Change verification process, so ToDenseVector is not used but direct literal translation is done

carlosdelest · carlosdelest · commit 1be036448ebd · 2025-09-03T18:34:18.000+02:00
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/knn-function.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/knn-function.csv-spec
@@ -308,7 +308,7 @@ required_capability: knn_function_v3
 required_capability: to_dense_vector_function
 
 from colors metadata _score
-| eval query = [0, 120, 0]
+| eval query = to_dense_vector([0, 120, 0])
 | where knn(rgb_vector, query, 10) 
 | sort _score desc, color asc
 | keep color, rgb_vector
@@ -333,7 +333,6 @@ required_capability: knn_function_v3
 required_capability: to_dense_vector_function
 
 from colors metadata _score
-| eval query = [0, 120, 0]
 | where knn(rgb_vector, "007800", 10) 
 | sort _score desc, color asc
 | keep color, rgb_vector
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-cosine-similarity.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-cosine-similarity.csv-spec
@@ -94,12 +94,12 @@ similarityWithRow
 required_capability: cosine_vector_similarity_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3]) 
 | eval similarity = round(v_cosine(vector, [0, 1, 2]), 3) 
 ;
 
-vector: integer      | similarity:double
-[1, 2, 3]      | 0.978  
+vector: dense_vector | similarity:double
+[1.0, 2.0, 3.0]      | 0.978  
 ;
 
 similarityWithVectorField
@@ -108,7 +108,7 @@ required_capability: to_dense_vector_function
 
 from colors
 | where color != "black" 
-| eval query = [0, 255, 255]
+| eval query = to_dense_vector([0, 255, 255])
 | eval similarity = v_cosine(rgb_vector, query) 
 | sort similarity desc, color asc 
 | limit 10
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-dot-product.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-dot-product.csv-spec
@@ -92,20 +92,20 @@ similarityWithRow
 required_capability: dot_product_vector_similarity_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3])
 | eval similarity = round(v_dot_product(vector, [0, 1, 2]), 3) 
 ;
 
-vector: integer | similarity:double
-[1, 2, 3]       | 4.5
+vector: dense_vector | similarity:double
+[1.0, 2.0, 3.0]      | 4.5
 ;
 
 similarityWithVectorField
 required_capability: dot_product_vector_similarity_function
 required_capability: to_dense_vector_function
 
 from colors
-| eval query = [0, 255, 255]
+| eval query = to_dense_vector([0, 255, 255])
 | eval similarity = v_dot_product(rgb_vector, query) 
 | sort similarity desc, color asc 
 | limit 10
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-hamming.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-hamming.csv-spec
@@ -91,20 +91,20 @@ similarityWithRow
 required_capability: hamming_vector_similarity_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3])
 | eval similarity = round(v_hamming(vector, [0, 1, 2]), 3) 
 ;
 
-vector: integer  | similarity:double
-[1, 2, 3]        | 4.0
+vector: dense_vector | similarity:double
+[1.0, 2.0, 3.0]      | 4.0
 ;
 
 similarityWithVectorField
 required_capability: hamming_vector_similarity_function
 required_capability: to_dense_vector_function
 
 from colors
-| eval query = [0, 255, 255]
+| eval query = to_dense_vector([0, 255, 255])
 | eval similarity = v_hamming(rgb_vector, query) 
 | sort similarity desc, color asc 
 | limit 10
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-l1-norm.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-l1-norm.csv-spec
@@ -91,20 +91,20 @@ similarityWithRow
 required_capability: l1_norm_vector_similarity_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3]) 
 | eval similarity = round(v_l1_norm(vector, [0, 1, 2]), 3) 
 ;
 
-vector: integer  | similarity:double
-[1, 2, 3]        | 3.0
+vector: dense_vector | similarity:double
+[1.0, 2.0, 3.0]      | 3.0
 ;
 
 similarityWithVectorField
 required_capability: l1_norm_vector_similarity_function
 required_capability: to_dense_vector_function
 
 from colors
-| eval query = [0, 255, 255]
+| eval query = to_dense_vector([0, 255, 255])
 | eval similarity = v_l1_norm(rgb_vector, query) 
 | sort similarity desc, color asc 
 | limit 10
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-l2-norm.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-l2-norm.csv-spec
@@ -91,20 +91,20 @@ similarityWithRow
 required_capability: l2_norm_vector_similarity_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3]) 
 | eval similarity = round(v_l2_norm(vector, [0, 1, 2]), 3) 
 ;
 
-vector: integer  | similarity:double
-[1, 2, 3]        | 1.732
+vector: dense_vector | similarity:double
+[1.0, 2.0, 3.0]      | 1.732
 ;
 
 similarityWithVectorField
 required_capability: l2_norm_vector_similarity_function
 required_capability: to_dense_vector_function
 
 from colors
-| eval query = [0, 255, 255]
+| eval query = to_dense_vector([0, 255, 255])
 | eval similarity = v_l2_norm(rgb_vector, query) 
 | sort similarity desc, color asc 
 | limit 10
diff --git a/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-magnitude.csv-spec b/x-pack/plugin/esql/qa/testFixtures/src/main/resources/vector-magnitude.csv-spec
@@ -90,10 +90,10 @@ magnitudeWithRow
 required_capability: magnitude_scalar_vector_function
 required_capability: to_dense_vector_function
  
-row vector = [1, 2, 3] 
+row vector = to_dense_vector([1, 2, 3]) 
 | eval magnitude = round(v_magnitude(vector), 3) 
 ;
 
-vector: integer      | magnitude:double
-[1, 2, 3]      | 3.742
+vector: dense_vector | magnitude:double
+[1.0, 2.0, 3.0]      | 3.742
 ;
diff --git a/x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/plugin/KnnFunctionIT.java b/x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/plugin/KnnFunctionIT.java
@@ -191,6 +191,23 @@ public void testKnnWithLookupJoin() {
         );
     }
 
+
+    public void testKnnIncorrectCasting() {
+        var query = String.format(Locale.ROOT, """
+            FROM test
+            | WHERE KNN(vector, "notcorrect", 5)
+            """);
+
+        var error = expectThrows(VerificationException.class, () -> run(query));
+        assertThat(
+            error.getMessage(),
+            containsString(
+                "line 3:13: [KNN] function cannot operate on [lookup_vector], supplied by an index [test_lookup] in non-STANDARD "
+                    + "mode [lookup]"
+            )
+        );
+    }
+
     @Before
     public void setup() throws IOException {
         assumeTrue("Needs KNN support", EsqlCapabilities.Cap.KNN_FUNCTION_V3.isEnabled());
diff --git a/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/analysis/Analyzer.java b/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/analysis/Analyzer.java
@@ -77,7 +77,6 @@
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.FromAggregateMetricDouble;
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToAggregateMetricDouble;
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToDateNanos;
-import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToDenseVector;
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToDouble;
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToInteger;
 import org.elasticsearch.xpack.esql.expression.function.scalar.convert.ToLong;
@@ -1686,21 +1685,28 @@ private static Expression processVectorFunction(org.elasticsearch.xpack.esql.cor
                                 floatVector = ((List<Number>) folded).stream().map(Number::floatValue).collect(Collectors.toList());
                             }
                         } else if (folded instanceof BytesRef hexString && arg.dataType() == KEYWORD) {
-                            byte[] bytes = HexFormat.of().parseHex(hexString.utf8ToString());
-                            floatVector = new ArrayList<>();
-                            for (byte value : bytes) {
-                                floatVector.add((float) value);
+                            try {
+                                byte[] bytes = HexFormat.of().parseHex(hexString.utf8ToString());
+                                floatVector = new ArrayList<>();
+                                for (byte value : bytes) {
+                                    floatVector.add((float) value);
+                                }
+                            } catch (IllegalArgumentException e) {
+                                throw new VerificationException(
+                                    "Error in ["
+                                        + vectorFunction.sourceText()
+                                        + "] for argument ["
+                                        + arg.sourceText()
+                                        + "]; dense_vectors must be a hex-encoded string: "
+                                        + e.getMessage()
+                                );
                             }
                         }
                         if (floatVector != null) {
                             Literal denseVector = new Literal(arg.source(), floatVector, DataType.DENSE_VECTOR);
                             newArgs.add(denseVector);
                             continue;
                         }
-                    } else if ((arg instanceof ToDenseVector == false) && (arg.dataType().isNumeric() || arg.dataType() == KEYWORD)) {
-                        // add casting function if it's not already there
-                        newArgs.add(new ToDenseVector(arg.source(), arg));
-                        continue;
                     }
                 }
                 newArgs.add(arg);
diff --git a/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/scalar/convert/ToDenseVector.java b/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/scalar/convert/ToDenseVector.java
@@ -7,29 +7,37 @@
 
 package org.elasticsearch.xpack.esql.expression.function.scalar.convert;
 
+import org.apache.lucene.util.BytesRef;
 import org.elasticsearch.common.io.stream.NamedWriteableRegistry;
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.compute.ann.ConvertEvaluator;
+import org.elasticsearch.xpack.esql.capabilities.PostAnalysisPlanVerificationAware;
 import org.elasticsearch.xpack.esql.capabilities.PostAnalysisVerificationAware;
+import org.elasticsearch.xpack.esql.capabilities.PostOptimizationVerificationAware;
+import org.elasticsearch.xpack.esql.common.Failure;
+import org.elasticsearch.xpack.esql.common.Failures;
 import org.elasticsearch.xpack.esql.core.expression.Expression;
+import org.elasticsearch.xpack.esql.core.expression.FoldContext;
 import org.elasticsearch.xpack.esql.core.tree.NodeInfo;
 import org.elasticsearch.xpack.esql.core.tree.Source;
 import org.elasticsearch.xpack.esql.core.type.DataType;
 import org.elasticsearch.xpack.esql.expression.function.Example;
 import org.elasticsearch.xpack.esql.expression.function.FunctionInfo;
 import org.elasticsearch.xpack.esql.expression.function.Param;
+import org.elasticsearch.xpack.esql.plan.logical.LogicalPlan;
 
 import java.io.IOException;
 import java.util.List;
 import java.util.Map;
+import java.util.function.BiConsumer;
 
 import static org.elasticsearch.xpack.esql.core.type.DataType.DENSE_VECTOR;
 import static org.elasticsearch.xpack.esql.core.type.DataType.DOUBLE;
 import static org.elasticsearch.xpack.esql.core.type.DataType.INTEGER;
 import static org.elasticsearch.xpack.esql.core.type.DataType.KEYWORD;
 import static org.elasticsearch.xpack.esql.core.type.DataType.LONG;
 
-public class ToDenseVector extends AbstractConvertFunction implements PostAnalysisVerificationAware {
+public class ToDenseVector extends AbstractConvertFunction implements PostAnalysisPlanVerificationAware {
     public static final NamedWriteableRegistry.Entry ENTRY = new NamedWriteableRegistry.Entry(
         Expression.class,
         "ToDenseVector",
@@ -60,6 +68,40 @@ public ToDenseVector(
         super(source, field);
     }
 
+    @Override
+    public BiConsumer<LogicalPlan, Failures> postAnalysisPlanVerification() {
+        return (lp, failures) -> {
+            Expression arg = children().get(0);
+            if (arg.foldable()) {
+                Object fold = arg.fold(FoldContext.small());
+                if ((fold instanceof List<?> list) && arg.dataType().isNumeric()) {
+                    if (list.size() <= 1) {
+                        failures.add(Failure.fail(
+                            this,
+                            "[" + sourceText() + "] requires at least two values to convert to a dense_vector"
+                        ));
+                    }
+                    return;
+                }
+                if ((arg.dataType() == KEYWORD) && fold instanceof BytesRef bytesRef) {
+                    if (bytesRef.length == 0) {
+                        failures.add(Failure.fail(
+                            this,
+                            "["
+                                + sourceText()
+                                + "] must be a non-empty hexadecimal string"));
+                    }
+                    return;
+                }
+                failures.add(Failure.fail(
+                    this,
+                    "["
+                        + sourceText()
+                        + "] must be a multi-valued input of numbers or an hexadecimal string"));
+            }
+        };
+    }
+
     private ToDenseVector(StreamInput in) throws IOException {
         super(in);
     }
diff --git a/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/scalar/convert/ToDenseVectorFromStringEvaluator.java b/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/scalar/convert/ToDenseVectorFromStringEvaluator.java
@@ -16,6 +16,7 @@
 import org.elasticsearch.compute.operator.DriverContext;
 import org.elasticsearch.compute.operator.EvalOperator;
 import org.elasticsearch.core.Releasables;
+import org.elasticsearch.xpack.esql.core.InvalidArgumentException;
 import org.elasticsearch.xpack.esql.core.tree.Source;
 
 import java.util.HexFormat;
@@ -54,24 +55,29 @@ public Block evalBlock(Block b) {
                     builder.appendNull();
                 } else {
                     scratch = block.getBytesRef(p, scratch);
-                    byte[] bytes = HexFormat.of().parseHex(scratch.utf8ToString());
-                    if (bytes.length == 0) {
-                        builder.appendNull();
-                        continue;
-                    }
-                    if (dimensions == 0) {
-                        dimensions = bytes.length;
-                    } else {
-                        if (bytes.length != dimensions) {
-                            throw new IllegalArgumentException("All dense_vector must have the same number of dimensions. Expected: "
-                                + dimensions + ", found: " + bytes.length);
+                    try {
+                        byte[] bytes = HexFormat.of().parseHex(scratch.utf8ToString());
+                        if (bytes.length == 0) {
+                            builder.appendNull();
+                            continue;
                         }
+                        if (dimensions == 0) {
+                            dimensions = bytes.length;
+                        } else {
+                            if (bytes.length != dimensions) {
+                                throw new IllegalArgumentException("All dense_vector must have the same number of dimensions. Expected: "
+                                    + dimensions + ", found: " + bytes.length);
+                            }
+                        }
+                        builder.beginPositionEntry();
+                        for (byte value : bytes) {
+                            builder.appendFloat(value);
+                        }
+                        builder.endPositionEntry();
+                    } catch (IllegalArgumentException  e) {
+                        registerException(e);
+                        builder.appendNull();
                     }
-                    builder.beginPositionEntry();
-                    for (byte value : bytes) {
-                        builder.appendFloat(value);
-                    }
-                    builder.endPositionEntry();
                 }
             }
             return builder.build();
diff --git a/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/vector/VectorSimilarityFunction.java b/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/vector/VectorSimilarityFunction.java
@@ -82,6 +82,17 @@ public int vectorArgumentsCount() {
         return 2;
     }
 
+//    @Override
+//    public BiConsumer<LogicalPlan, Failures> postAnalysisPlanVerification() {
+//        return (lp, failures) -> {
+//            for (Expression child : children()) {
+//                if (child instanceof PostAnalysisVerificationAware pa) {
+//                    pa.postAnalysisVerification(failures);
+//                }
+//            }
+//        };
+//    }
+
     @Override
     public final EvalOperator.ExpressionEvaluator.Factory toEvaluator(EvaluatorMapper.ToEvaluator toEvaluator) {
         return new SimilarityEvaluatorFactory(
diff --git a/x-pack/plugin/esql/src/test/java/org/elasticsearch/xpack/esql/analysis/AnalyzerTests.java b/x-pack/plugin/esql/src/test/java/org/elasticsearch/xpack/esql/analysis/AnalyzerTests.java

Original file line number	Diff line number	Diff line change
`@@ -90,10 +90,10 @@ magnitudeWithRow`
`90`	`90`	`required_capability: magnitude_scalar_vector_function`
`91`	`91`	`required_capability: to_dense_vector_function`
`92`	`92`
`93`		`-row vector = [1, 2, 3]`
	`93`	`+row vector = to_dense_vector([1, 2, 3])`
`94`	`94`	`\| eval magnitude = round(v_magnitude(vector), 3)`
`95`	`95`	`;`
`96`	`96`
`97`		`-vector: integer \| magnitude:double`
`98`		`-[1, 2, 3] \| 3.742`
	`97`	`+vector: dense_vector \| magnitude:double`
	`98`	`+[1.0, 2.0, 3.0] \| 3.742`
`99`	`99`	`;`