feat(i18n): translate the GUI strings left in English, guard against drift, document the workflow

[Guts]

Description

Widget text goes through Qt's own i18n, whose .ts files are refreshed by running pylupdate6 by hand. Three merged PRs — cancellation, parallel scan, driver-based format gating — added widget text without that step, so 8 strings had no French or Spanish translation at all and showed in English whatever the language selected:

Context	Source string
DicoGIS	Cancel
DicoGIS	Canceling...
DicoGIS	Folder scan canceled.
DicoGIS	Processing canceled. Partial results were saved to the output.
TabFiles	Not available: the installed GDAL build does not include the driver required for this format.
TabSettings	Parallelize folder scan (recommended only for network-mounted / high-latency storage; …)
TabSettings	Parallel scan: max workers (0 = auto)
TabSettings	Auto

Ran pylupdate6 over dicogis/ui, translated the 8 new entries in both locales, and recompiled the .qm with lrelease. The .ts files go from 68 to 76 messages, with 0 unfinished in FR and ES.

Verified at runtime, not just on paper: a QTranslator loading each compiled .qm returns the expected translation for all 8 — e.g. Cancel → Annuler / Cancelar. Existing translations are untouched; the FR/ES wording follows the file's conventions (French typographic space before :, none in Spanish).

Correcting my own earlier count, in case it was quoted anywhere: I had said "at least 9 strings". It is 8 — Fund & Support was already translated, my grep had missed it because the .ts escapes it as Fund & Support.

Guarding against the next drift

The real defect is that nothing enforced the pipeline being re-run, so tests/test_ui_i18n.py is added:

• test_every_widget_string_is_translated collects every self.tr() source string by walking the GUI modules with ast — so it needs neither PyQt6 nor GDAL and runs in any environment — and fails naming the offending strings.
• test_compiled_qm_matches_the_ts loads the compiled .qm and compares it against the .ts. Only the .qm is read at runtime, so a .ts updated without re-running lrelease changes nothing for the end user; this catches that half of the mistake. It skips when PyQt6 is absent.
• test_translatable_sources_are_found guards the guard: an extraction returning nothing would make the rest vacuously true.

Both fail on the previous state, listing exactly those 8 strings:

AssertionError: 8 string(s) have no FR translation. Re-run pylupdate6 (see CLAUDE.md),
fill the new <translation> entries and recompile the .qm: ['Auto', 'Cancel', …]

New page: docs/development/translation.md

The pipeline was only described in CLAUDE.md, an agent-facing file, so a human contributor had nowhere to learn it — which is how those 8 strings shipped untranslated in the first place. Modelled on qtribu's translation page, updated for PyQt6 and for this project:

• pylupdate6 ships with PyQt6 itself, so the gui extra suffices; only lrelease and Linguist need system packages (qt6-l10n-tools, linguist-qt6), which Ubuntu installs outside the default PATH;
• no .pro profile is involved, unlike pylupdate5 — pylupdate6 takes the source files directly;
• unlike qtribu, whose CI generates them, the compiled .qm are tracked here: they are package data and PyInstaller --add-data inputs, so a stale .qm ships a stale interface;
• dicogis_en.ts is a source-language reference, never compiled nor loaded;
• the second, separate mechanism — TextsManager and locale/lang_*.xml — is documented alongside, since which one applies depends on whether the string is GUI or pipeline output.

It also covers adding a new language and points at the new tests. Package names and binary paths were checked against the real packages rather than assumed: the Qt 6 Linguist binary is linguist, not linguist6.

Sphinx build verified: it succeeds, the page is registered in the index toctree, and the warning set is identical to master's (55 non-intersphinx warnings either way). One iteration was needed — a ### Requirements heading collided with development/ubuntu.md under autosectionlabel, so it is now ### Required tools.

Related issue(s)

None — found while reviewing the code ahead of the 4.0.0-beta13 release.

Type of change

• Bug fix (fix/… branch)
• New feature or enhancement (feature/… branch) — user-visible translations plus a new guard; the branch is named fix/… since it repairs shipped behaviour, rename or relabel as you prefer
• Documentation (docs/… branch or *.md/docs/** changes)
• Packaging / build (packaging/… branch or builder/** changes)
• Tooling / CI-CD (tooling/… branch or .github/** changes)
• Tests only (tests/** changes)

Checklist

• I have read the Contributing Guidelines
• pre-commit run --all-files passes locally — its two ruff hooks on the changed files: ruff-format reports them already formatted, ruff check --fix-only rewrites nothing, and ruff check reports no violation on the new test. It needed one # noqa: S314 with a stated reason, the parsed file being the project's own .ts; dicogis/utils/texts.py carries the same unsuppressed pattern on master, so tell me if you would rather I dropped the suppression for consistency
• I have added or updated tests covering my changes
• I have updated the documentation where relevant — the new page, plus the index toctree
• pytest passes locally — 226 passed on the non-GDAL modules and 41 on tests/ui, with the same 8 pre-existing sandbox failures as on master

AI/LLM disclosure

• Level: 7 — human specced, bots coded. The FR and ES translations are the model's own wording; a native reading is worth a minute before merging.
• Model: claude-opus-5
• Session: https://claude.ai/code/session_01WABM4NMiD2DhEVreKLxWLq

🤖 Generated with Claude Code

https://claude.ai/code/session_01WABM4NMiD2DhEVreKLxWLq

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 76.09%. Comparing base (b92f5b0) to head (e307c44).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #499   +/-   ##
=======================================
  Coverage   76.09%   76.09%           
=======================================
  Files          46       46           
  Lines        3409     3409           
  Branches      415      415           
=======================================
  Hits         2594     2594           
  Misses        695      695           
  Partials      120      120           

Flag	Coverage Δ
unittests	76.03% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud