|
| 1 | +package io.github.dfa1.vortex.reader.decode; |
| 2 | + |
| 3 | +import io.github.dfa1.vortex.core.DType; |
| 4 | +import io.github.dfa1.vortex.core.PType; |
| 5 | +import io.github.dfa1.vortex.encoding.EncodingId; |
| 6 | +import io.github.dfa1.vortex.proto.ALPMetadata; |
| 7 | +import io.github.dfa1.vortex.proto.PatchesMetadata; |
| 8 | +import io.github.dfa1.vortex.reader.ReadRegistry; |
| 9 | +import io.github.dfa1.vortex.reader.array.DoubleArray; |
| 10 | +import io.github.dfa1.vortex.reader.array.FloatArray; |
| 11 | +import org.junit.jupiter.api.Test; |
| 12 | + |
| 13 | +import java.lang.foreign.Arena; |
| 14 | +import java.lang.foreign.MemorySegment; |
| 15 | +import java.nio.ByteBuffer; |
| 16 | +import java.nio.ByteOrder; |
| 17 | + |
| 18 | +import static org.assertj.core.api.Assertions.assertThat; |
| 19 | +import static org.assertj.core.api.Assertions.assertThatThrownBy; |
| 20 | +import static org.assertj.core.api.Assertions.within; |
| 21 | + |
| 22 | +class AlpEncodingDecoderTest { |
| 23 | + |
| 24 | + private static final AlpEncodingDecoder SUT = new AlpEncodingDecoder(); |
| 25 | + private static final ReadRegistry REGISTRY = TestRegistry.ofDecoders(SUT, new PrimitiveEncodingDecoder()); |
| 26 | + |
| 27 | + private static final DType F64 = new DType.Primitive(PType.F64, false); |
| 28 | + private static final DType F32 = new DType.Primitive(PType.F32, false); |
| 29 | + |
| 30 | + private static MemorySegment leLongs(long... vs) { |
| 31 | + byte[] b = new byte[vs.length * 8]; |
| 32 | + ByteBuffer bb = ByteBuffer.wrap(b).order(ByteOrder.LITTLE_ENDIAN); |
| 33 | + for (long v : vs) { |
| 34 | + bb.putLong(v); |
| 35 | + } |
| 36 | + return MemorySegment.ofArray(b); |
| 37 | + } |
| 38 | + |
| 39 | + private static MemorySegment leInts(int... vs) { |
| 40 | + byte[] b = new byte[vs.length * 4]; |
| 41 | + ByteBuffer bb = ByteBuffer.wrap(b).order(ByteOrder.LITTLE_ENDIAN); |
| 42 | + for (int v : vs) { |
| 43 | + bb.putInt(v); |
| 44 | + } |
| 45 | + return MemorySegment.ofArray(b); |
| 46 | + } |
| 47 | + |
| 48 | + private static MemorySegment leDoubles(double... vs) { |
| 49 | + byte[] b = new byte[vs.length * 8]; |
| 50 | + ByteBuffer bb = ByteBuffer.wrap(b).order(ByteOrder.LITTLE_ENDIAN); |
| 51 | + for (double v : vs) { |
| 52 | + bb.putDouble(v); |
| 53 | + } |
| 54 | + return MemorySegment.ofArray(b); |
| 55 | + } |
| 56 | + |
| 57 | + @Test |
| 58 | + void accepts_floatsTrue_otherFalse() { |
| 59 | + // Given / When / Then |
| 60 | + assertThat(SUT.accepts(F64)).isTrue(); |
| 61 | + assertThat(SUT.accepts(F32)).isTrue(); |
| 62 | + assertThat(SUT.accepts(new DType.Primitive(PType.I64, false))).isFalse(); |
| 63 | + assertThat(SUT.accepts(new DType.Utf8(false))).isFalse(); |
| 64 | + } |
| 65 | + |
| 66 | + @Test |
| 67 | + void decode_nonPrimitiveDtype_throws() { |
| 68 | + // Given a Utf8 dtype on an ALP node |
| 69 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, ByteBuffer.wrap(new ALPMetadata(0, 0, null).encode()), |
| 70 | + new ArrayNode[0], new int[0]); |
| 71 | + DecodeContext ctx = new DecodeContext(node, new DType.Utf8(false), 1, |
| 72 | + new MemorySegment[0], REGISTRY, Arena.global()); |
| 73 | + |
| 74 | + // When / Then |
| 75 | + assertThatThrownBy(() -> SUT.decode(ctx)).hasMessageContaining("expected primitive dtype"); |
| 76 | + } |
| 77 | + |
| 78 | + @Test |
| 79 | + void decode_missingMetadata_defaultsToZeroExponents() { |
| 80 | + // Given no metadata — decoder falls back to exp_e=0, exp_f=0 (scale 1.0) |
| 81 | + ArrayNode enc = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{0}); |
| 82 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, null, new ArrayNode[]{enc}, new int[0]); |
| 83 | + DecodeContext ctx = new DecodeContext(node, F64, 2, new MemorySegment[]{leLongs(5L, 7L)}, REGISTRY, Arena.global()); |
| 84 | + |
| 85 | + // When |
| 86 | + DoubleArray result = (DoubleArray) SUT.decode(ctx); |
| 87 | + |
| 88 | + // Then |
| 89 | + assertThat(result.getDouble(0)).isCloseTo(5.0, within(1e-9)); |
| 90 | + assertThat(result.getDouble(1)).isCloseTo(7.0, within(1e-9)); |
| 91 | + } |
| 92 | + |
| 93 | + @Test |
| 94 | + void decode_f64_broadcastNoPatches_returnsConstant() { |
| 95 | + // Given a single encoded value but 4 logical rows (capacity < n) and no patches: |
| 96 | + // the decoder broadcasts it into a constant array |
| 97 | + ArrayNode enc = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{0}); |
| 98 | + byte[] meta = new ALPMetadata(2, 0, null).encode(); // exp_e=2 -> *0.01 |
| 99 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, ByteBuffer.wrap(meta), new ArrayNode[]{enc}, new int[0]); |
| 100 | + DecodeContext ctx = new DecodeContext(node, F64, 4, new MemorySegment[]{leLongs(123L)}, REGISTRY, Arena.global()); |
| 101 | + |
| 102 | + // When |
| 103 | + DoubleArray result = (DoubleArray) SUT.decode(ctx); |
| 104 | + |
| 105 | + // Then |
| 106 | + assertThat(result.length()).isEqualTo(4); |
| 107 | + for (int i = 0; i < 4; i++) { |
| 108 | + assertThat(result.getDouble(i)).as("index %d", i).isCloseTo(1.23, within(1e-9)); |
| 109 | + } |
| 110 | + } |
| 111 | + |
| 112 | + @Test |
| 113 | + void decode_f32_broadcastNoPatches_returnsConstant() { |
| 114 | + // Given single value, 3 rows, no patches |
| 115 | + ArrayNode enc = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{0}); |
| 116 | + byte[] meta = new ALPMetadata(1, 0, null).encode(); // exp_e=1 -> *0.1 |
| 117 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, ByteBuffer.wrap(meta), new ArrayNode[]{enc}, new int[0]); |
| 118 | + DecodeContext ctx = new DecodeContext(node, F32, 3, new MemorySegment[]{leInts(25)}, REGISTRY, Arena.global()); |
| 119 | + |
| 120 | + // When |
| 121 | + FloatArray result = (FloatArray) SUT.decode(ctx); |
| 122 | + |
| 123 | + // Then |
| 124 | + assertThat(result.length()).isEqualTo(3); |
| 125 | + for (int i = 0; i < 3; i++) { |
| 126 | + assertThat(result.getFloat(i)).as("index %d", i).isCloseTo(2.5f, within(1e-6f)); |
| 127 | + } |
| 128 | + } |
| 129 | + |
| 130 | + @Test |
| 131 | + void decode_f64_patches_withU8Indices() { |
| 132 | + // Given patches whose index child uses U8 storage — exercises the U8 arm of |
| 133 | + // readUnsigned (the encoder always emits U32 indices) |
| 134 | + PatchesMetadata pm = new PatchesMetadata(1L, 0L, io.github.dfa1.vortex.proto.PType.U8, null, null, null); |
| 135 | + byte[] meta = new ALPMetadata(2, 0, pm).encode(); // *0.01 |
| 136 | + |
| 137 | + ArrayNode enc = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{0}); |
| 138 | + ArrayNode idx = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{1}); |
| 139 | + ArrayNode val = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{2}); |
| 140 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, ByteBuffer.wrap(meta), |
| 141 | + new ArrayNode[]{enc, idx, val}, new int[0]); |
| 142 | + |
| 143 | + MemorySegment idxSeg = MemorySegment.ofArray(new byte[]{1}); // patch row 1 |
| 144 | + MemorySegment[] segs = {leLongs(100L, 0L, 300L), idxSeg, leDoubles(9.0)}; |
| 145 | + DecodeContext ctx = new DecodeContext(node, F64, 3, segs, REGISTRY, Arena.global()); |
| 146 | + |
| 147 | + // When |
| 148 | + DoubleArray result = (DoubleArray) SUT.decode(ctx); |
| 149 | + |
| 150 | + // Then |
| 151 | + assertThat(result.getDouble(0)).isCloseTo(1.0, within(1e-9)); |
| 152 | + assertThat(result.getDouble(1)).isCloseTo(9.0, within(1e-9)); // patched |
| 153 | + assertThat(result.getDouble(2)).isCloseTo(3.0, within(1e-9)); |
| 154 | + } |
| 155 | + |
| 156 | + @Test |
| 157 | + void decode_patches_nonUnsignedIndexPtype_throws() { |
| 158 | + // Given a signed (I32) patch-index ptype — readUnsigned rejects it |
| 159 | + PatchesMetadata pm = new PatchesMetadata(1L, 0L, io.github.dfa1.vortex.proto.PType.I32, null, null, null); |
| 160 | + byte[] meta = new ALPMetadata(2, 0, pm).encode(); |
| 161 | + |
| 162 | + ArrayNode enc = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{0}); |
| 163 | + ArrayNode idx = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{1}); |
| 164 | + ArrayNode val = ArrayNode.of(EncodingId.VORTEX_PRIMITIVE, null, new ArrayNode[0], new int[]{2}); |
| 165 | + ArrayNode node = ArrayNode.of(EncodingId.VORTEX_ALP, ByteBuffer.wrap(meta), |
| 166 | + new ArrayNode[]{enc, idx, val}, new int[0]); |
| 167 | + |
| 168 | + MemorySegment[] segs = {leLongs(100L, 0L), leInts(1), leDoubles(9.0)}; |
| 169 | + DecodeContext ctx = new DecodeContext(node, F64, 2, segs, REGISTRY, Arena.global()); |
| 170 | + |
| 171 | + // When / Then |
| 172 | + assertThatThrownBy(() -> SUT.decode(ctx)).hasMessageContaining("non-unsigned patch index ptype"); |
| 173 | + } |
| 174 | +} |
0 commit comments