Fix sparse parser. (#1262)

### What problem does this PR solve?

1. Fix: memory leak when parsing sparse vector.
2. Fix: parser reduce warning by adding new DataType: EmptyArray
3. Refactor: knn expr parser.

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Refactoring

src/function/cast/cast_function.cpp

@@ -25,6 +25,7 @@ import embedding_cast;
 import varchar_cast;
 import tensor_cast;
 import tensor_array_cast;
+import empty_array_cast;
 import logger;
 import stl;
 import sparse_cast;
@@ -249,6 +250,9 @@ BoundCastFunc CastFunction::GetBoundFunc(const DataType &source, const DataType
             UnrecoverableError(error_message);
             break;
         }
+        case kEmptyArray: {
+            return BindEmptyArrayCast(source, target);
+        }
         default:
             String error_message = fmt::format("Can't cast from {} to {}", source.ToString(), target.ToString());
             LOG_CRITICAL(error_message);

src/function/cast/empty_array_cast.cppm

@@ -0,0 +1,58 @@
+// Copyright(C) 2023 InfiniFlow, Inc. All rights reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+module;
+
+export module empty_array_cast;
+
+import stl;
+import data_type;
+import bound_cast_func;
+import column_vector_cast;
+import infinity_exception;
+import internal_types;
+import logical_type;
+import third_party;
+
+namespace infinity {
+
+struct EmptyTryCastToFixlen;
+
+export inline BoundCastFunc BindEmptyArrayCast(const DataType &source, const DataType &target) {
+    switch (target.type()) {
+        case LogicalType::kSparse: {
+            return BoundCastFunc(&ColumnVectorCast::TryCastColumnVector<EmptyArrayT, SparseT, EmptyTryCastToFixlen>);
+        }
+        default: {
+            UnrecoverableError("Not implemented");
+        }
+    }
+    return BoundCastFunc(nullptr);
+}
+
+struct EmptyTryCastToFixlen {
+    template <typename SourceType, typename TargetType>
+    static bool Run(SourceType, TargetType &) {
+        UnrecoverableError(fmt::format("Not support to cast from {} to {}", DataType::TypeToString<SourceType>(), DataType::TypeToString<TargetType>()));
+        return false;
+    }
+};
+
+template<>
+bool EmptyTryCastToFixlen::Run(EmptyArrayT, SparseT &target) {
+    target.nnz_ = 0;
+    return true;
+}
+
+}

src/parser/expr/constant_expr.cpp

@@ -130,8 +130,7 @@ std::string ConstantExpr::ToString() const {
             size_t nnz = double_sparse_array_.first.size();
             return SparseT::Sparse2StringT2(double_sparse_array_.first.data(), double_sparse_array_.second.data(), nnz);
         }
-        default: {
-            ParserError("Unexpected branch");
+        case LiteralType::kEmptyArray: {
             return {};
         }
     }
@@ -156,6 +155,7 @@ int32_t ConstantExpr::GetSizeInBytes() const {
             size += sizeof(int64_t);
             break;
         }
+        case LiteralType::kEmptyArray:
         case LiteralType::kNull: {
             break;
         }
@@ -221,6 +221,7 @@ void ConstantExpr::WriteAdv(char *&ptr) const {
             WriteBufAdv<int64_t>(ptr, integer_value_);
             break;
         }
+        case LiteralType::kEmptyArray:
         case LiteralType::kNull: {
             break;
         }
@@ -307,6 +308,7 @@ std::shared_ptr<ParsedExpr> ConstantExpr::ReadAdv(char *&ptr, int32_t maxbytes)
             const_expr->integer_value_ = integer_value;
             break;
         }
+        case LiteralType::kEmptyArray:
         case LiteralType::kNull: {
             break;
         }
@@ -401,6 +403,7 @@ nlohmann::json ConstantExpr::Serialize() const {
             j["value"] = integer_value_;
             break;
         }
+        case LiteralType::kEmptyArray:
         case LiteralType::kNull: {
             break;
         }
@@ -468,6 +471,7 @@ std::shared_ptr<ParsedExpr> ConstantExpr::Deserialize(const nlohmann::json &cons
             const_expr->integer_value_ = constant_expr["value"].get<int64_t>();
             break;
         }
+        case LiteralType::kEmptyArray:
         case LiteralType::kNull: {
             break;
         }

src/parser/expr/constant_expr.h

@@ -41,6 +41,7 @@ enum class LiteralType : int32_t {
     kInterval,
     kLongSparseArray,
     kDoubleSparseArray,
+    kEmptyArray,
 };
 
 class ConstantExpr : public ParsedExpr {

src/parser/expr/knn_expr.cpp

@@ -100,6 +100,104 @@ std::string KnnExpr::ToString() const {
     return expr_str;
 }
 
+bool KnnExpr::InitDistanceType(const char *distance_type) {
+    if (strcmp(distance_type, "l2") == 0) {
+        distance_type_ = infinity::KnnDistanceType::kL2;
+    } else if (strcmp(distance_type, "ip") == 0) {
+        distance_type_ = infinity::KnnDistanceType::kInnerProduct;
+    } else if (strcmp(distance_type, "cosine") == 0) {
+        distance_type_ = infinity::KnnDistanceType::kCosine;
+    } else if (strcmp(distance_type, "hamming") == 0) {
+        distance_type_ = infinity::KnnDistanceType::kHamming;
+    } else {
+        return false;
+    }
+    return true;
+}
+
+bool KnnExpr::InitEmbedding(const char *data_type, const ConstantExpr *query_vec) {
+    if (strcmp(data_type, "float") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemFloat;
+        if (!(query_vec->double_array_.empty())) {
+            dimension_ = query_vec->double_array_.size();
+            embedding_data_ptr_ = new float[dimension_];
+            for (long i = 0; i < dimension_; ++i) {
+                ((float *)(embedding_data_ptr_))[i] = query_vec->double_array_[i];
+            }
+        }
+        if (!(query_vec->long_array_.empty())) {
+            dimension_ = query_vec->long_array_.size();
+            embedding_data_ptr_ = new float[dimension_];
+            for (long i = 0; i < dimension_; ++i) {
+                ((float *)(embedding_data_ptr_))[i] = query_vec->long_array_[i];
+            }
+        }
+    } else if (strcmp(data_type, "tinyint") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->long_array_.size();
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemInt8;
+        embedding_data_ptr_ = new char[dimension_];
+
+        for (long i = 0; i < dimension_; ++i) {
+            ((char *)embedding_data_ptr_)[i] = query_vec->long_array_[i];
+        }
+    } else if (strcmp(data_type, "smallint") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->long_array_.size();
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemInt16;
+        embedding_data_ptr_ = new short int[dimension_];
+
+        for (long i = 0; i < dimension_; ++i) {
+            ((short int *)embedding_data_ptr_)[i] = query_vec->long_array_[i];
+        }
+    } else if (strcmp(data_type, "integer") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->long_array_.size();
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemInt32;
+        embedding_data_ptr_ = new int[dimension_];
+
+        for (long i = 0; i < dimension_; ++i) {
+            ((int *)embedding_data_ptr_)[i] = query_vec->long_array_[i];
+        }
+    } else if (strcmp(data_type, "bigint") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->long_array_.size();
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemInt64;
+        embedding_data_ptr_ = new long[dimension_];
+
+        memcpy(embedding_data_ptr_, (void *)query_vec->long_array_.data(), dimension_ * sizeof(long));
+    } else if (strcmp(data_type, "bit") == 0 and distance_type_ == infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->long_array_.size();
+        if (dimension_ % 8 == 0) {
+            embedding_data_type_ = infinity::EmbeddingDataType::kElemBit;
+            long embedding_size = dimension_ / 8;
+            char *char_ptr = new char[embedding_size];
+            uint8_t *data_ptr = reinterpret_cast<uint8_t *>(char_ptr);
+            embedding_data_ptr_ = char_ptr;
+            for (long i = 0; i < embedding_size; ++i) {
+                uint8_t embedding_unit = 0;
+                for (long bit_idx = 0; bit_idx < 8; ++bit_idx) {
+                    if (query_vec->long_array_[i * 8 + bit_idx] == 1) {
+                        embedding_unit |= (uint8_t(1) << bit_idx);
+                    } else if (query_vec->long_array_[i * 8 + bit_idx] == 0) {
+                        // no-op
+                    } else {
+                        return false;
+                    }
+                }
+                data_ptr[i] = embedding_unit;
+            }
+        } else {
+            return false;
+        }
+    } else if (strcmp(data_type, "double") == 0 and distance_type_ != infinity::KnnDistanceType::kHamming) {
+        dimension_ = query_vec->double_array_.size();
+        embedding_data_type_ = infinity::EmbeddingDataType::kElemDouble;
+        embedding_data_ptr_ = new double[dimension_];
+
+        memcpy(embedding_data_ptr_, (void *)query_vec->double_array_.data(), dimension_ * sizeof(double));
+    } else {
+        return false;
+    }
+    return true;
+}
+
 std::string KnnExpr::KnnDistanceType2Str(KnnDistanceType knn_distance_type) {
     switch (knn_distance_type) {
         case KnnDistanceType::kL2: {

src/parser/expr/knn_expr.h

@@ -14,10 +14,11 @@
 
 #pragma once
 
+#include "expr/constant_expr.h"
 #include "parsed_expr.h"
 #include "statement/statement_common.h"
 #include "type/complex/embedding_type.h"
-#include <vector>
+
 namespace infinity {
 
 enum class KnnDistanceType {
@@ -36,6 +37,10 @@ class KnnExpr : public ParsedExpr {
 
     [[nodiscard]] std::string ToString() const override;
 
+    bool InitDistanceType(const char *distance_type);
+
+    bool InitEmbedding(const char *data_type, const ConstantExpr *query_vec);
+
 public:
     static std::string KnnDistanceType2Str(KnnDistanceType knn_distance_type);