/**
 * Incremental reader for the `files` member of a legacy journal document.
 *
 * WHY A HAND-WRITTEN SCANNER AND NOT `JSON.parse`
 * -----------------------------------------------
 * The manifest builder's tail step reads ONE journal shard and needs six
 * fields per row. `JSON.parse` cannot give it those without first
 * materialising the entire document twice over: once as a UTF-8 string the
 * size of the file, and again as a parsed object graph. On a 200,000-entry
 * shard (~34 MB on disk) that pair costs ~200 MB of resident growth before the
 * caller has looked at a single row — and RSS, unlike heap, does not come back
 * when the garbage collector frees it. The transient allocation IS the peak,
 * and the peak is the budget.
 *
 * Streaming the document in 256 KB chunks and parsing each row's small value
 * object on its own keeps the transient set bounded by one chunk plus one row,
 * so the resident cost becomes the compact index the caller retains and
 * nothing else. Measured on the 200,000-entry bench fixture: 260 MB → 182 MB
 * of resident growth, and (perhaps counter-intuitively) faster than
 * `JSON.parse`, because 200,000 small parses beat one 34 MB parse plus the
 * `Object.entries` array it takes to walk the result.
 *
 * WHAT THIS DELIBERATELY IS NOT
 * -----------------------------
 * It is not a general JSON parser and must never be used as one. It tracks
 * just enough structure — string/escape state and container depth — to find
 * the top-level `files` object and slice out each of its members verbatim.
 * Every value is then handed to the real `JSON.parse`, so number, escape, and
 * unicode semantics are V8's rather than this file's. Anything malformed
 * surfaces as a `SyntaxError` from that parse, which callers already treat as
 * a corrupt shard.
 *
 * On-disk format is untouched: this reads exactly the bytes the existing
 * writer produces.
 */
/**
 * Invoked once per member of the `files` object, in document order.
 *
 * `valueJson` is the member's raw JSON text. The callback owns the decision to
 * parse it — which is the point, since a caller keeping four fields out of ten
 * should never pay to build the other six.
 */
export type JournalLedgerRowVisitor = (key: string, valueJson: string) => void;
/**
 * Stream the `files` members of the journal document at `filePath`.
 *
 * Stops reading as soon as the `files` object closes: a shard's `pulls` tail
 * is not the ledger, and on a large journal declining to read it is free.
 *
 * Throws whatever `fs` throws for an unreadable file, and `SyntaxError` for a
 * row whose JSON is malformed. A document with no top-level `files` member
 * simply yields nothing — the same thing an empty ledger yields, which is the
 * correct reading of a locator file.
 */
export declare function streamJournalLedgerRows(filePath: string, onRow: JournalLedgerRowVisitor): void;
//# sourceMappingURL=journal-ledger-stream.d.ts.map