import { createClient } from '@/lib/supabase/server' import { NextResponse } from 'next/server' import { requireCompanyId } from '@/lib/company/context' import { scopeTransactionsToAccount } from '@/lib/reconciliation/bank-reconciliation' const MAX_ROWS = 500 export async function GET(request: Request) { const supabase = await createClient() const { data: { user } } = await supabase.auth.getUser() if (!user) { return NextResponse.json({ error: 'Unauthorized' }, { status: 401 }) } const companyId = await requireCompanyId(supabase, user.id) const { searchParams } = new URL(request.url) const unmatched = searchParams.get('unmatched') === 'true' const reconciled = searchParams.get('reconciled') === 'true' const currency = searchParams.get('currency') || undefined // currency is interpolated into the PostgREST .or() filter below, so reject // anything that isn't a 3-letter ISO code. RLS still scopes results to the // company, but an unsanitized value could otherwise malform or widen the // filter (PostgREST filter injection). if (currency && !/^[A-Z]{3}$/.test(currency)) { return NextResponse.json({ error: 'Ogiltig valutakod' }, { status: 400 }) } const dateFrom = searchParams.get('date_from') || undefined const dateTo = searchParams.get('date_to') || undefined // When set, return only ignored rows — used by the reconciliation view to // surface a "Visa ignorerade" undo list. The default (no param) behaviour // continues to exclude ignored rows from unmatched results. const onlyIgnored = searchParams.get('only_ignored') === 'true' // account_number selects which cash account to scope to. We resolve it to a // cash_accounts.id (ledger_account is unique per company) and scope // transactions by that id, falling back to currency for legacy rows whose // cash_account_id hasn't been backfilled yet. This is what stops two // same-currency accounts from showing each other's transactions. const accountNumberParam = searchParams.get('account_number') || undefined let derivedCurrency = currency let cashAccountId: string | undefined // Only the primary account claims unassigned (NULL cash_account_id) rows, so // a secondary same-currency account's lists match its status card instead of // pooling the primary's unassigned rows. See scopeTransactionsToAccount. let includeUnassigned = true if (accountNumberParam) { const { data: cashAccount } = await supabase .from('cash_accounts') .select('id, currency, is_primary') .eq('company_id', companyId) .eq('ledger_account', accountNumberParam) .maybeSingle() if (cashAccount) { cashAccountId = cashAccount.id as string includeUnassigned = Boolean(cashAccount.is_primary) if (!derivedCurrency && cashAccount.currency) derivedCurrency = cashAccount.currency as string } } let query = supabase .from('transactions') .select('id, date, description, amount, currency, amount_sek, exchange_rate, reference, journal_entry_id, reconciliation_method, is_ignored') .eq('company_id', companyId) // unmatched and reconciled are mutually exclusive — unmatched wins if both set if (unmatched) { query = query.is('journal_entry_id', null) // Hide rows the user has explicitly suppressed from the reconciliation // view. Other callers (e.g. BookDirectlyDialog) also benefit — once // ignored, the row stops surfacing in the "to book" funnel everywhere. if (!onlyIgnored) query = query.eq('is_ignored', false) } else if (reconciled) { query = query.not('journal_entry_id', 'is', null) } if (onlyIgnored) query = query.eq('is_ignored', true) // Scope to the selected cash account. With a resolved id, match that account // OR legacy NULL rows of the same currency (so nothing disappears mid- // backfill). With only a currency (no account), filter by currency. With // neither (e.g. the company-wide only_ignored recovery list), no scope. // Shares one implementation with the reconciliation lib so the filter shape // can't drift between the status card and these lists. if (cashAccountId || derivedCurrency) { query = scopeTransactionsToAccount(query, cashAccountId, derivedCurrency ?? 'SEK', includeUnassigned) } if (dateFrom) query = query.gte('date', dateFrom) if (dateTo) query = query.lte('date', dateTo) // Fetch one extra row so we can tell the caller whether the result was truncated. query = query.order('date', { ascending: false }).limit(MAX_ROWS + 1) const { data, error } = await query if (error) { return NextResponse.json({ error: error.message }, { status: 500 }) } const rows = data || [] const hasMore = rows.length > MAX_ROWS const truncated = hasMore ? rows.slice(0, MAX_ROWS) : rows return NextResponse.json({ data: truncated, has_more: hasMore, limit: MAX_ROWS }) }