Data you already have
Rows the page holds as objects go into the engine as they are — typed, replaceable, with the grid keeping its view.
An application often has its data as objects already — from its own backend, an IPC call, a store. It does not have to write CSV for the engine:
await provider.loadRows("people", rows, schemaJson);rows is an array of objects keyed by field name (or that array as JSON text). Values are in the
notation of a result: decimals as strings, dates as YYYY-MM-DD, null for a missing value. The
schema is required — ingest never guesses a type — and the engine reads the rows as it reads the
same data from a CSV file, so a query answers the same either way.
Replacing
Loading under a name that is already loaded replaces the source, all at once. A grid showing
it runs its query again and keeps its view: the sort, the filter, and the focus, which follows the
record by its row-key (and stays at its place where the record is gone).
What does not fit
A value that does not fit its field refuses the load, with the row and the field named
(JSON row 1, field id: …). Nothing is loaded, and the rows that were there stay.
Changing single rows is not part of this: a replace takes all rows.