Skip to content

HLL cross-language test: add union and updatable CL files #790

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
leerho merged 2 commits into from
Nov 3, 2026
Merged

HLL cross-language test: add union and updatable CL files #790

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8d7e13b
Add union and updatable HLL fixtures to the cross-language test
leerho Sep 10, 2026
b466491
HLL cross-language test: check curMin and numAtCurMin in union CL files
leerho Nov 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
113 changes: 113 additions & 0 deletions src/test/java/org/apache/datasketches/hll/HllSketchCrossLanguageTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
import static org.apache.datasketches.common.UtilityIO.CHECK_CL_BINARY_FILES;
import static org.apache.datasketches.common.UtilityIO.CHECK_CPP_FILES;
import static org.apache.datasketches.common.UtilityIO.CHECK_GO_FILES;
import static org.apache.datasketches.common.UtilityIO.CHECK_RUST_FILES;
import static org.apache.datasketches.common.UtilityIO.CHECK_JAVA_FILES;
import static org.apache.datasketches.common.UtilityIO.GENERATE_JAVA_FILES;
import static org.apache.datasketches.common.UtilityIO.LS;
Expand All @@ -30,7 +31,14 @@
import static org.apache.datasketches.hll.TgtHllType.HLL_4;
import static org.apache.datasketches.hll.TgtHllType.HLL_6;
import static org.apache.datasketches.hll.TgtHllType.HLL_8;
import static java.lang.foreign.ValueLayout.JAVA_INT_UNALIGNED;
import static org.apache.datasketches.hll.PreambleUtil.COMPACT_FLAG_MASK;
import static org.apache.datasketches.hll.PreambleUtil.CUR_MIN_COUNT_INT;
import static org.apache.datasketches.hll.PreambleUtil.FLAGS_BYTE;
import static org.apache.datasketches.hll.PreambleUtil.HLL_CUR_MIN_BYTE;
import static org.apache.datasketches.hll.PreambleUtil.REBUILD_CURMIN_NUM_KXQ_MASK;
import static org.testng.Assert.assertEquals;
import static org.testng.Assert.assertFalse;
import static org.testng.Assert.assertTrue;

import java.io.IOException;
Expand All @@ -46,6 +54,9 @@
*/
public class HllSketchCrossLanguageTest {

private static final int LG_K = HllSketch.DEFAULT_LG_K;
private static final int UNION_DISTINCT = 6000; //0..999 and 2000..6999; the SET-mode input overlaps

@Test(groups = {GENERATE_JAVA_FILES}, priority = 0)
public void generateBinariesForCompatibilityTesting() throws IOException {
final int[] nArr = {0, 1, 10, 100, 1000, 10_000, 100_000, 1_000_000};
Expand Down Expand Up @@ -136,6 +147,108 @@ private static void deserializeHll(final GroupLanguage lang) {
}
}

/**
* The cross-language (CL) files above are update sketches only, serialized compact, at a single lgK. That
* leaves two gaps a cross-language reader can fall through, both of which produced real bugs
* in 2026-09: the deferred KxQ rebuild is reachable only through a union gadget, and the
* flags-byte bit 32 collision is only visible in the flags byte itself. These union CL files
* close both. One lgK is deliberate; the defect classes are not lgK-dependent.
*/
@Test(groups = {GENERATE_JAVA_FILES}, priority = 0)
public void generateUnionBinariesForCompatibilityTesting() throws IOException {
for (final TgtHllType type: new TgtHllType[] {HLL_4, HLL_6, HLL_8}) {
final HllUnion union = mixedModeUnion();
final String tag = tagOf(type);
putBytesToJavaPath(tag + "_union_java.sk", union.getResult(type).toCompactByteArray());
putBytesToJavaPath(tag + "_unionupd_java.sk", union.getResult(type).toUpdatableByteArray());
}
}

@Test(groups = {CHECK_JAVA_FILES}, priority = 1)
public void checkUnionJava() { deserializeUnionResults(GroupLanguage.JAVA); }

@Test(groups = {CHECK_CPP_FILES})
public void checkUnionCpp() { deserializeUnionResults(GroupLanguage.CPP); }

@Test(groups = {CHECK_GO_FILES})
public void checkUnionGo() { deserializeUnionResults(GroupLanguage.GO); }

@Test(groups = {CHECK_RUST_FILES})
public void checkUnionRust() { deserializeUnionResults(GroupLanguage.RUST); }

/**
* A union whose inputs span all three curModes, so the coupon-update path into a gadget with
* a pending KxQ rebuild is exercised. The second sketch stays in SET mode and overlaps the
* first entirely, so the true distinct count is UNION_DISTINCT.
*/
private static HllUnion mixedModeUnion() {
final HllUnion union = new HllUnion(LG_K);
final HllSketch big = new HllSketch(LG_K, HLL_8);
for (int i = 0; i < 1000; i++) { big.update(i); }
final HllSketch small = new HllSketch(LG_K, HLL_8);
for (int i = 0; i < 20; i++) { small.update(500 + i); } //stays in SET mode, fully overlapping
final HllSketch other = new HllSketch(LG_K, HLL_8);
for (int i = 0; i < 5000; i++) { other.update(2000 + i); }
union.update(big);
union.update(small);
union.update(other);
return union;
}

private static String tagOf(final TgtHllType type) {
return type == HLL_4 ? "hll4" : type == HLL_6 ? "hll6" : "hll8";
}

/**
* One body, reused by every language. Beyond deserializing, this asserts the two flags-byte
* invariants the 2026-09 fixes established: bit 32 is reserved and must never be written by
* any implementation, and the compact flag must reflect the form actually produced. It also
* checks curMin and numAtCurMin, which the deferred KxQ rebuild updates, against Java's own
* result for the same target type and form.
*/
private static void deserializeUnionResults(final GroupLanguage lang) {
final TgtHllType[] types = {HLL_4, HLL_6, HLL_8};
final String[] forms = {"_union", "_unionupd"};
final HllUnion expectedUnion = mixedModeUnion();
double firstEstimate = Double.NaN;
for (final String form: forms) {
for (final TgtHllType type: types) {
final String fileName = tagOf(type) + form + lang.sfx + ".sk";
final byte[] bytes = getFileBytes(lang.pth, fileName);
if (bytes.length == 0) { continue; }
final int flags = bytes[FLAGS_BYTE] & 0xFF;

//bit 32 collided across implementations: REBUILD_CURMIN_NUM_KXQ here, FULL_SIZE in C++.
//Java clears it before serializing and C++ no longer writes it, so no image carries it.
assertFalse((flags & REBUILD_CURMIN_NUM_KXQ_MASK) > 0,
fileName + " has reserved flags bit 32 set");

//HLL_6 and HLL_8 used to omit this on toCompactByteArray()
assertEquals((flags & COMPACT_FLAG_MASK) > 0, form.equals("_union"),
fileName + " compact flag does not match its serialized form");

//curMin and numAtCurMin are rebuilt together with KxQ, so check them directly
final HllSketch expected = expectedUnion.getResult(type);
final byte[] expectedBytes =
form.equals("_union") ? expected.toCompactByteArray() : expected.toUpdatableByteArray();
assertEquals(bytes[HLL_CUR_MIN_BYTE] & 0xFF, expectedBytes[HLL_CUR_MIN_BYTE] & 0xFF,
fileName + " curMin");
assertEquals(MemorySegment.ofArray(bytes).get(JAVA_INT_UNALIGNED, CUR_MIN_COUNT_INT),
MemorySegment.ofArray(expectedBytes).get(JAVA_INT_UNALIGNED, CUR_MIN_COUNT_INT),
fileName + " numAtCurMin");

final HllSketch sketch = HllSketch.heapify(MemorySegment.ofArray(bytes));
assertEquals(sketch.getLgConfigK(), LG_K);
assertTrue(!sketch.isEmpty());
assertEquals(sketch.getEstimate(), UNION_DISTINCT, UNION_DISTINCT * 0.02);

@proost proost Nov 2, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For more robust compatibility test, how about adding comparing "curMin", "numAtCurMin"?

@leerho leerho Nov 2, 2026

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed, those are exactly the fields the deferred rebuild updates, so checking them is more direct than relying on the estimate. Added (b466491): each union CL file's curMin and numAtCurMin are compared with the values in Java's own result for the same target type and form. Only the Java CL files exist today; the C++/Go/Rust checks will start comparing once those languages write union CL files.

//an incorrect KxQ rebuild shifts the estimate, and it must not depend on target type
if (Double.isNaN(firstEstimate)) { firstEstimate = sketch.getEstimate(); }
else { assertEquals(sketch.getEstimate(), firstEstimate, 1e-9, fileName); }
}
}
}

static void println(Object o) {
//System.out.println(o.toString());
}
Expand Down
Loading