[stevevanhooser]

Why

The bridge checks on main fail on NDI-matlab's Update-button rework (19b8ab573, 3bc99bf85, 749b1a266):

• updateDialog, packageStatus, pullRepo and findCalculators had no bridge entries;
• check, install, checkOnStartup and ndiPane had drifted;
• epochid2element, filename2epochid and setup had also drifted.

Bumping hashes without porting would have claimed parity Python didn't have, so this ports the change.

What

• ndi.update.packageStatus (new): one row per package, with MATLAB's fields, notes, levels and tick rules.
  • A git install lists NDI-python plus each clone in ~/.ndi/tools.
  • An installer install is a single row for NDI-python.
• ndi.update.install(Packages=[...]) updates only the named packages.
  • Unknown names raise ndi:update:install:unknownPackage.
  • Only changes in the named packages need consent.
  • The clone must be on main only if "NDI-python" is named.
  • Naming "NDI-python" alone pulls main and runs ndi_install.py without --update, so the dependency clones aren't touched.
  • A named dependency is pulled in ~/.ndi/tools, where the .pth file already points.
• Pulling now goes through _pull_repo, MATLAB's pullRepo. It stashes edits (with a ref check, so an older stash is never popped), runs git pull --no-edit, then stash pop. Local commits are merged rather than refused by --ff-only.
• ndi.update.check:
  • Dependencies gain CurrentVersion, LatestVersion and Url.
  • Calculators / CalculatorUpdateAvailable exist but are always empty / False. NDI-python has no NDIcalc clones; it gets ndicalc-vis-matlab from pip.
• ndi.gui.nav.update_dialog (new): the modal Qt package table.
  • Pure helpers testable without Qt: table_data, note_text, update_button_state.
  • UpdateDialog carries MATLAB's property names; ask() opens it and returns the choice.
  • NdiPane.check_for_updates opens it and installs the ticked packages. The old confirm-box update_dialog() goes, as MATLAB's old dialogs did.
• Bridge YAMLs:
  • New entries: packageStatus, pullRepo (ported_differently), findCalculators (matlab_only), updateDialog.
  • Ported and bumped: check, install, checkOnStartup, ndiPane.
  • Bumped with a reason, nothing to port:
    • epochid2element and filename2epochid: MATLAB's fixes for repeated ids/filenames are no-ops here, since Python already returns every match.
    • setup: every change since 03c79fa48 is in +conv/+vhlab, the Stacy legacy importer, which has no Python port.

Testing

• New tests/test_update_package_status.py (18 tests):

  • packageStatus on real throwaway git clones: up to date; behind; your changes left unticked; changes in a package that isn't updating; an untracked file in the way; another branch. Also installer and pip installs.
  • install(Packages=...):
    • only the named dependency is pulled;
    • another branch and edits in NDI-python don't block a dependency update;
    • named changes still need consent, and survive the update;
    • NDI-python alone needs main and the installer, and runs it without --update;
    • unknown and up-to-date names.
  • The dialog helpers; the Qt window offscreen (ticks, snap-back, button text and default); NdiPane.check_for_updates installs only the ticked packages.

• Existing update tests still pass (59 in the update group).

• Full suite: 6477 passed. One failure each in bridge completeness and bridge drift remains:

  • completeness: no entry for +ndi/+daq/+fun/ingestEpochDocuments.m;
  • drift: daq.system and session have moved in NDI-matlab.

  Both are fixed by Ingest: progress bars and parallel per-epoch work (mirrors NDI-matlab#1079) #345, which ports those. With Ingest: progress bars and parallel per-epoch work (mirrors NDI-matlab#1079) #345 and this PR merged, the bridge checks pass against NDI-matlab main.

---

Generated by Claude Code

[stevevanhooser]

MATLAB/Python bridge completeness now has 1 failure, down from the 2 that main has; the src/ndi/gui and src/ndi/update checks pass. The one left is +ndi/+daq/+fun/ingestEpochDocuments.m.

That function came in with NDI-matlab#1079, and its port and bridge entry are in #345. The daq.system / session drift in the hash check is the same story. No fix belongs here: once #345 merges, re-running this PR's CI should clear both.

---

Generated by Claude Code