* feat(transactions): "Ta bort underlag" detach action on a transaction (#2132)
Wrong receipt pinned, no way back: the DELETE
/api/transactions/[id]/attach-document route and its tests already existed,
but nothing in the UI called it. This wires it up, frontend only.
- Inbox card and history list: "Ta bort underlag" in the row menu, shown only
for writers on unbooked rows that carry a pin (canDetachDocument helper).
- Attach dialog: a small "Ta bort underlag" link beside the already-attached
hint, the one place the app previously admitted a doc was pinned.
- Page: handleDetachDocument confirms (useDestructiveConfirm, warning), then
DELETEs; 200 clears document_id in local state (list, dialog snapshot, and
the inbox card's optimistic override via a -unlinked window event) and
toasts; 409 renders the route's Swedish BFL message verbatim; other errors
map through get-error-message.
- Strings under tx_detach in sv.json and en.json.
- Tests: gate hidden when booked / read-only / no pin / no handler; 409
rendered unchanged; wiring and locale assertions.
Out of scope, follow-up: MCP detach tool (new pending-op type + CHECK
migration), detaching from the inbox for non-email docs, and clearing
invoice_inbox_items.matched_transaction_id on detach so the doc is offered
again by inbox-available.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YRXN5CqHrfuuDw5LLcSgTg
* fix(transactions): clear the inbox back-link when detaching underlag (#2132)
Skeptic finding on PR #2144: DELETE attach-document nulled only
transactions.document_id and left invoice_inbox_items.matched_transaction_id
pointing at the transaction. propagateUnderlagForBookedTransaction selects
on exactly that column at categorize / book / bulk-book time, so the
detached receipt would have been re-anchored onto the new verifikation as
immutable underlag (BFL 5 kap 7 §), and the doc never reappeared in
inbox-available for re-matching.
The route now clears the back-link for the detached document, scoped to
items not yet consumed by a verifikat (created_journal_entry_id null),
mirroring the invoice-inbox extension's unmatch. Best-effort like the POST
side: the pin removal is the primary effect. Three DELETE tests cover the
filters, the no-pin case, and a failing unlink. DECISIONS.md and the PR body
record the accepted bulk-booked-row limitation in the history list.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YRXN5CqHrfuuDw5LLcSgTg
* fix(transactions): detach reports a failed inbox unlink instead of success (#2132)
Swedish compliance review on PR #2144: the inbox back-link cleanup was
fire-and-forget, so a failed UPDATE returned 200 while leaving exactly the
stale matched_transaction_id that re-anchors a detached document onto the
next verifikation (BFL 5 kap 6-7 §).
The unlink is now scoped by transaction only (the unique index on
matched_transaction_id means at most one item points here, and a stale item
from the replace path would re-anchor just the same), runs even when nothing
was pinned so a retry is idempotent, and a failure answers 500 with an honest
Swedish partial-failure message, mirroring the POST side's propagation
failure. Tests updated accordingly.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YRXN5CqHrfuuDw5LLcSgTg
* fix(transactions): release inbox back-link before a compare-and-set pin clear (#2132)
Review findings on PR #2144, one pass:
- CodeRabbit (major): DELETE cleared the pin and then released the inbox
back-link scoped by transaction, so a POST landing in between could end up
as "new doc pinned, its inbox item unlinked". The release now runs FIRST,
and the pin clear is a compare-and-set on the document that was read
(.eq document_id, or .is null when nothing was pinned). Zero rows answers
409 "ändrades samtidigt" and keeps the newer pin. A failed release returns
500 before anything changed, so a retry is trivially idempotent.
- Compliance swarm (A.8.15): the unlink failure log carried the raw driver
error; it now logs errorCauseTag() only.
- CodeRabbit docstring check: JSDoc on handleDetachDocument.
Tests: order of the two writes, CAS filters for both pinned and empty
states, 409 on concurrent re-attach, coded-cause logging.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YRXN5CqHrfuuDw5LLcSgTg
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Mattsson
4c76fb10d7
