Optimize storage of tiny strings (up to 3 characters)

Author: bblanchon

CHANGELOG.md

@@ -5,7 +5,7 @@ HEAD
 ----
 
 * Fix conversion from static string to number
-* Slightly reduce code size
+* Optimize storage of tiny strings (up to 3 characters)
 
 v7.3.0 (2024-12-29)
 ------

extras/tests/JsonObject/set.cpp

@@ -100,13 +100,13 @@ TEST_CASE("JsonObject::set()") {
     JsonDocument doc3(&timebomb);
     JsonObject obj3 = doc3.to<JsonObject>();
 
-    obj1["a"_s] = 1;
-    obj1["b"_s] = 2;
+    obj1["alpha"_s] = 1;
+    obj1["beta"_s] = 2;
 
     bool success = obj3.set(obj1);
 
     REQUIRE(success == false);
-    REQUIRE(doc3.as<std::string>() == "{\"a\":1}");
+    REQUIRE(doc3.as<std::string>() == "{\"alpha\":1}");
   }
 
   SECTION("copy fails in the middle of an array") {

extras/tests/JsonVariant/as.cpp

@@ -199,7 +199,7 @@ TEST_CASE("JsonVariant::as()") {
     REQUIRE(variant.as<JsonString>() == "hello");
   }
 
-  SECTION("set(std::string(\"4.2\"))") {
+  SECTION("set(std::string(\"4.2\")) (tiny string optimization)") {
     variant.set("4.2"_s);
 
     REQUIRE(variant.as<bool>() == true);
@@ -211,6 +211,18 @@ TEST_CASE("JsonVariant::as()") {
     REQUIRE(variant.as<JsonString>().isStatic() == false);
   }
 
+  SECTION("set(std::string(\"123.45\"))") {
+    variant.set("123.45"_s);
+
+    REQUIRE(variant.as<bool>() == true);
+    REQUIRE(variant.as<long>() == 123L);
+    REQUIRE(variant.as<double>() == Approx(123.45));
+    REQUIRE(variant.as<const char*>() == "123.45"_s);
+    REQUIRE(variant.as<std::string>() == "123.45"_s);
+    REQUIRE(variant.as<JsonString>() == "123.45");
+    REQUIRE(variant.as<JsonString>().isStatic() == false);
+  }
+
   SECTION("set(\"true\")") {
     variant.set("true");
 

extras/tests/JsonVariant/set.cpp

@@ -63,6 +63,18 @@ TEST_CASE("JsonVariant::set() when there is enough memory") {
                          });
   }
 
+  SECTION("char* (tiny string optimization)") {
+    char str[16];
+
+    strcpy(str, "abc");
+    bool result = variant.set(str);
+    strcpy(str, "def");
+
+    REQUIRE(result == true);
+    REQUIRE(variant == "abc");  // stores by copy
+    REQUIRE(spy.log() == AllocatorLog{});
+  }
+
   SECTION("(char*)0") {
     bool result = variant.set(static_cast<char*>(0));
 

src/ArduinoJson/Variant/VariantContent.hpp

@@ -24,6 +24,7 @@ enum class VariantTypeBits : uint8_t {
 
 enum class VariantType : uint8_t {
   Null = 0,             // 0000 0000
+  TinyString = 0x02,    // 0000 0010
   RawString = 0x03,     // 0000 0011
   LinkedString = 0x04,  // 0000 0100
   OwnedString = 0x05,   // 0000 0101
@@ -46,6 +47,8 @@ inline bool operator&(VariantType type, VariantTypeBits bit) {
   return (uint8_t(type) & uint8_t(bit)) != 0;
 }
 
+const size_t tinyStringMaxLength = 3;
+
 union VariantContent {
   VariantContent() {}
 
@@ -61,6 +64,7 @@ union VariantContent {
   CollectionData asCollection;
   const char* asLinkedString;
   struct StringNode* asOwnedString;
+  char asTinyString[tinyStringMaxLength + 1];
 };
 
 #if ARDUINOJSON_USE_EXTENSIONS

src/ArduinoJson/Variant/VariantData.hpp

@@ -63,6 +63,9 @@ class VariantData {
       case VariantType::Object:
         return visit.visit(content_.asObject);
 
+      case VariantType::TinyString:
+        return visit.visit(JsonString(content_.asTinyString));
+
       case VariantType::LinkedString:
         return visit.visit(JsonString(content_.asLinkedString, true));
 
@@ -199,6 +202,9 @@ class VariantData {
       case VariantType::Int64:
         return static_cast<T>(extension->asInt64);
 #endif
+      case VariantType::TinyString:
+        str = content_.asTinyString;
+        break;
       case VariantType::LinkedString:
         str = content_.asLinkedString;
         break;
@@ -241,6 +247,9 @@ class VariantData {
       case VariantType::Int64:
         return convertNumber<T>(extension->asInt64);
 #endif
+      case VariantType::TinyString:
+        str = content_.asTinyString;
+        break;
       case VariantType::LinkedString:
         str = content_.asLinkedString;
         break;
@@ -281,6 +290,8 @@ class VariantData {
 
   JsonString asString() const {
     switch (type_) {
+      case VariantType::TinyString:
+        return JsonString(content_.asTinyString);
       case VariantType::LinkedString:
         return JsonString(content_.asLinkedString, true);
       case VariantType::OwnedString:
@@ -395,7 +406,8 @@ class VariantData {
 
   bool isString() const {
     return type_ == VariantType::LinkedString ||
-           type_ == VariantType::OwnedString;
+           type_ == VariantType::OwnedString ||
+           type_ == VariantType::TinyString;
   }
 
   size_t nesting(const ResourceManager* resources) const {
@@ -509,6 +521,14 @@ class VariantData {
     content_.asLinkedString = s;
   }
 
+  void setTinyString(const char* s, uint8_t n) {
+    ARDUINOJSON_ASSERT(type_ == VariantType::Null);  // must call clear() first
+    ARDUINOJSON_ASSERT(s);
+    type_ = VariantType::TinyString;
+    for (uint8_t i = 0; i < n + 1; i++)
+      content_.asTinyString[i] = s[i];
+  }
+
   void setOwnedString(StringNode* s) {
     ARDUINOJSON_ASSERT(type_ == VariantType::Null);  // must call clear() first
     ARDUINOJSON_ASSERT(s);

src/ArduinoJson/Variant/VariantImpl.hpp

@@ -31,6 +31,18 @@ inline bool VariantData::setString(TAdaptedString value,
     return true;
   }
 
+  if (value.size() <= tinyStringMaxLength) {
+    uint8_t length = static_cast<uint8_t>(value.size());
+    bool containsNul = false;
+    for (uint8_t i = 0; i < length; i++)
+      containsNul |= !value[i];
+    bool nulTerminated = value[length] == 0;
+    if (!containsNul && nulTerminated) {
+      setTinyString(value.data(), length);
+      return true;
+    }
+  }
+
   auto dup = resources->saveString(value);
   if (dup) {
     setOwnedString(dup);