From 5b77d45a277df4a3286f459bc98dd48277db1a94 Mon Sep 17 00:00:00 2001
From: Buqian Zheng <zhengbuqian@gmail.com>
Date: Sat, 29 Nov 2025 21:55:09 +0800
Subject: [PATCH] fix: [2.6] term expr to correctly handle in of string in json
 (#45956)

issue: https://github.com/milvus-io/milvus/issues/45887
pr: https://github.com/milvus-io/milvus/pull/45955

---------

Signed-off-by: Buqian Zheng <zhengbuqian@gmail.com>
---
 internal/core/src/exec/expression/Element.h | 36 +++++++++++++++++++--
 1 file changed, 34 insertions(+), 2 deletions(-)

diff --git a/internal/core/src/exec/expression/Element.h b/internal/core/src/exec/expression/Element.h
index af302d44d3..15004e5159 100644
--- a/internal/core/src/exec/expression/Element.h
+++ b/internal/core/src/exec/expression/Element.h
@@ -31,6 +31,18 @@
 namespace milvus {
 namespace exec {
 
+// Transparent string hash for heterogeneous lookup in unordered_dense::set
+// See: https://github.com/martinus/unordered_dense#324-heterogeneous-overloads-using-is_transparent
+struct StringHash {
+    using is_transparent = void;  // enable heterogeneous overloads
+    using is_avalanching = void;  // mark as high quality avalanching hash
+
+    [[nodiscard]] auto
+    operator()(std::string_view str) const noexcept -> uint64_t {
+        return ankerl::unordered_dense::hash<std::string_view>{}(str);
+    }
+};
+
 class BaseElement {
  public:
     virtual ~BaseElement() = default;
@@ -223,6 +235,13 @@ class FlatVectorElement : public MultiElement {
                     return true;
             }
         }
+        // Handle string_view -> string comparison when T is std::string
+        if constexpr (std::is_same_v<T, std::string>) {
+            if (auto sv = std::get_if<std::string_view>(&value)) {
+                return std::find(values_.begin(), values_.end(), *sv) !=
+                       values_.end();
+            }
+        }
         return false;
     }
 
@@ -242,6 +261,13 @@ class FlatVectorElement : public MultiElement {
 
 template <typename T>
 class SetElement : public MultiElement {
+    // Use transparent hash for std::string to enable heterogeneous lookup
+    // This allows O(1) lookup with string_view without string copy
+    using SetType = std::conditional_t<
+        std::is_same_v<T, std::string>,
+        ankerl::unordered_dense::set<T, StringHash, std::equal_to<>>,
+        ankerl::unordered_dense::set<T>>;
+
  public:
     explicit SetElement(const std::vector<proto::plan::GenericValue>& values) {
         for (auto& value : values) {
@@ -263,7 +289,13 @@ class SetElement : public MultiElement {
     bool
     In(const ValueType& value) const override {
         if (std::holds_alternative<T>(value)) {
-            return values_.count(std::get<T>(value)) > 0;
+            return values_.find(std::get<T>(value)) != values_.end();
+        }
+        // Handle string_view -> string comparison when T is std::string
+        if constexpr (std::is_same_v<T, std::string>) {
+            if (auto sv = std::get_if<std::string_view>(&value)) {
+                return values_.find(*sv) != values_.end();
+            }
         }
         return false;
     }
@@ -284,7 +316,7 @@ class SetElement : public MultiElement {
     }
 
  public:
-    ankerl::unordered_dense::set<T> values_;
+    SetType values_;
 };
 
 template <>