test: cover new decode/validity branches; fix S3034 dict bit-set masking (Sonar gate)

reader/src/main/java/io/github/dfa1/vortex/reader/decode/DictEncodingDecoder.java

@@ -204,7 +204,7 @@ private static long readCode(MemorySegment codes, long i, PType codePType) {
     private static void setBit(MemorySegment bits, long i) {
         long byteIdx = i >>> 3;
         byte cur = bits.get(ValueLayout.JAVA_BYTE, byteIdx);
-        bits.set(ValueLayout.JAVA_BYTE, byteIdx, (byte) (cur | (1 << (i & 7))));
+        bits.set(ValueLayout.JAVA_BYTE, byteIdx, (byte) ((cur & 0xff) | (1 << (i & 7))));
     }
 
     private static Array decodeUtf8DictLegacy(DecodeContext ctx, MemorySegment meta) {

reader/src/main/java/io/github/dfa1/vortex/reader/decode/RleEncodingDecoder.java

@@ -167,7 +167,10 @@ private static short[] readShorts(MemorySegment buf, int count, PType ptype) {
         return out;
     }
 
-    private static double[] readDoubles(MemorySegment buf, int count) {
+    // Package-private (not private) so the broadcast branch (cap < count, a
+    // ConstantEncoding values pool) is testable directly: through decode() a
+    // constant child always materializes to `count` full elements, so cap == count.
+    static double[] readDoubles(MemorySegment buf, int count) {
         double[] out = new double[count];
         long cap = SegmentBroadcast.capacity(buf, 8);
         for (int i = 0; i < count; i++) {
@@ -176,7 +179,7 @@ private static double[] readDoubles(MemorySegment buf, int count) {
         return out;
     }
 
-    private static float[] readFloats(MemorySegment buf, int count) {
+    static float[] readFloats(MemorySegment buf, int count) {
         float[] out = new float[count];
         long cap = SegmentBroadcast.capacity(buf, 4);
         for (int i = 0; i < count; i++) {

reader/src/main/java/io/github/dfa1/vortex/reader/layout/DictLayoutDecoder.java

@@ -197,7 +197,7 @@ private static BoolArray gatherRowValidity(Array codes, BoolArray codesValidity,
             if (valid) {
                 long byteIdx = i >>> 3;
                 byte cur = bits.get(ValueLayout.JAVA_BYTE, byteIdx);
-                bits.set(ValueLayout.JAVA_BYTE, byteIdx, (byte) (cur | (1 << (i & 7))));
+                bits.set(ValueLayout.JAVA_BYTE, byteIdx, (byte) ((cur & 0xff) | (1 << (i & 7))));
             }
         }
         return new MaterializedBoolArray(DType.BOOL, n, bits.asReadOnly());

reader/src/test/java/io/github/dfa1/vortex/reader/ScanIteratorChunkGridTest.java

@@ -1,5 +1,6 @@
 package io.github.dfa1.vortex.reader;
 
+import io.github.dfa1.vortex.core.error.VortexException;
 import io.github.dfa1.vortex.core.model.ColumnName;
 import io.github.dfa1.vortex.core.model.LayoutId;
 import io.github.dfa1.vortex.reader.ScanIterator.ChunkSpec;
@@ -12,6 +13,7 @@
 import java.util.Map;
 
 import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
 
 /// Unit tests for [ScanIterator#buildChunks(Map)] — the scan planner that merges each column's
 /// chunk grid into one split grid (the Rust reference's `StructReader::register_splits` union of
@@ -120,6 +122,21 @@ void disjointBoundariesSpliceAtTheMergedGrid() {
         assertThat(coveringFlat(result.get(2), B)).isSameAs(coveringFlat(result.get(3), B));
     }
 
+    @Test
+    void columnShorterThanMergedGridThrows() {
+        // Given two columns whose totals disagree — A spans 8 rows [4, 4] but B only 6 [3, 3]. A
+        // well-formed file keeps every column the same length; a malformed/untrusted layout may not.
+        // The merged grid {0, 3, 4, 6, 8} then produces a window [6, 8) that lies beyond B's last
+        // chunk, so the covering-chunk advance loop runs B's cursor off the end. The guard must
+        // surface that as a VortexException rather than an ArrayIndexOutOfBoundsException.
+        var columnFlats = flats(A, new long[]{4, 4}, B, new long[]{3, 3});
+
+        // When / Then
+        assertThatThrownBy(() -> ScanIterator.buildChunks(columnFlats))
+                .isInstanceOf(VortexException.class)
+                .hasMessageContaining("no chunk covering");
+    }
+
     @Test
     void emptyProjectionYieldsNoChunks() {
         // Given no columns

reader/src/test/java/io/github/dfa1/vortex/reader/VortexReaderDecodeChunkTest.java

@@ -310,6 +310,46 @@ void scan_disjointMixedGrid_streamsExactValuesAcrossMergedWindows(@TempDir Path
         assertThat(streamedB).isEqualTo(MIXED_B_EXPECTED);
     }
 
+    // A NESTED struct column "s" (a vortex.struct under the root struct) with two I64 fields x, y,
+    // stored as ONE full-range subtree beside the per-chunk column "a" [2, 3, 2]. collectFlats treats
+    // the whole struct subtree as a single full-range chunk source, so the scan decodes it once as a
+    // StructArray and slices it per window — the ScanIterator.sliceArray StructArray branch that no
+    // primitive-only file reaches.
+    private static final long[] NESTED_X = {1000, 1001, 1002, 1003, 1004, 1005, 1006};
+    private static final long[] NESTED_Y = {2000, 2001, 2002, 2003, 2004, 2005, 2006};
+    private static final List<Long> NESTED_A_EXPECTED =
+            List.of(10L, 11L, 12L, 13L, 14L, 15L, 16L);
+
+    @Test
+    void scan_nestedStructColumn_slicesEachFieldPerWindow(@TempDir Path tmp) throws Exception {
+        // Given a struct whose second column "s" is itself a struct {x, y} spanning all 7 rows, beside
+        // a per-chunk column "a" chunked [2, 3, 2]. Neither the Java nor JNI writer emits a nested
+        // struct beside a per-chunk column in one file, so it is hand-assembled here.
+        Path file = writeNestedStructFile(tmp);
+        var streamedA = new ArrayList<Long>();
+        var streamedX = new ArrayList<Long>();
+        var streamedY = new ArrayList<Long>();
+
+        // When streaming the whole scan — "s" is decoded once over the full range then sliced into
+        // each merged-grid window {0,2,5,7} via ScanIterator.sliceArray's StructArray arm, which
+        // recursively slices each field into the window.
+        try (var reader = VortexReader.open(file, registry());
+                var iter = reader.scan(ScanOptions.all())) {
+            iter.forEachRemaining(chunk -> {
+                streamedA.addAll(values(chunk.column("a")));
+                io.github.dfa1.vortex.reader.array.StructArray s = chunk.column("s");
+                streamedX.addAll(fieldValues(s, 0));
+                streamedY.addAll(fieldValues(s, 1));
+            });
+        }
+
+        // Then every field value survives the per-window struct slice, including the middle window
+        // [2,5) sliced at a nonzero offset within the single full-range struct.
+        assertThat(streamedA).isEqualTo(NESTED_A_EXPECTED);
+        assertThat(streamedX).containsExactly(1000L, 1001L, 1002L, 1003L, 1004L, 1005L, 1006L);
+        assertThat(streamedY).containsExactly(2000L, 2001L, 2002L, 2003L, 2004L, 2005L, 2006L);
+    }
+
     // ── Helpers ────────────────────────────────────────────────────────────────
 
     private static ReadRegistry registry() {
@@ -330,6 +370,101 @@ private static List<Map<String, List<Long>>> streamAll(VortexReader reader) {
         return out;
     }
 
+    /// Reads field `fieldIdx` of a (possibly sliced) [StructArray] as a Long list.
+    private static List<Long> fieldValues(io.github.dfa1.vortex.reader.array.StructArray s, int fieldIdx) {
+        return values(s.field(fieldIdx));
+    }
+
+    /// Assembles a struct file: column "a" chunked over [CHUNK_ROWS] (one flat per chunk) and a nested
+    /// struct column "s" {x, y} stored as one full-range struct subtree (flat x, flat y). Segment
+    /// order: a-chunk0..a-chunkN, then x, then y. Layout: flat=0, chunked=1, struct=2; array spec
+    /// vortex.primitive=0.
+    private static Path writeNestedStructFile(Path dir) throws Exception {
+        int numChunks = CHUNK_ROWS.length;
+        List<byte[]> segments = new ArrayList<>();
+        for (int k = 0; k < numChunks; k++) {
+            segments.add(primitiveSegment(COL_A[k], null));
+        }
+        int xSegIdx = segments.size();
+        segments.add(primitiveSegment(NESTED_X, null));
+        int ySegIdx = segments.size();
+        segments.add(primitiveSegment(NESTED_Y, null));
+
+        long[] segOffsets = new long[segments.size()];
+        long[] segLengths = new long[segments.size()];
+        long off = 0;
+        for (int i = 0; i < segments.size(); i++) {
+            segOffsets[i] = off;
+            segLengths[i] = segments.get(i).length;
+            off += segments.get(i).length;
+        }
+
+        ByteBuffer footerBuf = MalformedFiles.buildFooter(
+                new String[]{"vortex.primitive", "vortex.bool"},
+                new String[]{"vortex.flat", "vortex.chunked", "vortex.struct"},
+                segOffsets, segLengths);
+        ByteBuffer dtypeBuf = buildNestedStructDtype();
+        ByteBuffer layoutBuf = buildNestedStructLayout(CHUNK_ROWS, xSegIdx, ySegIdx);
+
+        return writeVtxFile(dir, "nested.vtx", segments, footerBuf, dtypeBuf, layoutBuf);
+    }
+
+    /// Root struct DType {a: I64 non-null, s: struct {x: I64, y: I64} non-null}.
+    private static ByteBuffer buildNestedStructDtype() {
+        var fbb = new FbsBuilder(512);
+        int primA = FbsPrimitive.createFbsPrimitive(fbb, FbsPType.I64, false);
+        int dtA = FbsDType.createFbsDType(fbb, FbsType.FbsPrimitive, primA);
+
+        int primX = FbsPrimitive.createFbsPrimitive(fbb, FbsPType.I64, false);
+        int dtX = FbsDType.createFbsDType(fbb, FbsType.FbsPrimitive, primX);
+        int primY = FbsPrimitive.createFbsPrimitive(fbb, FbsPType.I64, false);
+        int dtY = FbsDType.createFbsDType(fbb, FbsType.FbsPrimitive, primY);
+        int innerNames = FbsStruct_.createNamesVector(fbb,
+                new int[]{fbb.createString("x"), fbb.createString("y")});
+        int innerDtypes = FbsStruct_.createDtypesVector(fbb, new int[]{dtX, dtY});
+        int innerStruct = FbsStruct_.createFbsStruct_(fbb, innerNames, innerDtypes, false);
+        int dtS = FbsDType.createFbsDType(fbb, FbsType.FbsStruct_, innerStruct);
+
+        int names = FbsStruct_.createNamesVector(fbb,
+                new int[]{fbb.createString("a"), fbb.createString("s")});
+        int dtypes = FbsStruct_.createDtypesVector(fbb, new int[]{dtA, dtS});
+        int rootStruct = FbsStruct_.createFbsStruct_(fbb, names, dtypes, false);
+        int dt = FbsDType.createFbsDType(fbb, FbsType.FbsStruct_, rootStruct);
+        FbsDType.finishFbsDTypeBuffer(fbb, dt);
+        return MalformedFiles.slice(fbb);
+    }
+
+    /// Layout: root struct(2) → [chunked "a" (flat per chunk), struct(2) "s" → [flat x, flat y]].
+    /// Both "s" fields are single full-range flats, so the whole subtree is one full-range chunk
+    /// source alongside "a"'s per-chunk grid.
+    private static ByteBuffer buildNestedStructLayout(long[] chunkRows, int xSegIdx, int ySegIdx) {
+        var fbb = new FbsBuilder(1024);
+        int numChunks = chunkRows.length;
+        long total = 0;
+        for (long r : chunkRows) {
+            total += r;
+        }
+        int[] flatOffs = new int[numChunks];
+        for (int k = 0; k < numChunks; k++) {
+            int segV = FbsLayout.createSegmentsVector(fbb, new long[]{k});
+            flatOffs[k] = FbsLayout.createFbsLayout(fbb, 0, chunkRows[k], 0, 0, segV);
+        }
+        int chunkedChildV = FbsLayout.createChildrenVector(fbb, flatOffs);
+        int chunkedA = FbsLayout.createFbsLayout(fbb, 1, total, 0, chunkedChildV, 0);
+
+        int xSegV = FbsLayout.createSegmentsVector(fbb, new long[]{xSegIdx});
+        int flatX = FbsLayout.createFbsLayout(fbb, 0, total, 0, 0, xSegV);
+        int ySegV = FbsLayout.createSegmentsVector(fbb, new long[]{ySegIdx});
+        int flatY = FbsLayout.createFbsLayout(fbb, 0, total, 0, 0, ySegV);
+        int sChildV = FbsLayout.createChildrenVector(fbb, new int[]{flatX, flatY});
+        int structS = FbsLayout.createFbsLayout(fbb, 2, total, 0, sChildV, 0);
+
+        int structChildV = FbsLayout.createChildrenVector(fbb, new int[]{chunkedA, structS});
+        int structOff = FbsLayout.createFbsLayout(fbb, 2, total, 0, structChildV, 0);
+        FbsLayout.finishFbsLayoutBuffer(fbb, structOff);
+        return MalformedFiles.slice(fbb);
+    }
+
     private static List<Long> values(Array arr) {
         List<Long> out = new ArrayList<>();
         if (arr instanceof MaskedArray masked) {

reader/src/test/java/io/github/dfa1/vortex/reader/array/LazyRleArrayTest.java

@@ -521,6 +521,28 @@ void indexedClampAndEmptyChunk() {
             assertThat(empty.getFloat(0)).isZero();
         }
 
+        @Test
+        void materializeConstantRunAndEmptyChunk() {
+            // Given a constant-run chunk (1 distinct value) and, separately, an empty chunk
+            // (0 distinct values): materialize's processChunk fast path must emit the lone
+            // value — or 0.0f for the empty pool — for every row without touching indices.
+            var constant = new LazyRleFloatArray(F32, 3, new float[]{6.25f}, new int[1024],
+                    new long[]{0L}, 0L, 1L, 1, 0);
+            var empty = new LazyRleFloatArray(F32, 2, new float[0], new int[1024],
+                    new long[]{0L}, 0L, 0L, 1, 0);
+
+            // When / Then
+            try (var arena = Arena.ofConfined()) {
+                var constantResult = constant.materialize(arena);
+                assertThat(constantResult.getAtIndex(VortexFormat.LE_FLOAT, 0)).isEqualTo(6.25f);
+                assertThat(constantResult.getAtIndex(VortexFormat.LE_FLOAT, 2)).isEqualTo(6.25f);
+
+                var emptyResult = empty.materialize(arena);
+                assertThat(emptyResult.getAtIndex(VortexFormat.LE_FLOAT, 0)).isEqualTo(0.0f);
+                assertThat(emptyResult.getAtIndex(VortexFormat.LE_FLOAT, 1)).isEqualTo(0.0f);
+            }
+        }
+
         @Test
         void materializeDecodesEveryRow() {
             // Given a mixed chunk; materialize must emit each logical row once, honoring

reader/src/test/java/io/github/dfa1/vortex/reader/decode/DictEncodingDecoderTest.java

@@ -508,6 +508,53 @@ void codeOutOfRangeWithPoolValidity_throwsVortexException() {
                     .hasMessageContaining("out of range for pool validity");
         }
 
+        @Test
+        void codesBroadcast_slowPathSetsValidityBits() {
+            // Given — a single-element codes buffer (codesCap == 1 < rowCount == 4, the
+            // ConstantEncoding fan-out shape) forces rowValidity's broadcast slow path
+            // (the `i % codesCap` loop and the setBit S3034 fix site). The lone code points at
+            // valid pool slot 0, so every row is valid and setBit runs on each iteration.
+            ArrayNode codesNode = primitiveNode(0);
+            ArrayNode valuesNode = maskedPrimitiveNode(1, 2);
+            MemorySegment[] segs = {
+                    u8Codes(0),                             // 1 code, broadcast across 4 rows
+                    TestSegments.leInts(10, 20),
+                    boolBitmap(true, false)                 // pool slot 0 valid, slot 1 invalid
+            };
+
+            // When
+            Array result = decodeDict(DType.I32, PType.U8, 2, 4, segs, codesNode, valuesNode);
+
+            // Then — the broadcast code resolves to valid slot 0 for every row
+            MaskedArray masked = assertMasked(result);
+            assertValidity(masked, true, true, true, true);
+            assertIntValues(masked, 10, 10, 10, 10);
+        }
+
+        @ParameterizedTest(name = "codes={0}")
+        @org.junit.jupiter.params.provider.EnumSource(value = PType.class, names = {"U16", "U32"})
+        void poolNull_readsWiderCodeWidths(PType codePType) {
+            // Given — codes at the wider widths (U16/U32) with a pool-null slot. rowValidity must
+            // read each code at the correct stride via readCode's U16/U32 arms (only exercised when
+            // pool validity is present), then mask rows whose code points at the dead slot. Here
+            // codesCap == rowCount, so the fast validity loop runs.
+            ArrayNode codesNode = primitiveNode(0);
+            ArrayNode valuesNode = maskedPrimitiveNode(1, 2);
+            MemorySegment[] segs = {
+                    codeSegment(codePType, new long[]{0, 1, 0, 1}),
+                    TestSegments.leInts(10, 20),
+                    boolBitmap(true, false)                 // slot 1 invalid
+            };
+
+            // When
+            Array result = decodeDict(DType.I32, codePType, 2, 4, segs, codesNode, valuesNode);
+
+            // Then — rows referencing the dead slot 1 are null; their expanded value is still present
+            MaskedArray masked = assertMasked(result);
+            assertValidity(masked, true, false, true, false);
+            assertIntValues(masked, 10, 20, 10, 20);
+        }
+
         private Array decodeDict(DType dtype, PType codePType, int valuesLen, long rowCount,
                 MemorySegment[] segs, ArrayNode codesNode, ArrayNode valuesNode) {
             MemorySegment meta = MemorySegment.ofArray(