Files
accounted/scripts/fix-replacement-chars.ts
T
Jakob Wennberg ec27228a8e style: remove em/en dashes repo-wide, add CLAUDE.md rule against them (#890)
Em dashes (—) and en dashes (–) had spread across comments, docs, tests,
and a few UI strings, reading as AI-generated boilerplate rather than
house style. Replaced each with punctuation matching its context: colon
for explanatory clauses, comma for asides, plain hyphen for numeric/legal
ranges (e.g. "21-23§"), "to"/"till" for date ranges, parentheses for
paired-dash asides. messages/en.json and messages/sv.json were fixed by
hand together to keep sv/en in sync.

Left untouched where the dash is the functional subject rather than
decorative punctuation: date-range-parser.ts's separator regex,
charset-repair.ts's CP1252 byte-mapping table (and its test), the SIE
encoding mojibake docs, generic-csv.ts's minus-sign normalizer, the
agent system-prompt files that already instruct against em dashes, and
a golden iXBRL test fixture compared byte-for-byte.

Also fixes two bugs surfaced along the way: an off-by-one in
ApiKeysPanel's scope-label split (a leftover from an earlier partial
pass), and a charset-repair test that had lost the literal en-dash it
exists to verify.

Regenerated the agent atom seed migration (skills:generate) since 27
SKILL.md files changed. Added a CLAUDE.md rule against em/en dashes,
with an explicit carve-out for the functional-dash cases above.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-04 15:58:06 +02:00

353 lines
12 KiB
TypeScript

#!/usr/bin/env npx tsx
/**
* Find and repair U+FFFD (replacement character) corruption in user-facing
* text columns across the production database.
*
* Cause: prior to the fix in lib/import/sie-parser.ts:decodeBuffer, the SIE
* importer's encoding detector sampled only the first 4 KB of each file. When
* Swedish characters appeared only deeper in the file, the detector defaulted
* to UTF-8. Decoding a Windows-1252 byte like 0xD6 (Ö) as UTF-8 produces the
* replacement character U+FFFD: silently, because TextDecoder defaults to
* `fatal: false`. The mangled strings ended up in customers, suppliers,
* journal_entries, etc.
*
* Recovery is not deterministic: the original byte is lost. This script does
* heuristic substitution against a small Swedish-word dictionary:
*
* 1. Find rows where any text column contains U+FFFD.
* 2. For each U+FFFD-containing word, try substituting Å/Ä/Ö (uppercase
* context) or å/ä/ö (lowercase). Use the surrounding word casing to
* pick the case.
* 3. If exactly one substitution matches a known Swedish stem, apply it.
* 4. Otherwise log the row for manual review and leave it untouched.
*
* Idempotent: re-running matches zero rows once successful repairs are applied.
*
* Usage:
* # Preview every company
* npx tsx scripts/fix-replacement-chars.ts
*
* # Preview a single company
* npx tsx scripts/fix-replacement-chars.ts --company-id <uuid>
*
* # Apply
* npx tsx scripts/fix-replacement-chars.ts --commit
* npx tsx scripts/fix-replacement-chars.ts --company-id <uuid> --commit
*/
import { config } from 'dotenv'
config({ path: '.env.local' })
import { createClient, type SupabaseClient } from '@supabase/supabase-js'
import { recoverStringWithFFFD } from '../lib/import/shared/encoding'
function arg(name: string): string | undefined {
const i = process.argv.indexOf(`--${name}`)
return i >= 0 ? process.argv[i + 1] : undefined
}
const COMPANY_ID = arg('company-id') ?? null
const COMMIT = process.argv.includes('--commit')
const supabaseUrl = process.env.NEXT_PUBLIC_SUPABASE_URL
const serviceRoleKey = process.env.SUPABASE_SERVICE_ROLE_KEY
if (!supabaseUrl || !serviceRoleKey) {
console.error('Missing NEXT_PUBLIC_SUPABASE_URL or SUPABASE_SERVICE_ROLE_KEY in .env.local')
process.exit(1)
}
const supabase = createClient(supabaseUrl, serviceRoleKey) as SupabaseClient
const REPLACEMENT = '\uFFFD'
interface ColumnSpec {
table: string
/** Column for tenant filtering. Most have company_id; legacy tables may use user_id. */
tenantColumn: 'company_id' | 'user_id' | null
/** Free-text columns to scan for U+FFFD. */
columns: readonly string[]
/** Optional row filter: used to skip immutable/posted rows. */
filter?: (row: Record<string, unknown>) => boolean
/** Extra SELECTs needed by the filter: pulled but not scanned for U+FFFD. */
extraSelect?: readonly string[]
}
const TARGETS: readonly ColumnSpec[] = [
{
table: 'customers',
tenantColumn: 'company_id',
columns: ['name', 'address_line1', 'address_line2', 'city', 'country', 'notes'],
},
{
table: 'suppliers',
tenantColumn: 'company_id',
columns: ['name', 'address_line1', 'address_line2', 'city', 'country', 'category', 'notes'],
},
{
table: 'transactions',
tenantColumn: 'company_id',
columns: ['description', 'merchant_name', 'notes'],
},
{
table: 'chart_of_accounts',
tenantColumn: 'company_id',
columns: ['account_name', 'description'],
},
{
table: 'journal_entries',
tenantColumn: 'company_id',
columns: ['description'],
extraSelect: ['status'],
// Skip posted entries: they're legally immutable per BFL.
filter: (row) => row.status !== 'posted',
},
// Note: journal_entry_lines.line_description is intentionally NOT scanned.
// The table has no direct company_id column, and posted lines are immutable
// per the engine's enforcement triggers. If line descriptions need repair,
// run a targeted SQL query against drafts only.
{
table: 'voucher_gap_explanations',
tenantColumn: 'company_id',
columns: ['explanation'],
},
{
table: 'cost_centers',
tenantColumn: 'company_id',
columns: ['name'],
},
{
table: 'projects',
tenantColumn: 'company_id',
columns: ['name'],
},
{
table: 'receipts',
tenantColumn: 'company_id',
columns: ['merchant_name', 'representation_purpose'],
},
{
table: 'receipt_line_items',
tenantColumn: 'company_id',
columns: ['description'],
},
{
table: 'employees',
tenantColumn: 'company_id',
columns: ['first_name', 'last_name', 'address_line1', 'postal_code', 'city'],
},
{
table: 'invoices',
tenantColumn: 'company_id',
columns: ['your_reference', 'our_reference', 'notes', 'reverse_charge_text'],
},
{
table: 'invoice_items',
tenantColumn: 'company_id',
columns: ['description'],
},
{
table: 'supplier_invoices',
tenantColumn: 'company_id',
columns: ['notes'],
extraSelect: ['status'],
filter: (row) => !['paid', 'credited', 'reversed'].includes(String(row.status)),
},
{
table: 'supplier_invoice_items',
tenantColumn: 'company_id',
columns: ['description'],
},
{
table: 'categorization_templates',
tenantColumn: 'company_id',
columns: ['counterparty_name'],
},
{
table: 'companies',
tenantColumn: null, // top-level: filter directly on id when --company-id is given
columns: ['name', 'address_line1', 'address_line2', 'city'],
},
]
const PAGE_SIZE = 500
async function fetchAll(spec: ColumnSpec): Promise<Record<string, unknown>[]> {
const selectCols = ['id', ...spec.columns, ...(spec.extraSelect ?? [])]
if (spec.tenantColumn) selectCols.push(spec.tenantColumn)
const all: Record<string, unknown>[] = []
let from = 0
for (;;) {
let query = supabase
.from(spec.table)
.select(selectCols.join(','))
.order('id', { ascending: true })
.range(from, from + PAGE_SIZE - 1)
if (COMPANY_ID) {
if (spec.tenantColumn === 'company_id') {
query = query.eq('company_id', COMPANY_ID)
} else if (spec.table === 'companies') {
query = query.eq('id', COMPANY_ID)
}
}
const { data, error } = await query
if (error) throw new Error(`Failed to read ${spec.table}: ${error.message}`)
const rows = (data ?? []) as unknown as Record<string, unknown>[]
all.push(...rows)
if (rows.length < PAGE_SIZE) break
from += PAGE_SIZE
}
return all
}
interface Report {
table: string
scanned: number
affected: number
recovered: number
ambiguous: number
applied: number
failed: number
}
async function scanTable(spec: ColumnSpec): Promise<Report> {
const rows = await fetchAll(spec)
const filtered = spec.filter ? rows.filter(spec.filter) : rows
let affected = 0
let recovered = 0
let ambiguous = 0
let applied = 0
let failed = 0
for (const row of filtered) {
const updates: Record<string, string> = {}
const ambiguousFields: { field: string; value: string }[] = []
let rowHasFFFD = false
for (const col of spec.columns) {
const value = row[col]
if (typeof value !== 'string' || !value.includes(REPLACEMENT)) continue
rowHasFFFD = true
const fixed = recoverStringWithFFFD(value)
if (fixed !== null && fixed !== value) {
updates[col] = fixed
} else {
ambiguousFields.push({ field: col, value })
}
}
if (!rowHasFFFD) continue
affected++
const id = row.id as string
const tenantId = spec.tenantColumn ? row[spec.tenantColumn] : null
if (Object.keys(updates).length > 0) {
recovered++
console.log(`\n · ${spec.table}.${id}${tenantId ? ` (${spec.tenantColumn}=${tenantId})` : ''}`)
for (const [field, fixed] of Object.entries(updates)) {
const before = row[field] as string
console.log(` ${field}:`)
console.log(` before: ${JSON.stringify(before)}`)
console.log(` after : ${JSON.stringify(fixed)}`)
}
if (COMMIT) {
const { error } = await supabase.from(spec.table).update(updates).eq('id', id)
if (error) {
console.error(` FAILED: ${error.message}`)
failed++
} else {
applied++
}
}
}
if (ambiguousFields.length > 0) {
ambiguous++
console.log(
`\n · ${spec.table}.${id}${tenantId ? ` (${spec.tenantColumn}=${tenantId})` : ''}: AMBIGUOUS (manual review):`
)
for (const { field, value } of ambiguousFields) {
console.log(` ${field}: ${JSON.stringify(value)}`)
}
}
}
return {
table: spec.table,
scanned: filtered.length,
affected,
recovered,
ambiguous,
applied,
failed,
}
}
async function main() {
console.log('─────────────────────────────────────────────────────────')
console.log('U+FFFD repair across user-facing text columns')
console.log('─────────────────────────────────────────────────────────')
console.log('Supabase URL :', supabaseUrl)
console.log('Company :', COMPANY_ID ?? '(all)')
console.log('Mode :', COMMIT ? 'COMMIT (writes)' : 'DRY RUN (no writes)')
console.log('─────────────────────────────────────────────────────────\n')
const reports: Report[] = []
for (const spec of TARGETS) {
console.log(`[${spec.table}]`)
try {
const report = await scanTable(spec)
reports.push(report)
console.log(
` · Scanned ${report.scanned} rows; ${report.affected} affected (${report.recovered} recovered, ${report.ambiguous} ambiguous)`
)
} catch (err) {
console.error(` · FAILED: ${err instanceof Error ? err.message : err}`)
}
}
console.log('\n─────────────────────────────────────────────────────────')
console.log('Summary')
console.log('─────────────────────────────────────────────────────────')
const totals = reports.reduce(
(acc, r) => ({
scanned: acc.scanned + r.scanned,
affected: acc.affected + r.affected,
recovered: acc.recovered + r.recovered,
ambiguous: acc.ambiguous + r.ambiguous,
applied: acc.applied + r.applied,
failed: acc.failed + r.failed,
}),
{ scanned: 0, affected: 0, recovered: 0, ambiguous: 0, applied: 0, failed: 0 }
)
for (const r of reports) {
if (r.affected === 0) continue
console.log(
`${r.table.padEnd(28)}: ${r.affected.toString().padStart(4)} affected, ${r.recovered.toString().padStart(4)} recovered, ${r.ambiguous.toString().padStart(4)} ambiguous`
)
}
console.log('─────────────────────────────────────────────────────────')
console.log(
`TOTAL : ${totals.affected.toString().padStart(4)} affected, ${totals.recovered.toString().padStart(4)} recovered, ${totals.ambiguous.toString().padStart(4)} ambiguous`
)
if (COMMIT) {
console.log(`Applied : ${totals.applied}`)
console.log(`Failed : ${totals.failed}`)
} else {
console.log('\nDry run: no changes written. Re-run with --commit to apply.')
console.log(
'Ambiguous rows are NOT touched: review them by hand and update manually if needed.'
)
}
}
main().catch((err) => {
console.error('\nFATAL:', err instanceof Error ? err.message : err)
process.exit(1)
})