Files
accounted/lib/core/bookkeeping/journal-entry-references.ts
T
4bc007cdeb fix(reports): periodisk sammanställning nets a stornoed or corrected EU invoice voucher (#2354)
* fix(reports): periodisk sammanställning nets a stornoed or corrected EU invoice voucher (#2351)

The PS reads entries with status posted or reversed, so a stornoed EU
invoice voucher kept its 3308/3108 credit while the storno that nets it
was dropped: its source_type is 'storno' and no register row points at
it. A makulerad EU sale was over-reported while the account-based ruta 39
was zero.

Which invoice explains an entry is now resolved by one composed helper,
getInvoicesExplainingJournalEntries (lib/core/bookkeeping/
journal-entry-references.ts): the engine's own source_id, the invoice-side
rows, and the rättelse chain through correction_of_id / reverses_id, walked
upwards under the MAX_CHAIN_WALK cap correction-chain.ts already uses.
Parents outside the batch are fetched by id, company-scoped. The link
columns are followed rather than the storno's copied source_id because
correctEntry() copies none onto its storno or correction, and a storno's
copied source_id is polymorphic (a bank row on a bank booking).

getInvoiceReferencesForJournalEntries (the RPC mirror behind the underlag
surfaces) is unchanged: storno and correction are not doc-requiring source
types, so those surfaces have no such hole.

- INVOICE_SOURCED_ENTRY_TYPES / LINK_LOOKUP_CHUNK move to the helper module
  (the set now names every engine type whose source_id is an invoice).
- PS: one resolver call; reverses_id, correction_of_id in the select; the
  ZERO_NET_EXCLUDED text names makulering beside kreditfaktura.
- Tests: resolver cases (chain inheritance, out-of-batch fetch, own link
  wins, mirror, cycle, cap) and PS cases (the issue's storno nets to
  ZERO_NET_EXCLUDED, rättelse chain, later-period storno, mirror, linked
  import, gone invoice).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LvMaHcTnwAfxzgYD1fGYX1

* fix(bookkeeping): PR #2354 review: bound in-batch chain propagation at MAX_CHAIN_WALK

getInvoicesExplainingJournalEntries attributed an in-batch chain of
stornos/corrections all the way down through a recursive assign over
waitingOn, while parents fetched from outside the batch stopped at
MAX_CHAIN_WALK. The two paths now agree: every attribution carries its
depth from the root (0 for an entry resolved by its own source_id or
invoice-side link), propagation is an explicit queue instead of recursion,
and a descendant more than MAX_CHAIN_WALK links below the root resolves to
no invoice, both when it inherits from an already attributed parent and
when it is reached by propagation. Test: an in-batch chain of
MAX_CHAIN_WALK + 1 links, in both batch orders, attributes the links within
the cap and not the one beyond it, with no parent fetch.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

---------

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-06 21:10:11 +02:00

424 lines
18 KiB
TypeScript

import type { SupabaseClient } from '@supabase/supabase-js'
import { fetchAllRows } from '@/lib/supabase/fetch-all'
import { chunk } from '@/lib/utils'
import { NON_ISSUED_INVOICE_STATUSES_FILTER } from '@/lib/invoices/matchable-statuses'
import { MAX_CHAIN_WALK } from './correction-chain'
/**
* A followable reference from a verifikation back to its underlag: the customer
* or supplier invoice that identifies what the affärshändelse avser and who the
* motpart is.
*
* Surfacing these makes the verifieringskedja traceable from the verifikat side,
* not only from the invoice side (BFL 5 kap 7§: hänvisning till underlag;
* BFNAR 2013:2: the verification chain must be followable in both directions).
*
* Bank transactions are deliberately excluded: a bank line is the trace of the
* affärshändelse, not its underlag. Counting it as underlag would wrongly silence
* the "saknar underlag" warning for expenses that still genuinely need a kvitto.
*/
export type UnderlagReferenceType = 'invoice' | 'supplier_invoice'
export interface UnderlagReference {
type: UnderlagReferenceType
id: string
/** invoice_number / supplier_invoice_number: the UI builds the label from this. */
number: string
/**
* Retained source document owned by a referenced supplier invoice, if any.
*
* Set ONLY when the document is anchored to a journal entry
* (document_attachments.journal_entry_id IS NOT NULL), because that is the
* exact condition every missing-underlag surface uses: the
* verifikat_without_documents / transactions_without_documents RPCs,
* /api/documents/counts and the transactions list all require an anchored
* doc, since only anchored docs sit behind the WORM deletion guards. Handing
* out a floating doc here made the verifikat view display an underlag while
* the list kept warning "Underlag saknas" on the same row (support case
* 2026-07-27). The reference itself is still returned either way, so the
* verifieringskedja stays followable; only the attachment claim is withheld.
*/
document_id?: string
}
interface InvoiceRow {
id: string
invoice_number: string
}
interface SupplierInvoiceRow {
id: string
supplier_invoice_number: string
document_id?: string | null
/** Embedded document row; see UnderlagReference.document_id for why. */
document?: { journal_entry_id: string | null } | { journal_entry_id: string | null }[] | null
}
/** Columns every supplier-invoice lookup below needs, incl. the anchor check. */
const SUPPLIER_INVOICE_COLUMNS =
'id, supplier_invoice_number, document_id, document:document_attachments(journal_entry_id)'
/**
* A supplier invoice's document only counts as this verifikation's underlag
* when it is anchored to a journal entry: an unanchored doc is outside the WORM
* deletion guards, so the missing-underlag surfaces refuse to accept it and
* this resolver must refuse too.
*/
function anchoredDocumentId(row: SupplierInvoiceRow): string | undefined {
if (!row.document_id) return undefined
const document = Array.isArray(row.document) ? row.document[0] : row.document
return document?.journal_entry_id ? row.document_id : undefined
}
/**
* Resolve every customer/supplier invoice linked to a verifikation, across all
* the deterministic FK paths the engine uses to book one:
* - invoices.journal_entry_id (faktureringsmetod registration / direct)
* - invoice_payments.journal_entry_id (kontantmetod inbetalning / delbetalning)
* - supplier_invoices.registration_journal_entry_id / payment_journal_entry_id
* - supplier_invoice_payments.journal_entry_id (delbetalning)
*
* Every query is company-scoped (defense in depth alongside RLS). Results are
* deduplicated by id, so an invoice reachable via several paths appears once.
*/
export async function getJournalEntryUnderlagReferences(
supabase: SupabaseClient,
companyId: string,
journalEntryId: string,
): Promise<UnderlagReference[]> {
// --- Customer invoices ---------------------------------------------------
const invoices = new Map<string, string>()
// Direct link (faktureringsmetod registration, or invoices.journal_entry_id).
// Issued invoices only: a draft or cancelled invoice is no underlag, and the
// verifikat page counts these references as underlag (same verdict as the
// missing-underlag surfaces: NON_ISSUED_INVOICE_STATUSES).
const directInvoices = await fetchAllRows<InvoiceRow>(({ from, to }) =>
supabase.from('invoices').select('id, invoice_number')
.eq('company_id', companyId).eq('journal_entry_id', journalEntryId)
.not('status', 'in', NON_ISSUED_INVOICE_STATUSES_FILTER)
.order('id', { ascending: true }).range(from, to),
)
for (const inv of (directInvoices ?? []) as InvoiceRow[]) {
invoices.set(inv.id, inv.invoice_number)
}
// Payment rows (kontantmetod inbetalning, partial payments) → invoice_payments.
const paymentRows = await fetchAllRows<{ id: string; invoice_id: string | null }>(({ from, to }) =>
supabase.from('invoice_payments').select('id, invoice_id')
.eq('journal_entry_id', journalEntryId).order('id', { ascending: true }).range(from, to),
)
const paymentInvoiceIds = new Set<string>()
for (const row of (paymentRows ?? []) as { invoice_id: string | null }[]) {
if (row.invoice_id && !invoices.has(row.invoice_id)) paymentInvoiceIds.add(row.invoice_id)
}
if (paymentInvoiceIds.size > 0) {
const paidInvoices = await fetchAllRows<InvoiceRow>(({ from, to }) =>
supabase.from('invoices').select('id, invoice_number')
.eq('company_id', companyId).in('id', Array.from(paymentInvoiceIds))
.not('status', 'in', NON_ISSUED_INVOICE_STATUSES_FILTER)
.order('id', { ascending: true }).range(from, to),
)
for (const inv of (paidInvoices ?? []) as InvoiceRow[]) {
invoices.set(inv.id, inv.invoice_number)
}
}
// --- Supplier invoices ---------------------------------------------------
const supplierInvoices = new Map<string, { number: string; documentId?: string }>()
// Registration booking (accrual) on the invoice itself.
const registrationLinks = await fetchAllRows<SupplierInvoiceRow>(({ from, to }) =>
supabase.from('supplier_invoices').select(SUPPLIER_INVOICE_COLUMNS)
.eq('company_id', companyId).eq('registration_journal_entry_id', journalEntryId)
.order('id', { ascending: true }).range(from, to),
)
for (const si of (registrationLinks ?? []) as SupplierInvoiceRow[]) {
const documentId = anchoredDocumentId(si)
supplierInvoices.set(si.id, {
number: si.supplier_invoice_number,
...(documentId ? { documentId } : {}),
})
}
// Payment booking on the invoice itself.
const paymentLinks = await fetchAllRows<SupplierInvoiceRow>(({ from, to }) =>
supabase.from('supplier_invoices').select(SUPPLIER_INVOICE_COLUMNS)
.eq('company_id', companyId).eq('payment_journal_entry_id', journalEntryId)
.order('id', { ascending: true }).range(from, to),
)
for (const si of (paymentLinks ?? []) as SupplierInvoiceRow[]) {
const documentId = anchoredDocumentId(si)
supplierInvoices.set(si.id, {
number: si.supplier_invoice_number,
...(documentId ? { documentId } : {}),
})
}
// Partial-payment rows → supplier_invoice_payments.
const supplierPaymentRows = await fetchAllRows<{ id: string; supplier_invoice_id: string | null }>(
({ from, to }) => supabase.from('supplier_invoice_payments').select('id, supplier_invoice_id')
.eq('journal_entry_id', journalEntryId).order('id', { ascending: true }).range(from, to),
)
const supplierPaymentIds = new Set<string>()
for (const row of (supplierPaymentRows ?? []) as { supplier_invoice_id: string | null }[]) {
if (row.supplier_invoice_id && !supplierInvoices.has(row.supplier_invoice_id)) {
supplierPaymentIds.add(row.supplier_invoice_id)
}
}
if (supplierPaymentIds.size > 0) {
const paidSupplierInvoices = await fetchAllRows<SupplierInvoiceRow>(({ from, to }) =>
supabase.from('supplier_invoices').select(SUPPLIER_INVOICE_COLUMNS)
.eq('company_id', companyId).in('id', Array.from(supplierPaymentIds))
.order('id', { ascending: true }).range(from, to),
)
for (const si of (paidSupplierInvoices ?? []) as SupplierInvoiceRow[]) {
const documentId = anchoredDocumentId(si)
supplierInvoices.set(si.id, {
number: si.supplier_invoice_number,
...(documentId ? { documentId } : {}),
})
}
}
// --- Assemble ------------------------------------------------------------
const references: UnderlagReference[] = []
for (const [id, number] of invoices) references.push({ type: 'invoice', id, number })
for (const [id, supplierInvoice] of supplierInvoices) {
references.push({
type: 'supplier_invoice',
id,
number: supplierInvoice.number,
...(supplierInvoice.documentId ? { document_id: supplierInvoice.documentId } : {}),
})
}
return references
}
/**
* Batch form of the customer-invoice arm above, for the surfaces that decide
* "saknar underlag" for many verifikat at once: which register invoices point
* at each of the given journal entries, through the two links the register
* keeps (invoices.journal_entry_id for the registration booking,
* invoice_payments.journal_entry_id for a kontantmetod inbetalning, a
* delbetalning, or "matcha mot befintligt verifikat").
*
* An entry that appears in the result is backed by that invoice under BFL
* 5 kap 7 § (hänvisning till underlag): the invoice Accounted issued is the
* verifikation for the sale, and the payment row identifies the inbetalning.
* This is the TS mirror of the customer arm in the verifikat_without_documents
* / transactions_without_documents RPCs (migration 20260906135702, #2298):
* every TS surface (journal-list filter, documents/counts, transactions list)
* must reach the same verdict as the dashboard badge and the MCP tools.
*
* Values are invoice ids per journal entry id, direct link first and then
* payment rows in id order, deduplicated. Only entries with at least one link
* to an ISSUED invoice are present: a draft or cancelled invoice is no
* document, so it cannot back a verifikat (NON_ISSUED_INVOICE_STATUSES, the
* counterpart of the anchored-document requirement on the supplier arm).
* Every query is company-scoped (defense in depth alongside RLS).
*
* Callers pass at most one PostgREST `.in()` chunk (the ~150-id URL-length
* convention in lib/worklist/categories.ts). The two queries run in a fixed
* order (invoices, then invoice_payments) so queued test mocks stay simple.
*/
export async function getInvoiceReferencesForJournalEntries(
supabase: SupabaseClient,
companyId: string,
journalEntryIds: readonly string[],
): Promise<Map<string, string[]>> {
const result = new Map<string, string[]>()
if (journalEntryIds.length === 0) return result
const ids = [...journalEntryIds]
const add = (journalEntryId: string | null | undefined, invoiceId: string | null | undefined) => {
if (!journalEntryId || !invoiceId) return
const list = result.get(journalEntryId)
if (!list) result.set(journalEntryId, [invoiceId])
else if (!list.includes(invoiceId)) list.push(invoiceId)
}
const direct = await fetchAllRows<{ id: string; journal_entry_id: string | null }>(
({ from, to }) =>
supabase.from('invoices').select('id, journal_entry_id')
.eq('company_id', companyId).in('journal_entry_id', ids)
.not('status', 'in', NON_ISSUED_INVOICE_STATUSES_FILTER)
.order('id', { ascending: true }).range(from, to),
)
for (const row of direct) add(row.journal_entry_id, row.id)
// The invoice's status rides along as an inner embed so the filter drops
// payment rows of non-issued invoices server-side (one query, no id list).
const payments = await fetchAllRows<{
id: string
invoice_id: string | null
journal_entry_id: string | null
}>(({ from, to }) =>
supabase.from('invoice_payments').select('id, invoice_id, journal_entry_id, invoices!inner(status)')
.eq('company_id', companyId).in('journal_entry_id', ids)
.not('invoices.status', 'in', NON_ISSUED_INVOICE_STATUSES_FILTER)
.order('id', { ascending: true }).range(from, to),
)
for (const row of payments) add(row.journal_entry_id, row.invoice_id)
return result
}
/**
* Source types the invoice engine writes with `source_id` = the id of the
* register invoice (an `invoices` row; credit notes are rows there too) that
* the entry books: issuance and payment under faktureringsmetoden, the
* kontantmetod inbetalning, a credit note, and a reminder fee. For these the
* entry's own source columns are the link; no invoice-side row is needed.
* `rot_rut_payout` is deliberately absent: its source_id is the ROT/RUT
* request, not an invoice.
*/
export const INVOICE_SOURCED_ENTRY_TYPES: ReadonlySet<string> = new Set([
'invoice_created',
'invoice_paid',
'invoice_cash_payment',
'credit_note',
'reminder_fee',
])
/** Ids per PostgREST `.in()` filter (URL-length convention, lib/worklist/categories.ts). */
export const LINK_LOOKUP_CHUNK = 100
/** The columns {@link getInvoicesExplainingJournalEntries} reads off an entry. */
export interface ExplainableJournalEntry {
id: string
source_type: string | null
source_id: string | null
/** Storno: the entry this one cancels (reverseEntry / correctEntry). */
reverses_id?: string | null
/** Rättelse: the entry this one replaces (correctEntry). */
correction_of_id?: string | null
}
/** Literal select for the parent rows the chain walk fetches. */
const CHAIN_COLUMNS = 'id, source_type, source_id, reverses_id, correction_of_id'
/**
* Which register invoices explain each of the given journal entries, through
* every link the register keeps:
*
* 1. the engine's own entries: source_id IS the invoice id
* (INVOICE_SOURCED_ENTRY_TYPES);
* 2. the invoice side: invoices.journal_entry_id and
* invoice_payments.journal_entry_id (getInvoiceReferencesForJournalEntries);
* 3. the rättelse chain: a storno or correction carries reverses_id /
* correction_of_id and is explained by whatever explains the entry it
* cancels or replaces. Neither writer leaves a usable link of its own on
* the new entry (reverseEntry copies the original's polymorphic
* source_id, correctEntry copies nothing, and the register never points
* at a storno), so the chain is walked upwards until an attribution is
* found: the same links correctionChainDepth trusts, under the same
* MAX_CHAIN_WALK cap.
*
* A reader that followed only 1 and 2 kept a reversed original in its totals
* and dropped the storno that nets it (#2351): a makulerad EU sale stayed in
* the periodisk sammanställning while the account-based ruta 39 was zero.
*
* Values are invoice ids per entry id; an entry is present only when at least
* one invoice explains it. An engine entry is attributed by its source_id
* whether or not that invoice still exists (a missing one is a data defect
* for the caller to report, not a reason for silence), and a chain entry
* inherits its root's attribution the same way. An explicit link on the entry
* itself wins over an inherited one. Parents are fetched by id, company
* scoped and chunked, so the storno of a May invoice booked in June is
* resolved from June's entries alone.
*/
export async function getInvoicesExplainingJournalEntries(
supabase: SupabaseClient,
companyId: string,
entries: readonly ExplainableJournalEntry[],
): Promise<Map<string, string[]>> {
const result = new Map<string, string[]>()
if (entries.length === 0) return result
// Chain entries waiting for their parent's attribution, by parent id.
const waitingOn = new Map<string, string[]>()
// Links between an attributed entry and the root that explains it: 0 for
// an entry attributed by its own source_id or invoice-side link.
const depthOf = new Map<string, number>()
// Every id considered so far (the batch plus fetched parents): a cycle, or
// a parent shared by several children, is never fetched twice.
const seen = new Set<string>(entries.map((e) => e.id))
// Attribute an entry and every descendant waiting on it. Iterative, so a
// pathological in-batch chain cannot exhaust the stack, and with the depth
// carried along: a descendant more than MAX_CHAIN_WALK links below the root
// resolves to "no invoice", the cap the fetched walk below applies, so an
// in-batch chain and an out-of-batch chain of the same length agree.
const assign = (entryId: string, invoiceIds: string[], depth: number): void => {
const queue: [string, number][] = [[entryId, depth]]
for (let i = 0; i < queue.length; i++) {
const [id, d] = queue[i]
if (result.has(id)) continue
result.set(id, i === 0 ? invoiceIds : [...invoiceIds])
depthOf.set(id, d)
if (d >= MAX_CHAIN_WALK) continue
for (const child of waitingOn.get(id) ?? []) queue.push([child, d + 1])
}
}
let frontier: ExplainableJournalEntry[] = [...entries]
for (let hop = 0; frontier.length > 0; hop++) {
const unresolved: ExplainableJournalEntry[] = []
for (const entry of frontier) {
if (entry.source_id && INVOICE_SOURCED_ENTRY_TYPES.has(entry.source_type ?? '')) {
assign(entry.id, [entry.source_id], 0)
} else {
unresolved.push(entry)
}
}
for (const ids of chunk(unresolved.map((e) => e.id), LINK_LOOKUP_CHUNK)) {
const refs = await getInvoiceReferencesForJournalEntries(supabase, companyId, ids)
for (const [entryId, invoiceIds] of refs) assign(entryId, invoiceIds, 0)
}
const parentIds: string[] = []
for (const entry of unresolved) {
if (result.has(entry.id)) continue
const parentId = entry.correction_of_id ?? entry.reverses_id ?? null
if (!parentId) continue
const inherited = result.get(parentId)
if (inherited) {
const depth = (depthOf.get(parentId) ?? 0) + 1
if (depth <= MAX_CHAIN_WALK) assign(entry.id, [...inherited], depth)
continue
}
const waiting = waitingOn.get(parentId)
if (waiting) waiting.push(entry.id)
else waitingOn.set(parentId, [entry.id])
if (!seen.has(parentId)) {
seen.add(parentId)
parentIds.push(parentId)
}
}
if (parentIds.length === 0 || hop >= MAX_CHAIN_WALK) break
frontier = []
for (const ids of chunk(parentIds, LINK_LOOKUP_CHUNK)) {
const parents = await fetchAllRows<ExplainableJournalEntry>(({ from, to }) =>
supabase.from('journal_entries').select(CHAIN_COLUMNS)
.eq('company_id', companyId).in('id', ids)
.order('id', { ascending: true }).range(from, to),
)
frontier.push(...parents)
}
}
return result
}