_kit_footer, ui_kit_footer_at).
  • Headings in the accent, labels dim, values bright (ui_kit_heading,
    ui_kit_kv), plus ui_kit_row, ui_kit_cell and ui_kit_bar for
    lists, tables and meters.

Every screen, restyled

  • Desktop.
    • The top bar shows "HavenDOS" and the version in the accent colour,
      with today's date on the right.
    • The bottom bar has a "HavenDOS" start pill, the user and role, an
      alerts pill ("! 2 new", or "no alerts") and the clock.
    • Icons are solid tiles in a colour per category. The selected tile is
      drawn in the selection colour.
    • The info strip is just "RAM free / up", with no dashed rule.
  • Start menu, submenus and the right-click menu have header bands,
    highlighted rows and footers that list their keys. The right-click
    menu's "Display" item was renamed "Next theme", which is what it does.
  • The notification centre and the frame around boxed apps (solid
    edges, wallpaper around it).
  • Games.
    • Snake has a solid border and cells, a footer and a game-over panel.
    • Tetris has solid pieces, a [] ghost, and a sidebar panel with
      score, lines, level and next piece. The old sidebar ran past the
      edge on an 80-column screen; the new one doesn't.
    • Pong has solid walls, paddles and ball, and its key hint moved from
      the title bar into a footer.
  • Text Editor (renamed from "Editor"): the file line is quiet text,
    not a blue bar.
  • Paint, Music Player, Quiz and Calendar have theme bars and
    footers. Calendar marks today in the selection colour. Quiz shows a
    green or red result band.
  • PCI, USB, Boot Log, Self Test and Neofetch use the standard top
    bar. Their "press a key" lines are footers.
  • Terms and Privacy: no dashed rules; a solid scrollbar.
  • Shell: the welcome is a bar (a "HavenDOS" pill, "Shell" and the
    clock) and a fortune, without dashed rules. ls lost its 56-dash
    line.
  • Boot splash, setup, installer and login: solid progress bars,
    solid bands, no dashed separators.

New games

Open them from the Games menu, or from the shell by name or with open.

  • 2048 (2048). Arrow keys slide the tiles; tiles are coloured by
    value. The best score is kept while the system runs.
  • Breakout (breakout). Five rows of twelve bricks, three lives,
    Space to launch.
  • Tic-Tac-Toe (tictactoe, ttt). The computer makes mistakes
    by default. D switches it to perfect play (minimax), which never
    loses.
  • Memory (memory). Sixteen cards, eight pairs, with a move count.

New apps

  • Notes (notes). Up to 18 short notes, saved to
    HOME/<user>/NOTES.TXT. A adds a note, Enter edits it, D deletes it.
  • Stopwatch (stopwatch). Big block digits, laps (L), and a
    countdown timer that beeps (Tab switches mode).
  • Converter (convert): length, weight, temperature, data size,
    speed and time. It works in millionths with its own 64-bit division,
    because the kernel has no FPU code and no libgcc.

open

open on its own lists every app by id. open <id or name> opens one,
matched by id or by a case-insensitive name prefix. For example,
open tetris, open text or open sys.

Stale apps removed

Disk Info, Network Info and User Manager still described the 0.7-era
system: Disk Info reported RAMFS and Network Info reported VirtIO.

  • Disk Info now opens System > Storage.
  • Network Info now opens System > Network.
  • User Manager now opens Settings > Accounts.

Fixes

  • The Haven theme's title text is now light cyan, to suit the new
    bars.
  • Game scores in the top bar overlapped "ESC close". They have moved
    left.
  • The login screen's account tiles and its "Welcome to HavenDOS" choice
    used the bright accent as a background. They now use the selection
    colour like everything else.
  • Settings > About ran past the right edge on an 80-column screen. The
    three long lines are shorter.
  • A toast that expired while the shell was open left a stripe of desktop
    colour across the shell. It now blanks to black; the desktop repaints
    that row itself.
  • The shell now starts from a cleared screen, so nothing from the login
    screen shows through under the welcome.
  • Pong showed the shell line it was opened from on row 1.
  • The notification drawer was 25 columns wide and cut messages off. It
    is now 34.
  • The login "welcome back" screen said "VirtIO mounted" or "Disk: not
    found", and only knew two network cards. It now says whether a disk
    is mounted and whether a network adapter is up, and it centres on any
    console size.
  • Settings > Appearance previews a window the new way (title band, no
    dashed frame). Terms and Privacy has the standard bar, a panel and a
    footer.

Tests and checks

  • New suite test_games2 (33 checks):
    • 2048 slides and merges, including that a tile merges only once per
      move;
    • Tic-Tac-Toe's perfect computer never loses, checked over every line
      of play;
    • the converter's conversions;
    • udiv64 checked against the C library;
    • stopwatch formatting.
  • test_chrome was rewritten for the new bars: 58 checks, floor
    raised from 42 to 58.
  • test_theme gained the stubs the new bars need.
  • check_help.py holds the eight new commands and open to the
    command reference, like every other command.