Typed Import Pipeline
Problem
CSV hands you strings, so qty arrives as "12" and one typo in a thousand-row upload should not cost you the whole file — but it must not silently become NaN either.
Solution
Give Spell a coercion schema and let parseTypedRecords aggregate every failure:
ts
import { s } from '@vielzeug/spell';
import { parseTypedRecords, TabulaValidationError } from '@vielzeug/tabula';
const importSchema = s.object({
id: s.string(),
qty: s.coerce.number(),
shipped: s.coerce.boolean(), // 'true'/'1' → true, 'false'/'0' → false
});
try {
const orders = parseTypedRecords(text, importSchema);
await ingest(orders); // qty is a number, shipped is a boolean
} catch (error) {
if (error instanceof TabulaValidationError) {
for (const row of error.errors) {
report(`line ${row.line}: ${row.issues?.map((i) => (i as { message: string }).message).join(', ')}`);
}
}
}Every row is validated even after another fails, so the user fixes the whole upload in one pass. line counts physical lines including the header, matching what a spreadsheet shows.
Pitfalls
- Schema validation is synchronous; an async schema throws a
TypeErrorbefore any row is produced. Do the async work on the parsed result, not in the schema. s.coerce.boolean()maps'true'/'1'and'false'/'0'deliberately — an arbitrary non-empty string like'yes'fails validation instead of becomingtrueby accident.- Empty cells fail required fields; if blank means zero, add
.optional().pipe(...)or a default to the schema rather than pre-processing rows.