feat: add json serde for expressions

Author: Innocent

src/iceberg/expression/json_serde.cc

@@ -17,22 +17,33 @@
  * under the License.
  */
 
-#include <format>
 #include <ranges>
 #include <string>
-#include <utility>
 #include <vector>
 
 #include <nlohmann/json.hpp>
 
 #include "iceberg/expression/json_serde_internal.h"
 #include "iceberg/expression/literal.h"
+#include "iceberg/expression/predicate.h"
+#include "iceberg/expression/term.h"
+#include "iceberg/transform.h"
 #include "iceberg/util/checked_cast.h"
 #include "iceberg/util/json_util_internal.h"
 #include "iceberg/util/macros.h"
+#include "iceberg/util/transform_util.h"
 
 namespace iceberg {
 namespace {
+// JSON field names
+constexpr std::string_view kType = "type";
+constexpr std::string_view kTerm = "term";
+constexpr std::string_view kTransform = "transform";
+constexpr std::string_view kValue = "value";
+constexpr std::string_view kValues = "values";
+constexpr std::string_view kLeft = "left";
+constexpr std::string_view kRight = "right";
+constexpr std::string_view kChild = "child";
 // Expression type strings
 constexpr std::string_view kTypeTrue = "true";
 constexpr std::string_view kTypeFalse = "false";
@@ -58,6 +69,53 @@ constexpr std::string_view kTypeCountNull = "count-null";
 constexpr std::string_view kTypeCountStar = "count-star";
 constexpr std::string_view kTypeMin = "min";
 constexpr std::string_view kTypeMax = "max";
+constexpr std::string_view kTypeLiteral = "literal";
+constexpr std::string_view kTypeReference = "reference";
+
+/// Helper to check if a JSON term represents a transform
+bool IsTransformTerm(const nlohmann::json& json) {
+  return json.is_object() && json.contains(kType) &&
+         json[kType].get<std::string>() == kTransform && json.contains(kTerm);
+}
+
+/// Template helper to create predicates from JSON with the appropriate term type
+template <typename B>
+Result<std::unique_ptr<UnboundPredicate>> MakePredicateFromJson(
+    Expression::Operation op, std::shared_ptr<UnboundTerm<B>> term,
+    const nlohmann::json& json) {
+  if (IsUnaryOperation(op)) {
+    if (json.contains(kValue)) [[unlikely]] {
+      return JsonParseError("Unary predicate has invalid 'value' field: {}",
+                            SafeDumpJson(json));
+    }
+    if (json.contains(kValues)) [[unlikely]] {
+      return JsonParseError("Unary predicate has invalid 'values' field: {}",
+                            SafeDumpJson(json));
+    }
+    return UnboundPredicateImpl<B>::Make(op, std::move(term));
+  }
+
+  if (IsSetOperation(op)) {
+    std::vector<Literal> literals;
+    if (!json.contains(kValues) || !json[kValues].is_array()) [[unlikely]] {
+      return JsonParseError("Missing or invalid 'values' field for set operation: {}",
+                            SafeDumpJson(json));
+    }
+    for (const auto& val : json[kValues]) {
+      ICEBERG_ASSIGN_OR_RAISE(auto lit, LiteralFromJson(val));
+      literals.push_back(std::move(lit));
+    }
+    return UnboundPredicateImpl<B>::Make(op, std::move(term), std::move(literals));
+  }
+
+  // Literal predicate
+  if (!json.contains(kValue)) [[unlikely]] {
+    return JsonParseError("Missing 'value' field for literal predicate: {}",
+                          SafeDumpJson(json));
+  }
+  ICEBERG_ASSIGN_OR_RAISE(auto literal, LiteralFromJson(json[kValue]));
+  return UnboundPredicateImpl<B>::Make(op, std::move(term), std::move(literal));
+}
 }  // namespace
 
 bool IsUnaryOperation(Expression::Operation op) {
@@ -83,7 +141,7 @@ bool IsSetOperation(Expression::Operation op) {
 }
 
 Result<Expression::Operation> OperationTypeFromJson(const nlohmann::json& json) {
-  if (!json.is_string()) {
+  if (!json.is_string()) [[unlikely]] {
     return JsonParseError("Unable to create operation. Json value is not a string");
   }
   auto typeStr = json.get<std::string>();
@@ -123,27 +181,252 @@ nlohmann::json ToJson(Expression::Operation op) {
   return json;
 }
 
+nlohmann::json ToJson(const NamedReference& ref) { return ref.name(); }
+
+Result<std::unique_ptr<NamedReference>> NamedReferenceFromJson(
+    const nlohmann::json& json) {
+  if (json.is_object() && json.contains(kType) &&
+      json[kType].get<std::string>() == kTypeReference && json.contains(kTerm)) {
+    return NamedReference::Make(json[kTerm].get<std::string>());
+  }
+  if (!json.is_string()) [[unlikely]] {
+    return JsonParseError("Expected string for named reference");
+  }
+  return NamedReference::Make(json.get<std::string>());
+}
+
+nlohmann::json ToJson(const UnboundTransform& transform) {
+  auto& mutable_transform = const_cast<UnboundTransform&>(transform);
+  nlohmann::json json;
+  json[kType] = kTransform;
+  json[kTransform] = transform.transform()->ToString();
+  json[kTerm] = mutable_transform.reference()->name();
+  return json;
+}
+
+Result<std::unique_ptr<UnboundTransform>> UnboundTransformFromJson(
+    const nlohmann::json& json) {
+  if (IsTransformTerm(json)) {
+    ICEBERG_ASSIGN_OR_RAISE(auto transform_str,
+                            GetJsonValue<std::string>(json, kTransform));
+    ICEBERG_ASSIGN_OR_RAISE(auto transform, TransformFromString(transform_str));
+    ICEBERG_ASSIGN_OR_RAISE(auto ref, NamedReferenceFromJson(json[kTerm]));
+    return UnboundTransform::Make(std::move(ref), std::move(transform));
+  }
+  return JsonParseError("Invalid unbound transform json: {}", SafeDumpJson(json));
+}
+
+nlohmann::json ToJson(const Literal& literal) {
+  if (literal.IsNull()) {
+    return nullptr;
+  }
+
+  const auto type_id = literal.type()->type_id();
+  const auto& value = literal.value();
+
+  switch (type_id) {
+    case TypeId::kBoolean:
+      return std::get<bool>(value);
+    case TypeId::kInt:
+      return std::get<int32_t>(value);
+    case TypeId::kDate:
+      return TransformUtil::HumanDay(std::get<int32_t>(value));
+    case TypeId::kLong:
+      return std::get<int64_t>(value);
+    case TypeId::kTime:
+      return TransformUtil::HumanTime(std::get<int64_t>(value));
+    case TypeId::kTimestamp:
+      return TransformUtil::HumanTimestamp(std::get<int64_t>(value));
+    case TypeId::kTimestampTz:
+      return TransformUtil::HumanTimestampWithZone(std::get<int64_t>(value));
+    case TypeId::kFloat:
+      return std::get<float>(value);
+    case TypeId::kDouble:
+      return std::get<double>(value);
+    case TypeId::kString:
+      return std::get<std::string>(value);
+    case TypeId::kBinary:
+    case TypeId::kFixed: {
+      const auto& bytes = std::get<std::vector<uint8_t>>(value);
+      std::string hex;
+      hex.reserve(bytes.size() * 2);
+      for (uint8_t byte : bytes) {
+        hex += std::format("{:02X}", byte);
+      }
+      return hex;
+    }
+    case TypeId::kDecimal: {
+      return literal.ToString();
+    }
+    case TypeId::kUuid:
+      return std::get<Uuid>(value).ToString();
+    default:
+      nlohmann::json json;
+      return json;
+  }
+}
+
+Result<Literal> LiteralFromJson(const nlohmann::json& json) {
+  // Unwrap {"type": "literal", "value": <actual>} wrapper
+  if (json.is_object() && json.contains(kType) &&
+      json[kType].get<std::string>() == kTypeLiteral && json.contains(kValue)) {
+    return LiteralFromJson(json[kValue]);
+  }
+  if (json.is_null()) {
+    return Literal::Null(nullptr);
+  }
+  if (json.is_boolean()) {
+    return Literal::Boolean(json.get<bool>());
+  }
+  if (json.is_number_integer()) {
+    return Literal::Long(json.get<int64_t>());
+  }
+  if (json.is_number_float()) {
+    return Literal::Double(json.get<double>());
+  }
+  if (json.is_string()) {
+    // All strings are returned as String literals.
+    // Conversion to binary/date/time/etc. happens during binding
+    // when schema type information is available.
+    return Literal::String(json.get<std::string>());
+  }
+  return JsonParseError("Unsupported literal JSON type");
+}
+
+nlohmann::json TermToJson(const Term& term) {
+  switch (term.kind()) {
+    case Term::Kind::kReference:
+      return ToJson(static_cast<const NamedReference&>(term));
+    case Term::Kind::kTransform:
+      return ToJson(static_cast<const UnboundTransform&>(term));
+    default:
+      return nullptr;
+  }
+}
+
+nlohmann::json ToJson(const UnboundPredicate& pred) {
+  nlohmann::json json;
+  json[kType] = ToJson(pred.op());
+
+  // Get term and literals by casting to the appropriate impl type
+  std::span<const Literal> literals;
+
+  if (auto* ref_pred = dynamic_cast<const UnboundPredicateImpl<BoundReference>*>(&pred)) {
+    json[kTerm] = TermToJson(*ref_pred->term());
+    literals = ref_pred->literals();
+  } else if (auto* transform_pred =
+                 dynamic_cast<const UnboundPredicateImpl<BoundTransform>*>(&pred)) {
+    json[kTerm] = TermToJson(*transform_pred->term());
+    literals = transform_pred->literals();
+  }
+
+  if (!IsUnaryOperation(pred.op())) {
+    if (IsSetOperation(pred.op())) {
+      nlohmann::json values = nlohmann::json::array();
+      for (const auto& lit : literals) {
+        values.push_back(ToJson(lit));
+      }
+      json[kValues] = std::move(values);
+    } else if (!literals.empty()) {
+      json[kValue] = ToJson(literals[0]);
+    }
+  }
+  return json;
+}
+
+Result<std::unique_ptr<UnboundPredicate>> UnboundPredicateFromJson(
+    const nlohmann::json& json) {
+  ICEBERG_ASSIGN_OR_RAISE(auto op, OperationTypeFromJson(json[kType]));
+
+  const auto& term_json = json[kTerm];
+
+  if (IsTransformTerm(term_json)) {
+    ICEBERG_ASSIGN_OR_RAISE(auto term, UnboundTransformFromJson(term_json));
+    return MakePredicateFromJson<BoundTransform>(op, std::move(term), json);
+  }
+
+  ICEBERG_ASSIGN_OR_RAISE(auto term, NamedReferenceFromJson(term_json));
+  return MakePredicateFromJson<BoundReference>(op, std::move(term), json);
+}
+
 Result<std::shared_ptr<Expression>> ExpressionFromJson(const nlohmann::json& json) {
-  // Handle boolean
+  // Handle boolean constants
   if (json.is_boolean()) {
     return json.get<bool>()
                ? internal::checked_pointer_cast<Expression>(True::Instance())
                : internal::checked_pointer_cast<Expression>(False::Instance());
   }
-  return JsonParseError("Only booleans are currently supported.");
+
+  if (!json.is_object()) [[unlikely]] {
+    return JsonParseError("Expression must be boolean or object");
+  }
+
+  ICEBERG_ASSIGN_OR_RAISE(auto op, OperationTypeFromJson(json[kType]));
+
+  switch (op) {
+    case Expression::Operation::kAnd: {
+      if (!json.contains(kLeft) || !json.contains(kRight)) [[unlikely]] {
+        return JsonParseError("AND expression missing 'left' or 'right' field");
+      }
+      ICEBERG_ASSIGN_OR_RAISE(auto left, ExpressionFromJson(json[kLeft]));
+      ICEBERG_ASSIGN_OR_RAISE(auto right, ExpressionFromJson(json[kRight]));
+      ICEBERG_ASSIGN_OR_RAISE(auto result, And::Make(std::move(left), std::move(right)));
+      return std::shared_ptr<Expression>(std::move(result));
+    }
+    case Expression::Operation::kOr: {
+      if (!json.contains(kLeft) || !json.contains(kRight)) [[unlikely]] {
+        return JsonParseError("OR expression missing 'left' or 'right' field");
+      }
+      ICEBERG_ASSIGN_OR_RAISE(auto left, ExpressionFromJson(json[kLeft]));
+      ICEBERG_ASSIGN_OR_RAISE(auto right, ExpressionFromJson(json[kRight]));
+      ICEBERG_ASSIGN_OR_RAISE(auto result, Or::Make(std::move(left), std::move(right)));
+      return std::shared_ptr<Expression>(std::move(result));
+    }
+    case Expression::Operation::kNot: {
+      if (!json.contains(kChild)) [[unlikely]] {
+        return JsonParseError("NOT expression missing 'child' field");
+      }
+      ICEBERG_ASSIGN_OR_RAISE(auto child, ExpressionFromJson(json[kChild]));
+      ICEBERG_ASSIGN_OR_RAISE(auto result, Not::Make(std::move(child)));
+      return std::shared_ptr<Expression>(std::move(result));
+    }
+    default:
+      // All other operations are predicates
+      return UnboundPredicateFromJson(json);
+  }
 }
 
 nlohmann::json ToJson(const Expression& expr) {
   switch (expr.op()) {
     case Expression::Operation::kTrue:
       return true;
-
     case Expression::Operation::kFalse:
       return false;
+    case Expression::Operation::kAnd: {
+      const auto& and_expr = static_cast<const And&>(expr);
+      nlohmann::json json;
+      json[kType] = ToJson(expr.op());
+      json[kLeft] = ToJson(*and_expr.left());
+      json[kRight] = ToJson(*and_expr.right());
+      return json;
+    }
+    case Expression::Operation::kOr: {
+      const auto& or_expr = static_cast<const Or&>(expr);
+      nlohmann::json json;
+      json[kType] = ToJson(expr.op());
+      json[kLeft] = ToJson(*or_expr.left());
+      json[kRight] = ToJson(*or_expr.right());
+      return json;
+    }
+    case Expression::Operation::kNot: {
+      const auto& not_expr = static_cast<const Not&>(expr);
+      nlohmann::json json;
+      json[kType] = ToJson(expr.op());
+      json[kChild] = ToJson(*not_expr.child());
+      return json;
+    }
     default:
-      // TODO(evindj): This code will be removed as we implemented the full expression
-      // serialization.
-      ICEBERG_CHECK_OR_DIE(false, "Only booleans are currently supported.");
+      return ToJson(dynamic_cast<const UnboundPredicate&>(expr));
   }
 }