fix(l10n): stamp a missing l10nSource when a connected translation is opened #345

Merged

dominik.polakovics merged 1 commit from afk/343 into main

2026-10-23 09:04:03 +02:00

dominik.polakovics commented

2026-10-23 08:58:59 +02:00

Owner

Closes #343

What

Opening a connected translation in the editor (GET /api/page, GET /api/page/elements, via openPage) now stamps, after the existing l10n.Repair:

every translated element, nested element and item without l10nSource with its source element's current hash (l10n.StampMissing, built on walkElements, sourceParam, hashOf, setSourceParam);
the front matter's page-level l10nSource when missing (l10n.StampMissingPage → pageops.TranslationPageStamp, same pageFormFields as TranslationPageChanged).

The write is persisted with the existing repair translation snapshot. Edge cases:

An existing hash, even a stale one, is never overwritten.
A free translation is never stamped.
An element with no same-uid, same-name source element gets no stamp.
The stamp runs under the same condition as the repair: the source's uids must be sound.
A present value that is not a string is treated as no hash and is stamped.
The tree still does not write, so before the first open it shows translated.

Rule owners

"Is this element stamped" stays sourceParam.
"Does this front matter carry a page hash" moves into pageSource, which PageChanged and StampMissingPage share.
Hashing is only Hash/ItemHash/PageHash.

Docs and skills

scaffold/site/AGENTS.md: "Change tracking", "Change tracking of the page's own fields" and "Translations by hand" now say the editor stamps a missing hash on first open.
docs/api.md: GET /api/page "Languages" and "Writes", the translations overview and pageChanged.
docs/architecture.md: "Multilingual sites".
Skills: I re-read edit-content and migrate-site. Neither contradicts the new behaviour, since both point at AGENTS.md, so they are unchanged.

Tests

l10n: TestStampMissing and TestStampMissingPage.
pageops: TestTranslationPageStamp.
server:
- TestOpenStampsMissingSourceHashes is the regression from the report. It failed before this change and passes now.
- TestOpenStampTracksParamAndTitleChanges
- TestOpenStampEdgeCases
- TestOpenStampSnapshot
go vet ./... and go test ./... are green. web/ is untouched.

🤖 Generated with Claude Code

Closes #343 ## What Opening a connected translation in the editor (`GET /api/page`, `GET /api/page/elements`, via `openPage`) now stamps, after the existing `l10n.Repair`: - every translated element, nested element and item without `l10nSource` with its source element's current hash (`l10n.StampMissing`, built on `walkElements`, `sourceParam`, `hashOf`, `setSourceParam`); - the front matter's page-level `l10nSource` when missing (`l10n.StampMissingPage` → `pageops.TranslationPageStamp`, same `pageFormFields` as `TranslationPageChanged`). The write is persisted with the existing `repair translation` snapshot. Edge cases: - An existing hash, even a stale one, is never overwritten. - A free translation is never stamped. - An element with no same-uid, same-name source element gets no stamp. - The stamp runs under the same condition as the repair: the source's uids must be sound. - A present value that is not a string is treated as no hash and is stamped. - The tree still does not write, so before the first open it shows `translated`. ## Rule owners - "Is this element stamped" stays `sourceParam`. - "Does this front matter carry a page hash" moves into `pageSource`, which `PageChanged` and `StampMissingPage` share. - Hashing is only `Hash`/`ItemHash`/`PageHash`. ## Docs and skills - `scaffold/site/AGENTS.md`: "Change tracking", "Change tracking of the page's own fields" and "Translations by hand" now say the editor stamps a missing hash on first open. - `docs/api.md`: GET /api/page "Languages" and "Writes", the translations overview and `pageChanged`. - `docs/architecture.md`: "Multilingual sites". - Skills: I re-read `edit-content` and `migrate-site`. Neither contradicts the new behaviour, since both point at AGENTS.md, so they are unchanged. ## Tests - `l10n`: `TestStampMissing` and `TestStampMissingPage`. - `pageops`: `TestTranslationPageStamp`. - `server`: - `TestOpenStampsMissingSourceHashes` is the regression from the report. It failed before this change and passes now. - `TestOpenStampTracksParamAndTitleChanges` - `TestOpenStampEdgeCases` - `TestOpenStampSnapshot` - `go vet ./...` and `go test ./...` are green. `web/` is untouched. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dominik.polakovics added 1 commit

2026-10-23 08:58:59 +02:00

fix(l10n): stamp a missing l10nSource when a connected translation is opened

ci / build-and-test (pull_request) Successful in 3m52s

Details

ci / release (pull_request) Has been skipped

Details

c6bb7c302f

A connected translation written without change-tracking hashes (by hand,
by `harness lang add`, imported) was never marked changed, however much
its source changed. Opening it in the editor (GET /api/page, GET
/api/page/elements) now stamps every translated element, nested element
and item that has no l10nSource with its source element's current hash,
and the front matter with the page-level hash, persisted with the
existing "repair translation" snapshot. An existing hash is never
overwritten; free translations are left alone.

- l10n.StampMissing and l10n.StampMissingPage reuse sourceParam, hashOf,
  PageHash and StampPageChanges; the page-level "carries a hash" rule
  moves into one helper (pageSource) shared with PageChanged.
- pageops.TranslationPageStamp uses the same properties-form fields as
  TranslationPageChanged.
- AGENTS.md "Translations", docs/api.md GET /api/page and
  docs/architecture.md describe the stamp.

Closes #343

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

dominik.polakovics commented

2026-10-23 09:04:02 +02:00

Author

Owner

[autoland] verdict: pass

PASS — signal: ci / build-and-test success (run 365; release skipped). Title is Conventional Commits, body carries Closes #343. Diff matches the issue: l10n.StampMissing / StampMissingPage (sourceParam stays the one 'is stamped' owner, pageSource shared with PageChanged, hashing only via hashOf/PageHash), pageops.TranslationPageStamp over the same pageFormFields as TranslationPageChanged, wired into openPage (both GET /api/page and GET /api/page/elements) under the existing repair-translation snapshot. Edge cases (stale hash kept, free translation untouched, no same-uid/same-name source skipped) are tested at l10n, pageops and server level, plus the #343 regression. AGENTS.md, docs/api.md and docs/architecture.md updated; skills unaffected (they point at AGENTS.md). Base moved by #344 (supervisor only, disjoint); merge-tree is clean.

[autoland] verdict: pass PASS — signal: ci / build-and-test success (run 365; release skipped). Title is Conventional Commits, body carries Closes #343. Diff matches the issue: l10n.StampMissing / StampMissingPage (sourceParam stays the one 'is stamped' owner, pageSource shared with PageChanged, hashing only via hashOf/PageHash), pageops.TranslationPageStamp over the same pageFormFields as TranslationPageChanged, wired into openPage (both GET /api/page and GET /api/page/elements) under the existing repair-translation snapshot. Edge cases (stale hash kept, free translation untouched, no same-uid/same-name source skipped) are tested at l10n, pageops and server level, plus the #343 regression. AGENTS.md, docs/api.md and docs/architecture.md updated; skills unaffected (they point at AGENTS.md). Base moved by #344 (supervisor only, disjoint); merge-tree is clean.

dominik.polakovics merged commit de2ff381db into main

2026-10-23 09:04:03 +02:00

dominik.polakovics referenced this pull request from a commit

2026-10-23 09:04:05 +02:00

Merge pull request 'fix(l10n): stamp a missing l10nSource when a connected translation is opened' (#345) from afk/343 into main