Use last lexigraphical key for parquet schema inference

engine/table/src/main/java/io/deephaven/engine/table/impl/locations/impl/KnownLocationKeyFinder.java

@@ -3,12 +3,11 @@
  */
 package io.deephaven.engine.table.impl.locations.impl;
 
-import io.deephaven.base.verify.Require;
 import io.deephaven.engine.table.impl.locations.ImmutableTableLocationKey;
 import org.jetbrains.annotations.NotNull;
 
 import java.util.Arrays;
-import java.util.Collections;
+import java.util.Comparator;
 import java.util.List;
 import java.util.function.Consumer;
 
@@ -18,17 +17,35 @@
 public final class KnownLocationKeyFinder<TLK extends ImmutableTableLocationKey>
         implements TableLocationKeyFinder<TLK> {
 
+    /**
+     * Creates a copy of the keys from {@code finder}. If {@code comparator} is not {@code null}, the keys will be
+     * sorted according to that {@code comparator}.
+     *
+     * @param finder the finder
+     * @param comparator the comparator
+     * @return the known location finder
+     * @param <TLK> the table location key type
+     */
+    public static <TLK extends ImmutableTableLocationKey> KnownLocationKeyFinder<TLK> copyFrom(
+            TableLocationKeyFinder<TLK> finder, Comparator<TLK> comparator) {
+        final RecordingLocationKeyFinder<TLK> recordingFinder = new RecordingLocationKeyFinder<>();
+        finder.findKeys(recordingFinder);
+        final List<TLK> mutableKeys = recordingFinder.getRecordedKeys();
+        if (comparator != null) {
+            mutableKeys.sort(comparator);
+        }
+        return new KnownLocationKeyFinder<>(mutableKeys);
+    }
+
     private final List<TLK> knownKeys;
 
     @SafeVarargs
     public KnownLocationKeyFinder(@NotNull final TLK... knownKeys) {
-        Require.elementsNeqNull(knownKeys, "knownKeys");
-        this.knownKeys = knownKeys.length == 0
-                ? Collections.emptyList()
-                : Collections.unmodifiableList(
-                        knownKeys.length == 1
-                                ? Collections.singletonList(knownKeys[0])
-                                : Arrays.asList(knownKeys));
+        this(Arrays.asList(knownKeys));
+    }
+
+    public KnownLocationKeyFinder(List<TLK> knownKeys) {
+        this.knownKeys = List.copyOf(knownKeys);
     }
 
     /**

extensions/parquet/table/src/main/java/io/deephaven/parquet/table/ParquetTools.java

@@ -687,10 +687,9 @@ public static Table readPartitionedTable(
     public static Table readPartitionedTableInferSchema(
             @NotNull final TableLocationKeyFinder<ParquetTableLocationKey> locationKeyFinder,
             @NotNull final ParquetInstructions readInstructions) {
-        final RecordingLocationKeyFinder<ParquetTableLocationKey> initialKeys = new RecordingLocationKeyFinder<>();
-        locationKeyFinder.findKeys(initialKeys);
-        final List<ParquetTableLocationKey> foundKeys = initialKeys.getRecordedKeys();
-        if (foundKeys.isEmpty()) {
+        final KnownLocationKeyFinder<ParquetTableLocationKey> sortedKeys =
+                KnownLocationKeyFinder.copyFrom(locationKeyFinder, Comparator.naturalOrder());
+        if (sortedKeys.getKnownKeys().isEmpty()) {
             if (readInstructions.isRefreshing()) {
                 throw new IllegalArgumentException(
                         "Unable to infer schema for a refreshing partitioned parquet table when there are no initial parquet files");
@@ -699,7 +698,7 @@ public static Table readPartitionedTableInferSchema(
         }
         // TODO (https://github.com/deephaven/deephaven-core/issues/877): Support schema merge when discovering multiple
         // parquet files
-        final ParquetTableLocationKey firstKey = foundKeys.get(0);
+        final ParquetTableLocationKey firstKey = sortedKeys.getKnownKeys().get(0);
         final Pair<List<ColumnDefinition<?>>, ParquetInstructions> schemaInfo = convertSchema(
                 firstKey.getFileReader().getSchema(),
                 firstKey.getMetadata().getFileMetaData().getKeyValueMetaData(),
@@ -723,7 +722,7 @@ public static Table readPartitionedTableInferSchema(
                     ColumnDefinition.ColumnType.Partitioning));
         }
         allColumns.addAll(schemaInfo.getFirst());
-        return readPartitionedTable(readInstructions.isRefreshing() ? locationKeyFinder : initialKeys,
+        return readPartitionedTable(readInstructions.isRefreshing() ? locationKeyFinder : sortedKeys,
                 schemaInfo.getSecond(), TableDefinition.of(allColumns));
     }