[tsaubergine]

Replaces both submodules with Open Ocean Software's continuous packages from openocean-build (apt.oceansoft.org). IvPico now has no submodules.

Changes

• init.sh installs libopenocean-messages-dev and libivp-helm-dev.
  • If apt doesn't have them yet, it adds apt.oceansoft.org/continuous/<distro>-<dist>, the same way it adds the Goby source.
  • The key comes from openocean-build's keys/continuous.asc, and the script stops unless its fingerprint is DAF0 8DD7 788E 6C84 C6F8 E8D3 E59E 1A81 FECD B228.
  • It no longer checks out submodules.
• openocean-messages: the external/openocean-messages submodule is removed.
  • The package is 0.0.1: upstream renumbered from 0.1.0 without changing the messages. So find_package(openocean_messages 0.0.1 …).
• MOOS-IvP: the external/moos-ivp submodule is removed in favour of libivp-helm-dev.
  • That package is the helm libraries built without MOOS, from GobySoft's ivp-without-moos.
  • It installs under its own prefix, /usr/lib/<arch>/ivp-helm, apart from a full MOOS-IvP, so find_package(IVP …) gets a HINTS for it.
  • The libraries are found at run time through its ld.so.conf.d entry.
• build.sh builds only IvPico.
• CI no longer installs git, which was only needed for the submodules.
• src/test/missions/record.sh shallow-clones ivp-without-moos (override with MOOS_IVP_BRANCH) to build all of MOOS-IvP for recording. It also adds MOOS-IvP's bin/ to PATH, which is where the current build puts MOOSDB.
• Alpha launch: a new get_images.sh downloads the background images that pMarineViewer and set_datum_warp.sh used to read from the submodule. These are forrest19 and MIT_SP from MOOS-IvP's ivp/data, about 21 MB, saved into an ignored images/ directory. viewer.launch and set_datum_warp.sh now look there.
• READMEs updated to match.

Testing

• Clean Ubuntu 26.04 and 24.04 containers (init.sh, build.sh, ctest): both packages install from apt.oceansoft.org, and all 15 tests pass, including the replays of MOOS-IvP missions and the Goby closed-loop missions.
• With both packages installed here, CMake finds IVP at /usr/lib/x86_64-linux-gnu/ivp-helm/lib/cmake/IVP, and goby_helm_ivp loads libhelmivp.so.0 from that prefix.
• get_images.sh downloads the four files, and set_datum_warp.sh picks MIT_SP.tif for the MIT datum and forrest19.tif for the default.
• CI on all four jobs.
  • The first run's Ubuntu 24.04 arm64 job failed replay_s7_golf, from an uninitialised m_mark_time in MOOS-IvP's BHV_PeriodicSurface.
  • That's fixed in BHV_PeriodicSurface: initialise m_mark_time moos-ivp#4, which went into the continuous libivp-helm (24.8.1+git5535.gc535209). The re-run passes.
• record.sh with the cloned branch: not run here.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VLcYK4e29B5FiPbDDJy5fp

[tsaubergine]

Ubuntu 24.04 (arm64) fails replay_s7_golf; the other three jobs pass:

FAIL  bhv_periodic_surface PENDING_SURFACE (1 changes, 1 off)
        pHelmIvP iteration 20: "-2147483648.00000", IvPico iteration 20: "2.14748e+09"

The cause is an uninitialised read in MOOS-IvP, not IvPico.

• BHV_PeriodicSurface::onRunState() posts PENDING_SURFACE = m_period - (m_curr_time - m_mark_time) before its m_first_iteration block sets m_mark_time.
• The constructor never initialises m_mark_time, so the first run-state iteration posts a value computed from whatever was in memory. That value overflows postIntMessage's (int) conversion: INT_MIN in the recording, INT_MAX here.
• Both the recording and the replay posted this undefined value. They matched while IvPico built the helm libraries itself, and stopped matching with the packaged libivp-helm0 on 24.04 arm64. A re-run wouldn't settle anything, since it could pass by luck.

The fix is to initialise m_mark_time in BHV_PeriodicSurface's constructor, in MOOS-IvP (ivp/src/lib_behaviors-marine/BHV_PeriodicSurface.cpp). None exists yet, and where it goes (GobySoft's ivp-without-moos, which the package is built from, or a separate upstream fix) is being decided before anything is pushed.

---

Generated by Claude Code