@gscdump/lakehouse 4.2.0 → 4.2.1

package/dist/partition-prune.d.mts

@@ -11,13 +11,9 @@ export type ManifestPartitionFilter = (partitions: IcebergFieldSummary[] | undef
 /**
  * One identity/dims value to prune an `identity`-transform partition field by.
  *
- * `'int32'`-encoded fields are deliberately NOT pruned here (mirrors the
- * original engine behavior): an INT column's bound bytes are a 4-byte int, but
- * per-team catalogs are single-tenant, so identity pruning on them saves ~nothing
- * — the per-file partition check in the resolver remains the authoritative
- * correctness filter. Only `'string'`-encoded identity fields are pruned by
- * lexicographic UTF-8 bound comparison (the truncated-string-stats case R2 SQL
- * needs the workaround for).
+ * Identity summaries use the declared field encoding: UTF-8 for strings and
+ * four-byte little-endian for INT. Unknown bounds keep the manifest. The
+ * per-file partition check remains the authoritative correctness filter.
  */
 export interface PartitionValueMatch {
   /** Partition field name as declared in the partition spec (e.g. `'site_id'`). */

package/dist/partition-prune.mjs

@@ -7,11 +7,12 @@ function decodeString(bytes) {
 	const u = toUint8(bytes);
 	return u == null ? null : UTF8_DECODER.decode(u);
 }
-function decodeMonthInt(bytes) {
+function decodeInt32(bytes) {
 	const u = toUint8(bytes);
-	if (u == null) return null;
+	if (u?.byteLength !== 4) return null;
 	return new DataView(u.buffer, u.byteOffset, u.byteLength).getInt32(0, true);
 }
+const decodeMonthInt = decodeInt32;
 function buildManifestPartitionFilter(partitionSpec, matches, wantedMonths) {
 	const fieldIndex = (name) => partitionSpec.findIndex((f) => f.name === name || f.sourceColumn === name);
 	const monthFieldIndex = partitionSpec.findIndex((f) => f.transform === "month");
@@ -23,6 +24,14 @@ function buildManifestPartitionFilter(partitionSpec, matches, wantedMonths) {
 			value: String(match.value)
 		}];
 	});
+	const intMatches = matches.flatMap((match) => {
+		if (match.encoding !== "int32" || typeof match.value !== "number" || !Number.isInteger(match.value) || match.value < -2147483648 || match.value > 2147483647) return [];
+		const index = fieldIndex(match.field);
+		return index < 0 || partitionSpec[index]?.transform !== "identity" ? [] : [{
+			index,
+			value: match.value
+		}];
+	});
 	const wantedMonthValues = wantedMonths ? [...wantedMonths].filter(Number.isFinite).sort((a, b) => a - b) : [];
 	const hasWantedMonthInRange = (lo, hi) => {
 		let left = 0;
@@ -43,6 +52,13 @@ function buildManifestPartitionFilter(partitionSpec, matches, wantedMonths) {
 			const hi = decodeString(summary.upper_bound);
 			if (lo != null && hi != null && (match.value < lo || match.value > hi)) return false;
 		}
+		for (const match of intMatches) {
+			const summary = partitions[match.index];
+			if (!summary || summary.lower_bound == null || summary.upper_bound == null) continue;
+			const lo = decodeInt32(summary.lower_bound);
+			const hi = decodeInt32(summary.upper_bound);
+			if (lo != null && hi != null && lo <= hi && (match.value < lo || match.value > hi)) return false;
+		}
 		if (wantedMonthValues.length > 0 && monthFieldIndex >= 0) {
 			const monthSummary = partitions[monthFieldIndex];
 			if (monthSummary && (monthSummary.lower_bound != null || monthSummary.upper_bound != null)) {

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@gscdump/lakehouse",
   "type": "module",
-  "version": "4.2.0",
+  "version": "4.2.1",
   "description": "Dataset-agnostic Iceberg lakehouse layer + dataset registry for R2 Data Catalog producers (ADR-0021).",
   "author": {
     "name": "Harlan Wilton",