github dekajoule/oos v0.1.15
OOS 0.1.15

one hour ago

Released on 2026-11-21. Fixes from the forum's reports on 0.1.14 (the picture while the trigger level turns, level
50 % with AC coupling, the USB stick going back, SSH on IPv6) and from W4R's pull requests (Run after Single, the I2C,
UART, CAN and LIN decoders, the spectrum's fundamental, CSV references), a stricter web server, and the memory above
256 MiB for hardware segments, off until a probe on the unit finds it. Under the hood the acquisition loop runs on a
simulated unit on the host, the model and the store are crates of their own, and the board family is a parameter of
the build, the tools and CI; none of this changes what the scope does. All of it is tested on the host; marked
"(unit)" where it also ran on a DHO924S. A build of main from the same day ran on a DHO924S from its kexec start and
its first power-on, which confirmed it.

Added:

  • System > Probe optional memory probes the memory above 256 MiB for every segment size of the Acquire menu's
    depths and switches it on when every size held; System > Opt. memory switches its use off and on again (on
    without a probe's yes probes first). Off by default; kept in system.conf (mem.upper, mem.upper.found)
    (First Steps).
  • Hardware segments can use the acquisition memory above 256 MiB where the unit has it: twice as many segments of a
    size. The stock software uses only 256 MiB; the other half was found by observation and measurement on a unit. It
    is off until System > Probe optional memory found it on the unit (next entry), and a short probe at the first
    recording of a segment size the button did not probe checks it again. System info shows whether it is in use
    ("Optional memory") (horizontal-config.md)
    (unit).
  • A Windows button under the zoom magnifier opens the Windows menu (First Steps).

Interface:

  • GET /cmd (experimental): a request with one command a remote may not give is refused as a whole, and none of its
    commands runs. Before, the allowed ones ran anyway (docs/http.md).
  • Every answer of the web server has Cache-Control: no-store, also a 503 busy, and the 202 of an MCP notification
    has no Content-Type instead of an empty one (docs/http.md).
  • The web server tells the methods of HTTP apart: each route takes the method of its reference (and HEAD where it
    takes GET, but /cmd and the live pictures), another method gets 405 with an Allow header, and a HEAD gets
    no body. Before, every
    method but POST was taken as GET, so POST /wave.csv gave the CSV and GET /mcp an MCP parse error
    (docs/http.md).
  • POST /update (stable) answers a refused package with 409 instead of 200 (the body still says REFUSED: <reason>),
    and POST /waveform (stable) answers a request without a slot from 1 to 10 with 400 instead of 409. A client can
    tell from the status that nothing was installed or loaded (docs/http.md).
  • GET /lxi/identification (stable) names the unit's model, as *IDN? does. It named every unit a DHO924S, so a
    DHO804 or DHO914 was found as one in an LXI discovery (docs/http.md).

Removed:

  • Builds without some decoders: OOS_DECODERS, OOS_DECODERS_OFF and the Cargo features dec-* are gone, and every
    decoder is always in oos-ui, as it was in every release image. System > Features leaves decoders out at run time
    (protocol-modules.md, ADR 0071).

Changed:

  • A picker longer than the screen scrolls by its bar, a finger and the knob, as the other lists do (ADR 0006): a WLAN
    scan lists every network it found, not the 17 strongest, and the Bluetooth pickers every device, not the first 18.
  • The full-screen button at the top right of the plot stays in place: a drag that starts on it no longer moves it
    (First Steps).

Fixed:

  • Level 50 % (a press on the trigger level knob, Trigger > Level 50 %) with AC or LF reject trigger coupling sets the
    level halfway through the signal as the trigger sees it, around the signal's mean. It set the middle in volts, which
    lay above the whole swing of a signal that does not cross 0 V, and the trigger did not fire
    (First Steps).
  • The CAN and CAN-FD decoders keep their bit timing on a line that is slow on its way back to recessive: they
    resynchronise on the edges to dominant alone, as a node does. Resynchronised on every edge, they sampled the bit
    behind a slow edge in its neighbour: on CAN_H at 500 kbit/s, back at the threshold 1.1 us late, a frame of every
    cycle had a wrong CRC or a wrong identifier, and frames were missed (Features) (unit).
  • The start from the stock system (the one-shot) and card-install.py take a card only when its partition table also
    has the stock backup partition, as the update already did: OOS keeps its previous image right after it.
  • A CSV file loaded into a reference takes the time between its samples over its whole time column, not from its first
    two lines: a file that writes its times with a few digits read every frequency up to a percent off, and one whose
    first two lines have the same rounded time was refused. In a file of several sweeps the longest one gives the step
    (First Steps).
  • At short time bases, no stray trace now and then from a record joined with older memory before the trigger (a step
    about 38 us left of the trigger at 10 us/div, 1 kpts). The first record after each arm of a long batch is left out,
    which costs one record of 26 to 32 (unit)
    (docs/protocols/acquisition-start.md).
  • The I2C decoder no longer calls the last byte of a read an error: a controller ends every read by not acknowledging
    its last byte, so on a working bus every read was red on the band and an Error (NACK) event of a search. That byte
    is still marked N; a byte not acknowledged anywhere else stays an error (Features).
  • An update that brings another device tree starts by kexec again when the loader seeded the kernel's address
    randomisation: the seed is carried into the new tree as 0, which kexec-tools requires (issue #193; host test; the
    developer's unit writes no seed, so it has not run on a unit).
  • The LIN decoder reads a bus at a rate that Decode > Baud does not list: it measures the bit time of each frame on
    its sync field, as a node does, and Baud is only for a record without a readable sync. A frame whose sync is not
    0x55 is listed with its sync as the error, not left as a break alone, and a byte that comes later than a frame may
    last is no longer taken for its checksum. Found on a capture of a real bus at 18350 bit/s, which read nothing but
    breaks at every Baud but 19200 (Features).
  • A time measurement drawn with Labels on trace whose span runs off the right of a narrow view has
    its label at the right edge of the plot, not at the left; with its edges far off the screen a build with overflow
    checks no longer stops (Measurements and Analysers).
  • The channel traces no longer drop out of single pictures while the scope runs with measurements that are computed in
    the background: the answer of such a measurement cleared the picture until the next record came, so on a DHO924S a
    third of the pictures had no traces, and with persistence the collected picture started over (unit)
    (Measurements and Analysers).
  • A setting changed while running, such as a step of the trigger level or the position, keeps the picture on the
    screen until the next record replaces it. Before, the screen went empty at once. With a Normal sweep and a level
    outside the signal it stayed empty, so the trigger looked broken (unit) (First Steps).
  • :REFerence:VSCale 2, :VOFFset 0.1, :COLor RED, :LABel:CONTent and :SHOW 1|0 without a slot set the
    selected reference slot, as docs/scpi.md says. A value that read like a slot (a number, or a word
    starting with R) was taken for one and refused with -224.
  • On the live acquisition, Run after a Single (:RUN after :SINGle, or Run/Stop after the Single key) goes back to
    the sweep before the Single, Auto or Normal; it went to Auto, so a Normal sweep was lost after every Single. A stored
    replay and SIM still go to Normal (SCPI and VXI-11).
  • A screenshot through MCP or /screen.bmp no longer stops the screen while the picture is encoded: the screen is
    copied, and the picture made beside the frame loop.
  • A SINGLE after Default (*RST), after recalling a set-up or after leaving a reference or SIM session ends with its
    record again: the set-up that came in set the hardware session's counters back, and the second such SINGLE in one
    run of the scope waited forever (First Steps) (unit).
  • The spectrum page and :SPECtrum:ITEM? F1 find the fundamental of a square or a ramp with few periods on the
    screen: a 10 kHz square at 50 us/div read 16.2 kHz, a ramp 12.3 kHz, and THD, SNR, SINAD, SFDR and ENOB were
    figures of the wrong lines. The window of the distortion figures follows the periods in the record (unit)
    (Measurements and Analysers).
  • SSH listens on IPv4 only while net.ipv6 is off (the default), as the web remote, SCPI and VXI-11 do. It listened on
    IPv6 too, where only the firewall kept connections out (docs/network.md).
  • The guide no longer says that the USB stick writes any version: the stick keeps a build on the card that is as new
    as its own or newer and only selects it, so going back to an older build needs an empty oos-reinstall.txt next
    to the .GEL, and makes no rollback image (Updates and Going Back).
  • After STOP the record on screen is the newest one the hardware took, and it stays the record measured and read
    out: a record of the whole memory is no longer replaced by a view read again about 90 ms later, and *OPC? after
    :STOP answers once the record of the STOP is in place, so :WAV:PREamble? and :WAV:DATA? describe the same
    record (SCPI and VXI-11).
  • The UART decoder lists no false first characters when a record begins inside a character: the first falling edge
    of the record was taken for a start bit, and with characters back to back the wrong framing ran on (FF, ED in front
    of a 9600 Bd text line). Up to four real characters at the very start of such a record are left out rather than
    guessed; a record that begins on the idle line a few bit times before a character of a regular stream keeps it
    (unit) (Features).
  • Opening the web page /update no longer stops the screen for as long as oos-update status runs: the page is made
    beside the frame loop.
  • The web page /update shows "no answer within 10 s, stopped" when oos-update status hangs, instead of waiting for
    it and keeping the web server's threads for large answers busy.
  • A web client that does not read its answer (a page, a refusal, the result of a waveform upload) no longer holds
    the screen for up to 2 s per answer: the answer goes out as the client takes it (ADR 0066).

Built by CI from the tag, on arm64, signed with the community key of the repository.
Still to add before this draft is published, the sources a release carries:

linux-6.12.115-base.tar.xz
dropbear-2026.94.tar.bz2
openssh-10.6p2.tar.gz
libmnl-1.0.5.tar.bz2
libnftnl-1.2.9.tar.xz
nftables-1.1.3.tar.xz
kexec-tools-2.0.32.tar.xz
e2fsprogs-1.47.2.tar.xz
wpa_supplicant-2.11.tar.gz
mesa_26.0.8.orig.tar.*|mesa-26.0.8.tar.*
dbus-1.16.2.tar.xz
bluez-5.82.tar.xz
expat-2.10.0.tar.xz
glib-2.84.4.tar.xz
tests-machine-id.patch
flaky-tests.patch
pcre2-10.46.tar.bz2
gettext-0.24.1.tar.xz
cwd.patch
skip-tests-musl.patch
skip-meaningless-pthread-rwlock-test.patch
libnl-3.11.0.tar.gz
libnl3-musl.patch
glibc_2.42-2ubuntu4.dsc
glibc_2.42.orig.tar.*
glibc_2.42-2ubuntu4.debian.tar.*
busybox_*.dsc
busybox_*.orig.tar.*
busybox_*.debian.tar.*
glibc_*.dsc
glibc_*.orig.tar.*
glibc_*.debian.tar.*
gcc-[0-9]*_*.dsc
gcc-[0-9]*_*.orig.tar.*
gcc-[0-9]*_*.debian.tar.*
libdrm_*.dsc
libdrm_*.orig.tar.*
libdrm_*.debian.tar.*
zlib_*.dsc
zlib_*.orig.tar.*
zlib_*.debian.tar.*
SHA256SUMS

Don't miss a new oos release

NewReleases is sending notifications on new releases.