|
| 1 | +package io.github.dfa1.vortex.writer.encode; |
| 2 | + |
| 3 | +import io.github.dfa1.vortex.core.DType; |
| 4 | +import io.github.dfa1.vortex.encoding.DTypes; |
| 5 | +import io.github.dfa1.vortex.reader.ReadRegistry; |
| 6 | +import io.github.dfa1.vortex.reader.array.Array; |
| 7 | +import io.github.dfa1.vortex.reader.array.DoubleArray; |
| 8 | +import io.github.dfa1.vortex.reader.array.FloatArray; |
| 9 | +import io.github.dfa1.vortex.reader.array.IntArray; |
| 10 | +import io.github.dfa1.vortex.reader.array.LongArray; |
| 11 | +import io.github.dfa1.vortex.reader.decode.AlpRdEncodingDecoder; |
| 12 | +import io.github.dfa1.vortex.reader.decode.BitpackedEncodingDecoder; |
| 13 | +import io.github.dfa1.vortex.reader.decode.DecodeContext; |
| 14 | +import io.github.dfa1.vortex.reader.decode.DeltaEncodingDecoder; |
| 15 | +import io.github.dfa1.vortex.reader.decode.EncodingDecoder; |
| 16 | +import io.github.dfa1.vortex.reader.decode.FrameOfReferenceEncodingDecoder; |
| 17 | +import io.github.dfa1.vortex.reader.decode.PrimitiveEncodingDecoder; |
| 18 | +import io.github.dfa1.vortex.reader.decode.TestRegistry; |
| 19 | +import io.github.dfa1.vortex.reader.decode.ZigZagEncodingDecoder; |
| 20 | +import org.junit.jupiter.params.ParameterizedTest; |
| 21 | +import org.junit.jupiter.params.provider.MethodSource; |
| 22 | + |
| 23 | +import java.util.Random; |
| 24 | +import java.util.stream.Stream; |
| 25 | + |
| 26 | +import static org.assertj.core.api.Assertions.assertThat; |
| 27 | + |
| 28 | +/// Property-based `decode(encode(x)) == x` round-trips for the lossless transform encodings. |
| 29 | +/// |
| 30 | +/// Seeded random inputs (plus curated edges: 0, ±1, MIN/MAX, near-extremes) flow through the |
| 31 | +/// real encoder → the writer's `EncodeResult` → the reader's decoder, asserting exact identity. |
| 32 | +/// Integer encodings (Delta, FrameOfReference, ZigZag) are checked for value identity including |
| 33 | +/// the wrap-around cases where the internal subtraction/shift overflows; AlpRd (the real-double |
| 34 | +/// bit-split codec) is checked **bit-exact** since it stores the full IEEE-754 content. |
| 35 | +class RoundTripPropertyTest { |
| 36 | + |
| 37 | + private static final DeltaEncodingEncoder DELTA = new DeltaEncodingEncoder(); |
| 38 | + private static final FrameOfReferenceEncodingEncoder FOR = new FrameOfReferenceEncodingEncoder(); |
| 39 | + private static final ZigZagEncodingEncoder ZIGZAG = new ZigZagEncodingEncoder(); |
| 40 | + private static final AlpRdEncodingEncoder ALPRD = new AlpRdEncodingEncoder(); |
| 41 | + |
| 42 | + private static final ReadRegistry REGISTRY = TestRegistry.ofDecoders( |
| 43 | + new DeltaEncodingDecoder(), new FrameOfReferenceEncodingDecoder(), |
| 44 | + new ZigZagEncodingDecoder(), new AlpRdEncodingDecoder(), |
| 45 | + new BitpackedEncodingDecoder(), new PrimitiveEncodingDecoder()); |
| 46 | + |
| 47 | + private static Array roundTrip(EncodingEncoder encoder, EncodingDecoder decoder, DType dtype, Object data, int n) { |
| 48 | + EncodeResult enc = encoder.encode(dtype, data, EncodeTestHelper.testCtx()); |
| 49 | + DecodeContext ctx = DecodeTestHelper.toDecodeContext(enc, n, dtype, REGISTRY); |
| 50 | + return decoder.decode(ctx); |
| 51 | + } |
| 52 | + |
| 53 | + // ── Delta ────────────────────────────────────────────────────────────────── |
| 54 | + |
| 55 | + @ParameterizedTest |
| 56 | + @MethodSource("longArrays") |
| 57 | + void delta_i64(long[] data) { |
| 58 | + assertLongs(roundTrip(DELTA, new DeltaEncodingDecoder(), DTypes.I64, data, data.length), data); |
| 59 | + } |
| 60 | + |
| 61 | + @ParameterizedTest |
| 62 | + @MethodSource("intArrays") |
| 63 | + void delta_i32(int[] data) { |
| 64 | + assertInts(roundTrip(DELTA, new DeltaEncodingDecoder(), DTypes.I32, data, data.length), data); |
| 65 | + } |
| 66 | + |
| 67 | + // ── FrameOfReference ─────────────────────────────────────────────────────── |
| 68 | + |
| 69 | + @ParameterizedTest |
| 70 | + @MethodSource("longArrays") |
| 71 | + void frameOfReference_i64(long[] data) { |
| 72 | + assertLongs(roundTrip(FOR, new FrameOfReferenceEncodingDecoder(), DTypes.I64, data, data.length), data); |
| 73 | + } |
| 74 | + |
| 75 | + @ParameterizedTest |
| 76 | + @MethodSource("intArrays") |
| 77 | + void frameOfReference_i32(int[] data) { |
| 78 | + assertInts(roundTrip(FOR, new FrameOfReferenceEncodingDecoder(), DTypes.I32, data, data.length), data); |
| 79 | + } |
| 80 | + |
| 81 | + // ── ZigZag ───────────────────────────────────────────────────────────────── |
| 82 | + |
| 83 | + @ParameterizedTest |
| 84 | + @MethodSource("longArrays") |
| 85 | + void zigzag_i64(long[] data) { |
| 86 | + assertLongs(roundTrip(ZIGZAG, new ZigZagEncodingDecoder(), DTypes.I64, data, data.length), data); |
| 87 | + } |
| 88 | + |
| 89 | + @ParameterizedTest |
| 90 | + @MethodSource("intArrays") |
| 91 | + void zigzag_i32(int[] data) { |
| 92 | + assertInts(roundTrip(ZIGZAG, new ZigZagEncodingDecoder(), DTypes.I32, data, data.length), data); |
| 93 | + } |
| 94 | + |
| 95 | + // ── AlpRd (bit-exact) ────────────────────────────────────────────────────── |
| 96 | + |
| 97 | + @ParameterizedTest |
| 98 | + @MethodSource("doubleArrays") |
| 99 | + void alpRd_f64_bitExact(double[] data) { |
| 100 | + DoubleArray r = (DoubleArray) roundTrip(ALPRD, new AlpRdEncodingDecoder(), DTypes.F64, data, data.length); |
| 101 | + for (int i = 0; i < data.length; i++) { |
| 102 | + assertThat(canon(r.getDouble(i))).as("index %d value %s", i, data[i]).isEqualTo(canon(data[i])); |
| 103 | + } |
| 104 | + } |
| 105 | + |
| 106 | + @ParameterizedTest |
| 107 | + @MethodSource("floatArrays") |
| 108 | + void alpRd_f32_bitExact(float[] data) { |
| 109 | + FloatArray r = (FloatArray) roundTrip(ALPRD, new AlpRdEncodingDecoder(), DTypes.F32, data, data.length); |
| 110 | + for (int i = 0; i < data.length; i++) { |
| 111 | + assertThat(canon(r.getFloat(i))).as("index %d value %s", i, data[i]).isEqualTo(canon(data[i])); |
| 112 | + } |
| 113 | + } |
| 114 | + |
| 115 | + // ── assertions ───────────────────────────────────────────────────────────── |
| 116 | + |
| 117 | + private static void assertLongs(Array result, long[] expected) { |
| 118 | + LongArray a = (LongArray) result; |
| 119 | + for (int i = 0; i < expected.length; i++) { |
| 120 | + assertThat(a.getLong(i)).as("index %d", i).isEqualTo(expected[i]); |
| 121 | + } |
| 122 | + } |
| 123 | + |
| 124 | + private static void assertInts(Array result, int[] expected) { |
| 125 | + IntArray a = (IntArray) result; |
| 126 | + for (int i = 0; i < expected.length; i++) { |
| 127 | + assertThat(a.getInt(i)).as("index %d", i).isEqualTo(expected[i]); |
| 128 | + } |
| 129 | + } |
| 130 | + |
| 131 | + private static long canon(double d) { |
| 132 | + return d == 0.0 ? 0L : Double.doubleToRawLongBits(d); // ±0 collapse, matches ALP family |
| 133 | + } |
| 134 | + |
| 135 | + private static int canon(float f) { |
| 136 | + return f == 0.0f ? 0 : Float.floatToRawIntBits(f); |
| 137 | + } |
| 138 | + |
| 139 | + // ── generators (seeded) ──────────────────────────────────────────────────── |
| 140 | + |
| 141 | + static Stream<long[]> longArrays() { |
| 142 | + Random rng = new Random(0xD17AL); |
| 143 | + Stream.Builder<long[]> b = Stream.builder(); |
| 144 | + b.add(new long[]{0L, 1L, -1L, Long.MIN_VALUE, Long.MAX_VALUE, Long.MIN_VALUE + 1, Long.MAX_VALUE - 1}); |
| 145 | + for (int n = 0; n < 50; n++) { |
| 146 | + int len = 1 + rng.nextInt(80); |
| 147 | + long[] a = new long[len]; |
| 148 | + for (int i = 0; i < len; i++) { |
| 149 | + a[i] = switch (rng.nextInt(3)) { |
| 150 | + case 0 -> rng.nextLong(); // full-range: forces wrap-around |
| 151 | + case 1 -> (long) (rng.nextInt(2001) - 1000); // small near-constant: clean delta/FoR |
| 152 | + default -> 1_000_000L + rng.nextInt(1000); // tight cluster around a reference |
| 153 | + }; |
| 154 | + } |
| 155 | + b.add(a); |
| 156 | + } |
| 157 | + return b.build(); |
| 158 | + } |
| 159 | + |
| 160 | + static Stream<int[]> intArrays() { |
| 161 | + Random rng = new Random(0xD17BL); |
| 162 | + Stream.Builder<int[]> b = Stream.builder(); |
| 163 | + b.add(new int[]{0, 1, -1, Integer.MIN_VALUE, Integer.MAX_VALUE, Integer.MIN_VALUE + 1, Integer.MAX_VALUE - 1}); |
| 164 | + for (int n = 0; n < 50; n++) { |
| 165 | + int len = 1 + rng.nextInt(80); |
| 166 | + int[] a = new int[len]; |
| 167 | + for (int i = 0; i < len; i++) { |
| 168 | + a[i] = switch (rng.nextInt(3)) { |
| 169 | + case 0 -> rng.nextInt(); |
| 170 | + case 1 -> rng.nextInt(2001) - 1000; |
| 171 | + default -> 1_000_000 + rng.nextInt(1000); |
| 172 | + }; |
| 173 | + } |
| 174 | + b.add(a); |
| 175 | + } |
| 176 | + return b.build(); |
| 177 | + } |
| 178 | + |
| 179 | + static Stream<double[]> doubleArrays() { |
| 180 | + Random rng = new Random(0xA1F64L); |
| 181 | + Stream.Builder<double[]> b = Stream.builder(); |
| 182 | + b.add(new double[]{0.0, -0.0, 1.0, -1.0, 0.5, -0.25, 3.14159, |
| 183 | + Double.MIN_VALUE, Double.MAX_VALUE, Double.MIN_NORMAL, 1e-300, 1e300, |
| 184 | + Double.NaN, Double.POSITIVE_INFINITY, Double.NEGATIVE_INFINITY}); |
| 185 | + for (int n = 0; n < 50; n++) { |
| 186 | + int len = 1 + rng.nextInt(80); |
| 187 | + double[] a = new double[len]; |
| 188 | + for (int i = 0; i < len; i++) { |
| 189 | + a[i] = rng.nextInt(2) == 0 |
| 190 | + ? Double.longBitsToDouble(rng.nextLong()) |
| 191 | + : rng.nextGaussian() * Math.pow(10, rng.nextInt(20) - 10); |
| 192 | + } |
| 193 | + b.add(a); |
| 194 | + } |
| 195 | + return b.build(); |
| 196 | + } |
| 197 | + |
| 198 | + static Stream<float[]> floatArrays() { |
| 199 | + Random rng = new Random(0xA1F32L); |
| 200 | + Stream.Builder<float[]> b = Stream.builder(); |
| 201 | + b.add(new float[]{0.0f, -0.0f, 1.0f, -1.0f, 0.5f, -0.25f, 3.14159f, |
| 202 | + Float.MIN_VALUE, Float.MAX_VALUE, Float.MIN_NORMAL, 1e-30f, 1e30f, |
| 203 | + Float.NaN, Float.POSITIVE_INFINITY, Float.NEGATIVE_INFINITY}); |
| 204 | + for (int n = 0; n < 50; n++) { |
| 205 | + int len = 1 + rng.nextInt(80); |
| 206 | + float[] a = new float[len]; |
| 207 | + for (int i = 0; i < len; i++) { |
| 208 | + a[i] = rng.nextInt(2) == 0 |
| 209 | + ? Float.intBitsToFloat(rng.nextInt()) |
| 210 | + : (float) (rng.nextGaussian() * Math.pow(10, rng.nextInt(12) - 6)); |
| 211 | + } |
| 212 | + b.add(a); |
| 213 | + } |
| 214 | + return b.build(); |
| 215 | + } |
| 216 | +} |
0 commit comments