parquet 0.6.1 → 0.6.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: dd9295880f123b0ed979fb970d789bf63461007e8c087641b47391baa02bef29
-  data.tar.gz: e5ccd26a48e7b800412049e9d44e7f8c6fb5c12dbe25c74980467c822fb114c6
+  metadata.gz: dfd19103b2414e7feeaa6d1ec3c9a9c25ce42cf5c8362baa37e3b9d8d5245f82
+  data.tar.gz: c5c1170dbdc3635577738a568688c36adc9670710f4b0d570fae29294e337754
 SHA512:
-  metadata.gz: f275d82733cd6b1658ed7b6212618830e2a12e6831bcfcc26beeae79691ddc972b22b0b0660f10db9b0bb7a03e2af29da15de28d037d5896891fbd678b3ecca9
-  data.tar.gz: 8fbfd33a92350c84409c0e879745fbc5a07d60f0a2862a2993a4780eb5c110be7b3918bfdfe6721e60e44635792108a1a3eeed460a189076bd03056a27764e50
+  metadata.gz: c9bf72b4e708c750ab7ae30afd97aef7f456a4249904fe3eb74f916557e28ca1a53bc262a6492db38c38162ab3e3f684e30f0c70dabbaf8f8f4145ef4d9af259
+  data.tar.gz: 164c5b0569d3d13242bcff7c09d66edf67b279d8289f97def043000a508b4333dd1387a4f47517be09023cd270cf3b6dfd57fdf658ac1b52e25f3f5b2b5ca30c

data/Cargo.lock

@@ -225,6 +225,10 @@ dependencies = [
 name = "arrow-schema"
 version = "55.2.0"
 source = "git+https://github.com/njaremko/arrow-rs?branch=nathan_06-24-remove_primitive_map_key_assertion_on_record_reader#54858bf019ff3faeb8f5b562da8c01012162aef0"
+dependencies = [
+ "serde",
+ "serde_json",
+]
 
 [[package]]
 name = "arrow-select"
@@ -1092,6 +1096,7 @@ dependencies = [
  "rb-sys-env 0.2.2",
  "tempfile",
  "thiserror",
+ "uuid",
 ]
 
 [[package]]

data/ext/parquet-core/Cargo.toml

@@ -7,7 +7,7 @@ edition = "2021"
 arrow = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24-remove_primitive_map_key_assertion_on_record_reader" }
 arrow-array = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24-remove_primitive_map_key_assertion_on_record_reader" }
 arrow-buffer = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24-remove_primitive_map_key_assertion_on_record_reader" }
-arrow-schema = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24-remove_primitive_map_key_assertion_on_record_reader" }
+arrow-schema = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24-remove_primitive_map_key_assertion_on_record_reader", features = ["canonical_extension_types"]}
 bytes = "1.5"
 indexmap = "2.2"
 jiff = "0.2"
@@ -17,7 +17,7 @@ parquet = { git = "https://github.com/njaremko/arrow-rs", branch = "nathan_06-24
 rand = "0.9.1"
 serde = { version = "1.0", features = ["derive"] }
 thiserror = "2.0"
+uuid = { version = "1.0", features = ["v4"] }
 
 [dev-dependencies]
-uuid = { version = "1.0", features = ["v4"] }
 tempfile = "3.8"

data/ext/parquet-core/src/arrow_conversion.rs

@@ -7,6 +7,7 @@
 
 use crate::{ParquetError, ParquetValue, Result};
 use arrow_array::{builder::*, Array, ArrayRef, ListArray, MapArray, StructArray};
+use arrow_schema::extension::Uuid as ArrowUuid;
 use arrow_schema::{DataType, Field};
 use bytes::Bytes;
 use indexmap::IndexMap;
@@ -14,7 +15,11 @@ use ordered_float::OrderedFloat;
 use std::sync::Arc;
 
 /// Convert a single value from an Arrow array at the given index to a ParquetValue
-pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<ParquetValue> {
+pub fn arrow_to_parquet_value(
+    field: &Field,
+    array: &dyn Array,
+    index: usize,
+) -> Result<ParquetValue> {
     use arrow_array::*;
 
     if array.is_null(index) {
@@ -72,7 +77,6 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
             let array = downcast_array::<Float64Array>(array)?;
             Ok(ParquetValue::Float64(OrderedFloat(array.value(index))))
         }
-
         // String and binary types
         DataType::Utf8 => {
             let array = downcast_array::<StringArray>(array)?;
@@ -86,9 +90,15 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
         }
         DataType::FixedSizeBinary(_) => {
             let array = downcast_array::<FixedSizeBinaryArray>(array)?;
-            Ok(ParquetValue::Bytes(Bytes::copy_from_slice(
-                array.value(index),
-            )))
+            let value = array.value(index);
+            match field.try_extension_type::<ArrowUuid>() {
+                Ok(_) => {
+                    let uuid = uuid::Uuid::from_slice(value)
+                        .map_err(|e| ParquetError::Conversion(format!("Invalid UUID: {}", e)))?;
+                    Ok(ParquetValue::Uuid(uuid))
+                }
+                Err(_) => Ok(ParquetValue::Bytes(Bytes::copy_from_slice(value))),
+            }
         }
 
         // Date and time types
@@ -140,6 +150,10 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
                 let array = downcast_array::<Time64MicrosecondArray>(array)?;
                 Ok(ParquetValue::TimeMicros(array.value(index)))
             }
+            arrow_schema::TimeUnit::Nanosecond => {
+                let array = downcast_array::<Time64NanosecondArray>(array)?;
+                Ok(ParquetValue::TimeNanos(array.value(index)))
+            }
             _ => Err(ParquetError::Conversion(format!(
                 "Unsupported time64 unit: {:?}",
                 unit
@@ -173,13 +187,13 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
         }
 
         // Complex types
-        DataType::List(_) => {
+        DataType::List(item_field) => {
             let array = downcast_array::<ListArray>(array)?;
             let list_values = array.value(index);
 
             let mut values = Vec::with_capacity(list_values.len());
             for i in 0..list_values.len() {
-                values.push(arrow_to_parquet_value(&list_values, i)?);
+                values.push(arrow_to_parquet_value(item_field, &list_values, i)?);
             }
 
             Ok(ParquetValue::List(values))
@@ -192,10 +206,20 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
             let keys = map_value.column(0);
             let values = map_value.column(1);
 
+            let key_field = map_value
+                .fields()
+                .iter().find(|f| f.name() == "key")
+                .ok_or_else(|| ParquetError::Conversion("No value field found".to_string()))?;
+
+            let value_field = map_value
+                .fields()
+                .iter().find(|f| f.name() == "value")
+                .ok_or_else(|| ParquetError::Conversion("No value field found".to_string()))?;
+
             let mut map_vec = Vec::with_capacity(keys.len());
             for i in 0..keys.len() {
-                let key = arrow_to_parquet_value(keys, i)?;
-                let value = arrow_to_parquet_value(values, i)?;
+                let key = arrow_to_parquet_value(key_field, keys, i)?;
+                let value = arrow_to_parquet_value(value_field, values, i)?;
                 map_vec.push((key, value));
             }
 
@@ -207,7 +231,7 @@ pub fn arrow_to_parquet_value(array: &dyn Array, index: usize) -> Result<Parquet
             let mut map = IndexMap::new();
             for (col_idx, field) in array.fields().iter().enumerate() {
                 let column = array.column(col_idx);
-                let value = arrow_to_parquet_value(column, index)?;
+                let value = arrow_to_parquet_value(field, column, index)?;
                 map.insert(Arc::from(field.name().as_str()), value);
             }
 
@@ -1108,7 +1132,7 @@ mod tests {
         let array = parquet_values_to_arrow_array(values.clone(), &field).unwrap();
 
         for (i, expected) in values.iter().enumerate() {
-            let actual = arrow_to_parquet_value(array.as_ref(), i).unwrap();
+            let actual = arrow_to_parquet_value(&field, array.as_ref(), i).unwrap();
             assert_eq!(&actual, expected);
         }
     }

data/ext/parquet-core/src/reader.rs

@@ -163,8 +163,10 @@ where
                     // Extract values from current row
                     let mut row_values = Vec::with_capacity(batch.num_columns());
 
-                    for column in batch.columns() {
-                        let value = match arrow_to_parquet_value(column, self.current_row) {
+                    let schema = batch.schema();
+                    for (i, column) in batch.columns().iter().enumerate() {
+                        let field = schema.field(i);
+                        let value = match arrow_to_parquet_value(field, column, self.current_row) {
                             Ok(v) => v,
                             Err(e) => return Some(Err(e)),
                         };
@@ -228,12 +230,13 @@ where
                 let mut columns = Vec::with_capacity(batch.num_columns());
 
                 for (idx, column) in batch.columns().iter().enumerate() {
-                    let column_name = self.schema.field(idx).name().to_string();
+                    let field = self.schema.field(idx);
+                    let column_name = field.name().to_string();
 
                     // Convert entire column to ParquetValues
                     let mut values = Vec::with_capacity(column.len());
                     for row_idx in 0..column.len() {
-                        match arrow_to_parquet_value(column, row_idx) {
+                        match arrow_to_parquet_value(field, column, row_idx) {
                             Ok(value) => values.push(value),
                             Err(e) => return Some(Err(e)),
                         }

data/ext/parquet-core/src/schema.rs

@@ -72,6 +72,7 @@ pub enum PrimitiveType {
     TimestampNanos(Option<Arc<str>>),
     TimeMillis,
     TimeMicros,
+    TimeNanos,
 
     // Fixed-length byte array
     FixedLenByteArray(i32),
@@ -146,6 +147,7 @@ impl PrimitiveType {
             PrimitiveType::TimestampNanos(_) => "TimestampNanos",
             PrimitiveType::TimeMillis => "TimeMillis",
             PrimitiveType::TimeMicros => "TimeMicros",
+            PrimitiveType::TimeNanos => "TimeNanos",
             PrimitiveType::FixedLenByteArray(_) => "FixedLenByteArray",
         }
     }

data/ext/parquet-core/src/value.rs

@@ -2,6 +2,7 @@ use bytes::Bytes;
 use indexmap::IndexMap;
 use num::BigInt;
 use std::sync::Arc;
+use uuid::Uuid;
 
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub enum ParquetValue {
@@ -22,6 +23,7 @@ pub enum ParquetValue {
     Boolean(bool),
     String(Arc<str>),
     Bytes(Bytes),
+    Uuid(Uuid),
 
     // Date/Time types
     Date32(i32), // Days since epoch
@@ -40,6 +42,7 @@ pub enum ParquetValue {
     // Time types
     TimeMillis(i32), // Time of day in milliseconds since midnight
     TimeMicros(i64), // Time of day in microseconds since midnight
+    TimeNanos(i64),  // Time of day in nanoseconds since midnight
 
     // Complex types
     List(Vec<ParquetValue>),
@@ -68,6 +71,7 @@ impl std::hash::Hash for ParquetValue {
             ParquetValue::Boolean(b) => b.hash(state),
             ParquetValue::String(s) => s.hash(state),
             ParquetValue::Bytes(b) => b.hash(state),
+            ParquetValue::Uuid(u) => u.hash(state),
             ParquetValue::Date32(d) => d.hash(state),
             ParquetValue::Date64(d) => d.hash(state),
             ParquetValue::Decimal128(d, scale) => {
@@ -96,6 +100,7 @@ impl std::hash::Hash for ParquetValue {
             }
             ParquetValue::TimeMillis(t) => t.hash(state),
             ParquetValue::TimeMicros(t) => t.hash(state),
+            ParquetValue::TimeNanos(t) => t.hash(state),
             ParquetValue::List(l) => l.hash(state),
             ParquetValue::Map(m) => m.hash(state),
             ParquetValue::Record(r) => {
@@ -133,6 +138,7 @@ impl ParquetValue {
             ParquetValue::Boolean(_) => "Boolean",
             ParquetValue::String(_) => "String",
             ParquetValue::Bytes(_) => "Bytes",
+            ParquetValue::Uuid(_) => "Uuid",
             ParquetValue::Date32(_) => "Date32",
             ParquetValue::Date64(_) => "Date64",
             ParquetValue::Decimal128(_, _) => "Decimal128",
@@ -143,6 +149,7 @@ impl ParquetValue {
             ParquetValue::TimestampNanos(_, _) => "TimestampNanos",
             ParquetValue::TimeMillis(_) => "TimeMillis",
             ParquetValue::TimeMicros(_) => "TimeMicros",
+            ParquetValue::TimeNanos(_) => "TimeNanos",
             ParquetValue::List(_) => "List",
             ParquetValue::Map(_) => "Map",
             ParquetValue::Record(_) => "Record",

data/ext/parquet-core/src/writer.rs

@@ -235,6 +235,7 @@ where
             (Date64(_), DataType::Date64) => 8,
             (TimeMillis(_), DataType::Time32(_)) => 4,
             (TimeMicros(_), DataType::Time64(_)) => 8,
+            (TimeNanos(_), DataType::Time64(_)) => 8,
             (TimestampSecond(_, _), DataType::Timestamp(_, _)) => 8,
             (TimestampMillis(_, _), DataType::Timestamp(_, _)) => 8,
             (TimestampMicros(_, _), DataType::Timestamp(_, _)) => 8,
@@ -364,7 +365,9 @@ where
             writer.write(&batch)?;
 
             // Check if we need to flush based on memory usage
-            if writer.in_progress_size() >= self.memory_threshold {
+            if writer.in_progress_size() >= self.memory_threshold
+                || writer.memory_size() >= self.memory_threshold
+            {
                 writer.flush()?;
             }
         } else {
@@ -496,6 +499,7 @@ fn validate_value_against_field(value: &ParquetValue, field: &Field, path: &str)
         (Date64(_), DataType::Date64) => Ok(()),
         (TimeMillis(_), DataType::Time32(_)) => Ok(()),
         (TimeMicros(_), DataType::Time64(_)) => Ok(()),
+        (TimeNanos(_), DataType::Time64(_)) => Ok(()),
         (TimestampSecond(_, _), DataType::Timestamp(_, _)) => Ok(()),
         (TimestampMillis(_, _), DataType::Timestamp(_, _)) => Ok(()),
         (TimestampMicros(_, _), DataType::Timestamp(_, _)) => Ok(()),
@@ -591,10 +595,16 @@ fn schema_node_to_arrow_field(node: &SchemaNode) -> Result<Field> {
             name,
             primitive_type,
             nullable,
-            ..
+            format,
         } => {
             let data_type = primitive_type_to_arrow(primitive_type)?;
-            Ok(Field::new(name, data_type, *nullable))
+            let field = Field::new(name, data_type, *nullable);
+            let extended_field = if format.as_deref() == Some("uuid") {
+                field.with_extension_type(arrow_schema::extension::Uuid)
+            } else {
+                field
+            };
+            Ok(extended_field)
         }
         SchemaNode::List {
             name,
@@ -671,6 +681,7 @@ fn primitive_type_to_arrow(ptype: &crate::PrimitiveType) -> Result<DataType> {
         Date32 => DataType::Date32,
         TimeMillis => DataType::Time32(arrow_schema::TimeUnit::Millisecond),
         TimeMicros => DataType::Time64(arrow_schema::TimeUnit::Microsecond),
+        TimeNanos => DataType::Time64(arrow_schema::TimeUnit::Nanosecond),
         TimestampMillis(tz) => DataType::Timestamp(
             arrow_schema::TimeUnit::Millisecond,
             // PARQUET SPEC: ANY timezone (e.g., "+09:00", "America/New_York") means

data/ext/parquet-core/tests/arrow_conversion_tests.rs

@@ -99,7 +99,7 @@ fn test_decimal256_large_values() {
 
     // Verify roundtrip
     for i in 0..4 {
-        let value = arrow_to_parquet_value(array.as_ref(), i).unwrap();
+        let value = arrow_to_parquet_value(&field, array.as_ref(), i).unwrap();
         match (i, value) {
             (0, ParquetValue::Decimal256(v, _)) => assert_eq!(v, large_positive.clone()),
             (1, ParquetValue::Decimal256(v, _)) => assert_eq!(v, large_negative.clone()),
@@ -173,7 +173,7 @@ fn test_timestamp_with_timezone() {
 
     // Verify roundtrip preserves timezone
     for i in 0..3 {
-        let value = arrow_to_parquet_value(array.as_ref(), i).unwrap();
+        let value = arrow_to_parquet_value(&field, array.as_ref(), i).unwrap();
         match value {
             ParquetValue::TimestampMillis(_, Some(tz)) => {
                 assert_eq!(tz.as_ref(), "America/New_York");
@@ -209,7 +209,7 @@ fn test_nested_list_of_lists() {
     assert_eq!(array.len(), 1);
 
     // Verify roundtrip
-    let value = arrow_to_parquet_value(array.as_ref(), 0).unwrap();
+    let value = arrow_to_parquet_value(&outer_field, array.as_ref(), 0).unwrap();
     match value {
         ParquetValue::List(items) => assert_eq!(items.len(), 5),
         _ => panic!("Expected list"),
@@ -357,7 +357,7 @@ fn test_unsupported_arrow_types() {
     )
     .unwrap();
 
-    let result = arrow_to_parquet_value(&array, 0);
+    let result = arrow_to_parquet_value(&Field::new("int", DataType::Int32, false), &array, 0);
     assert!(result.is_err());
     assert!(result
         .unwrap_err()

data/ext/parquet-ruby-adapter/Cargo.toml

@@ -20,3 +20,4 @@ rb-sys = { version = "0.9", features = ["stable-api-compiled-fallback"] }
 tempfile = "^3.15"
 thiserror = "2.0"
 indexmap = "2.2"
+uuid = "*"

data/ext/parquet-ruby-adapter/src/converter.rs

@@ -41,27 +41,6 @@ impl RubyValueConverter {
             .map(|cache| cache.stats())
     }
 
-    /// Convert a Ruby value to ParquetValue with type hint
-    /// This is the primary conversion method that handles all Ruby types
-    pub fn to_parquet_with_type_hint(
-        &mut self,
-        value: Value,
-        type_hint: Option<&parquet_core::PrimitiveType>,
-    ) -> Result<ParquetValue> {
-        // Handle nil values
-        if value.is_nil() {
-            return Ok(ParquetValue::Null);
-        }
-
-        // If we have a type hint, use it to guide conversion
-        if let Some(hint) = type_hint {
-            return self.convert_with_type_hint(value, hint);
-        }
-
-        // Otherwise, infer type from Ruby value
-        self.infer_and_convert(value)
-    }
-
     /// Convert a Ruby value to ParquetValue with schema hint
     /// This handles both primitive and complex types
     pub fn to_parquet_with_schema_hint(
@@ -115,7 +94,7 @@ impl RubyValueConverter {
         use parquet_core::PrimitiveType::*;
 
         // Special handling for UUID format
-        if let (Binary, Some("uuid")) = (type_hint, format) {
+        if let (FixedLenByteArray(16), Some("uuid")) = (type_hint, format) {
             return self.convert_to_uuid_binary(value);
         }
 
@@ -156,6 +135,7 @@ impl RubyValueConverter {
             Date64 => self.convert_to_date64(value, None),
             TimeMillis => self.convert_to_time_millis(value),
             TimeMicros => self.convert_to_time_micros(value),
+            TimeNanos => self.convert_to_time_nanos(value),
             TimestampSecond(schema_tz) => {
                 self.convert_to_timestamp_second_with_tz(value, schema_tz.as_deref())
             }
@@ -484,32 +464,19 @@ impl RubyValueConverter {
 
         // Convert value to string
         let uuid_str: String = value
-            .funcall("to_s", ())
-            .and_then(TryConvert::try_convert)
+            .to_r_string()
+            .map_err(|e: MagnusError| {
+                ParquetError::Conversion(format!("Failed to convert to UUID string: {}", e))
+            })?
+            .to_string()
             .map_err(|e: MagnusError| {
                 ParquetError::Conversion(format!("Failed to convert to UUID string: {}", e))
             })?;
 
-        // Remove hyphens and validate length
-        let clean_uuid = uuid_str.replace('-', "");
-        if clean_uuid.len() != 32 {
-            return Err(ParquetError::Conversion(format!(
-                "Invalid UUID format: expected 32 hex characters (ignoring hyphens), got {}",
-                clean_uuid.len()
-            )));
-        }
-
-        // Parse hex string to bytes
-        let mut bytes = Vec::with_capacity(16);
-        for i in 0..16 {
-            let hex_byte = &clean_uuid[i * 2..i * 2 + 2];
-            let byte = u8::from_str_radix(hex_byte, 16).map_err(|_| {
-                ParquetError::Conversion(format!("Invalid hex character in UUID: {}", hex_byte))
-            })?;
-            bytes.push(byte);
-        }
-
-        Ok(ParquetValue::Bytes(bytes.into()))
+        let parsed = uuid::Uuid::parse_str(&uuid_str)
+            .map_err(|e| ParquetError::Conversion(format!("Failed to parse UUID: {}", e)))?;
+        let bytes = Bytes::copy_from_slice(parsed.as_bytes());
+        Ok(ParquetValue::Bytes(bytes))
     }
 
     fn convert_to_date32(&self, value: Value, date_format: Option<&str>) -> Result<ParquetValue> {
@@ -692,6 +659,38 @@ impl RubyValueConverter {
         )))
     }
 
+    fn convert_to_time_nanos(&self, value: Value) -> Result<ParquetValue> {
+        if value.is_nil() {
+            return Ok(ParquetValue::Null);
+        }
+
+        // Convert to microseconds since midnight
+        let ruby = Ruby::get()
+            .map_err(|_| ParquetError::Conversion("Failed to get Ruby runtime".to_string()))?;
+        if value.is_kind_of(ruby.class_time()) {
+            let hour: i64 = value
+                .funcall("hour", ())
+                .map_err(|e| ParquetError::Conversion(e.to_string()))?;
+            let min: i64 = value
+                .funcall("min", ())
+                .map_err(|e| ParquetError::Conversion(e.to_string()))?;
+            let sec: i64 = value
+                .funcall("sec", ())
+                .map_err(|e| ParquetError::Conversion(e.to_string()))?;
+            let nsec: i64 = value
+                .funcall("nsec", ())
+                .map_err(|e| ParquetError::Conversion(e.to_string()))?;
+
+            let nanos = (hour * 3600 + min * 60 + sec) * 1_000_000_000 + nsec;
+            return Ok(ParquetValue::TimeNanos(nanos));
+        }
+
+        Err(ParquetError::Conversion(format!(
+            "Cannot convert {} to time_micros",
+            value.class()
+        )))
+    }
+
     // Timestamp conversion methods that respect schema timezone
     fn convert_to_timestamp_second_with_tz(
         &self,
@@ -1399,21 +1398,8 @@ pub fn parquet_to_ruby(value: ParquetValue) -> Result<Value> {
         ParquetValue::Float32(OrderedFloat(f)) => Ok((f as f64).into_value_with(&ruby)),
         ParquetValue::Float64(OrderedFloat(f)) => Ok(f.into_value_with(&ruby)),
         ParquetValue::String(s) => Ok(s.into_value_with(&ruby)),
-        ParquetValue::Bytes(b) => {
-            // Check if this is a UUID (16 bytes)
-            if b.len() == 16 {
-                // Format as UUID string
-                let uuid_str = format!(
-                    "{:02x}{:02x}{:02x}{:02x}-{:02x}{:02x}-{:02x}{:02x}-{:02x}{:02x}-{:02x}{:02x}{:02x}{:02x}{:02x}{:02x}",
-                    b[0], b[1], b[2], b[3], b[4], b[5], b[6], b[7],
-                    b[8], b[9], b[10], b[11], b[12], b[13], b[14], b[15]
-                );
-                Ok(uuid_str.into_value_with(&ruby))
-            } else {
-                // Regular bytes - convert to string
-                Ok(ruby.str_from_slice(&b).as_value())
-            }
-        }
+        ParquetValue::Uuid(u) => Ok(u.to_string().into_value_with(&ruby)),
+        ParquetValue::Bytes(b) => Ok(ruby.enc_str_new(&b, ruby.ascii8bit_encoding()).as_value()),
         ParquetValue::Date32(days) => {
             // Convert days since epoch to Date object
             let _ = ruby.require("date");
@@ -1528,6 +1514,14 @@ pub fn parquet_to_ruby(value: ParquetValue) -> Result<Value> {
                 .map_err(|e| ParquetError::Conversion(e.to_string()))?;
             apply_timezone(time, &tz)
         }
+        ParquetValue::TimeNanos(nanos) => {
+            let time_class = ruby.class_time();
+            let secs = nanos / 1_000_000_000;
+            let nsec = nanos % 1_000_000_000;
+            time_class
+                .funcall("at", (secs, nsec, Symbol::new("nanosecond")))
+                .map_err(|e| ParquetError::Conversion(e.to_string()))
+        }
         ParquetValue::TimestampNanos(nanos, tz) => {
             let time_class = ruby.class_time();
             let secs = nanos / 1_000_000_000;

data/ext/parquet-ruby-adapter/src/reader.rs

@@ -2,6 +2,7 @@ use magnus::value::ReprValue;
 use magnus::{Error as MagnusError, IntoValue, RArray, RHash, Ruby, TryConvert, Value};
 use parquet_core::reader::Reader;
 
+use crate::StringCache;
 use crate::{
     converter::parquet_to_ruby,
     io::{RubyIOReader, ThreadSafeRubyIOReader},
@@ -101,6 +102,12 @@ pub fn each_row(
     })?;
     let mut row_count = 0u64;
 
+    let mut cache = StringCache::new(true);
+    let interned_column_names = column_names
+        .iter()
+        .map(|name| cache.intern(name.clone()))
+        .collect::<Vec<_>>();
+
     for row_result in row_iter {
         let row = row_result
             .map_err(|e| MagnusError::new(ruby.exception_runtime_error(), e.to_string()))?;
@@ -120,11 +127,11 @@ pub fn each_row(
             ParserResultType::Hash => {
                 let hash: RHash = ruby.hash_new();
                 for (idx, value) in row.into_iter().enumerate() {
-                    if idx < column_names.len() {
+                    if idx < interned_column_names.len() {
                         let ruby_value = parquet_to_ruby(value).map_err(|e| {
                             MagnusError::new(ruby.exception_runtime_error(), e.to_string())
                         })?;
-                        hash.aset(column_names[idx].as_str(), ruby_value)?;
+                        hash.aset(interned_column_names[idx].as_ref(), ruby_value)?;
                     }
                 }
                 hash.as_value()

data/ext/parquet-ruby-adapter/src/schema.rs

@@ -1,8 +1,9 @@
 use magnus::value::ReprValue;
 use magnus::{Error as MagnusError, RArray, RHash, Ruby, Symbol, TryConvert, Value};
-use parquet_core::{ParquetError, PrimitiveType, Result, Schema, SchemaNode};
+use parquet_core::{ParquetError, PrimitiveType, Schema, SchemaNode};
 
 use crate::utils::parse_string_or_symbol;
+use crate::RubyAdapterError;
 
 /// Ruby schema builder that converts Ruby hash/array representations to Parquet schemas
 pub struct RubySchemaBuilder;
@@ -13,18 +14,18 @@ impl RubySchemaBuilder {
     }
 
     /// Parse a Ruby schema definition (hash) into a SchemaNode
-    fn parse_schema_node(&self, name: String, schema_def: Value) -> Result<SchemaNode> {
+    fn parse_schema_node(
+        &self,
+        name: String,
+        schema_def: Value,
+    ) -> Result<SchemaNode, RubyAdapterError> {
         // If it's a Hash, parse it as a complex type
         if let Ok(hash) = <RHash as TryConvert>::try_convert(schema_def) {
             return self.parse_hash_schema_node(name, hash);
         }
 
         // Otherwise, try to parse as a simple type symbol
-        if let Ok(type_sym) = <Symbol as TryConvert>::try_convert(schema_def) {
-            let type_str = type_sym.name().map_err(|e: MagnusError| {
-                ParquetError::Conversion(format!("Failed to get symbol name: {}", e))
-            })?;
-
+        if let Ok(type_str) = schema_def.to_r_string()?.to_string() {
             // Check if it's a complex type with angle brackets
             if type_str.contains('<') {
                 return self.parse_complex_type_string(name, type_str.to_string(), true);
@@ -40,22 +41,24 @@ impl RubySchemaBuilder {
             });
         }
 
-        Err(ParquetError::Schema(format!(
+        Err(RubyAdapterError::InvalidInput(format!(
             "Expected Hash or Symbol for schema definition, got {}",
             schema_def.class()
         )))
     }
 
     /// Parse a Ruby hash schema node
-    fn parse_hash_schema_node(&self, name: String, hash: RHash) -> Result<SchemaNode> {
+    fn parse_hash_schema_node(
+        &self,
+        name: String,
+        hash: RHash,
+    ) -> Result<SchemaNode, RubyAdapterError> {
         // Get the type field
-        let type_sym: Symbol = hash
-            .fetch::<_, Symbol>(Symbol::new("type"))
+        let type_sym: Value = hash
+            .fetch::<_, Value>(Symbol::new("type"))
             .map_err(|e| ParquetError::Schema(format!("Schema missing 'type' field: {}", e)))?;
 
-        let type_str = type_sym.name().map_err(|e: MagnusError| {
-            ParquetError::Conversion(format!("Failed to get type name: {}", e))
-        })?;
+        let type_str = type_sym.to_r_string()?.to_string()?;
 
         // Get nullable field (default to true)
         let nullable = hash
@@ -142,6 +145,15 @@ impl RubySchemaBuilder {
 
             // Primitive types
             primitive_type => {
+                if format.as_deref() == Some("uuid") {
+                    return Ok(SchemaNode::Primitive {
+                        name,
+                        primitive_type: PrimitiveType::FixedLenByteArray(16),
+                        nullable,
+                        format,
+                    });
+                }
+
                 // Get precision and scale for decimal types
                 let precision = hash
                     .fetch::<_, Value>(Symbol::new("precision"))
@@ -196,7 +208,7 @@ impl RubySchemaBuilder {
         name: String,
         type_str: String,
         nullable: bool,
-    ) -> Result<SchemaNode> {
+    ) -> Result<SchemaNode, RubyAdapterError> {
         if type_str.starts_with("list<") && type_str.ends_with('>') {
             let inner_type = &type_str[5..type_str.len() - 1];
             let item_name = format!("{}_item", name);
@@ -229,7 +241,7 @@ impl RubySchemaBuilder {
             let inner = &type_str[4..type_str.len() - 1];
             let parts: Vec<&str> = inner.split(',').map(|s| s.trim()).collect();
             if parts.len() != 2 {
-                return Err(ParquetError::Schema(format!(
+                return Err(RubyAdapterError::InvalidInput(format!(
                     "Invalid map type: {}",
                     type_str
                 )));
@@ -255,7 +267,7 @@ impl RubySchemaBuilder {
                 }),
             })
         } else {
-            Err(ParquetError::Schema(format!(
+            Err(RubyAdapterError::InvalidInput(format!(
                 "Unknown complex type: {}",
                 type_str
             )))
@@ -263,7 +275,7 @@ impl RubySchemaBuilder {
     }
 
     /// Parse a field definition from a Ruby hash
-    fn parse_field_definition(&self, field_hash: RHash) -> Result<SchemaNode> {
+    fn parse_field_definition(&self, field_hash: RHash) -> Result<SchemaNode, RubyAdapterError> {
         let name: String = field_hash
             .fetch(Symbol::new("name"))
             .map_err(|e| ParquetError::Schema(format!("Field missing 'name': {}", e)))?;
@@ -274,7 +286,7 @@ impl RubySchemaBuilder {
             self.parse_schema_node(name, field_hash.as_value())
         } else {
             // This might be a simplified definition - look for known field patterns
-            Err(ParquetError::Schema(format!(
+            Err(RubyAdapterError::InvalidInput(format!(
                 "Field '{}' missing 'type' definition",
                 name
             )))
@@ -288,7 +300,7 @@ impl RubySchemaBuilder {
         precision: Option<u8>,
         scale: Option<i8>,
         timezone: Option<String>,
-    ) -> Result<PrimitiveType> {
+    ) -> Result<PrimitiveType, RubyAdapterError> {
         // Check if it's a decimal type with parentheses notation like "decimal(5,2)"
         if type_str.starts_with("decimal(") && type_str.ends_with(')') {
             let params = &type_str[8..type_str.len() - 1]; // Extract "5,2" from "decimal(5,2)"
@@ -324,6 +336,14 @@ impl RubySchemaBuilder {
             }
         }
 
+        if type_str.starts_with("fixed_len_byte_array(") && type_str.ends_with(')') {
+            let params = &type_str[20..type_str.len() - 1];
+            let len = params.parse::<i32>().map_err(|_| {
+                ParquetError::Schema(format!("Invalid fixed_len_byte_array length: {}", params))
+            })?;
+            return Ok(PrimitiveType::FixedLenByteArray(len));
+        }
+
         match type_str.as_str() {
             "boolean" | "bool" => Ok(PrimitiveType::Boolean),
             "int8" => Ok(PrimitiveType::Int8),
@@ -356,8 +376,9 @@ impl RubySchemaBuilder {
                 // PARQUET SPEC: timezone presence means UTC storage (isAdjustedToUTC = true)
                 Ok(PrimitiveType::TimestampNanos(timezone.map(Into::into)))
             }
-            "time32" | "time_millis" => Ok(PrimitiveType::TimeMillis),
-            "time64" | "time_micros" => Ok(PrimitiveType::TimeMicros),
+            "time_millis" => Ok(PrimitiveType::TimeMillis),
+            "time_micros" => Ok(PrimitiveType::TimeMicros),
+            "time_nanos" => Ok(PrimitiveType::TimeNanos),
             "decimal" => {
                 // Use provided precision/scale or defaults
                 let p = precision.unwrap_or(38);
@@ -380,7 +401,7 @@ impl RubySchemaBuilder {
                 let s = scale.unwrap_or(0);
                 Ok(PrimitiveType::Decimal256(p, s))
             }
-            _ => Err(ParquetError::Schema(format!(
+            _ => Err(RubyAdapterError::InvalidInput(format!(
                 "Unknown primitive type: {}",
                 type_str
             ))),
@@ -396,7 +417,7 @@ impl Default for RubySchemaBuilder {
 
 /// Wrapper functions for Ruby FFI since SchemaBuilderTrait requires Send + Sync
 /// and Ruby Value is not Send/Sync
-pub fn ruby_schema_to_parquet(schema_def: Value) -> Result<Schema> {
+pub fn ruby_schema_to_parquet(schema_def: Value) -> Result<Schema, RubyAdapterError> {
     let builder = RubySchemaBuilder::new();
 
     // The Ruby schema should be a hash with a root struct
@@ -430,7 +451,7 @@ pub fn ruby_schema_to_parquet(schema_def: Value) -> Result<Schema> {
         let mut unique_names = std::collections::HashSet::new();
         for name in &field_names {
             if !unique_names.insert(name) {
-                return Err(ParquetError::Schema(format!(
+                return Err(RubyAdapterError::InvalidInput(format!(
                     "Duplicate field names in root level schema: {:?}",
                     field_names
                 )));
@@ -443,7 +464,7 @@ pub fn ruby_schema_to_parquet(schema_def: Value) -> Result<Schema> {
             fields: field_nodes,
         }
     } else {
-        return Err(ParquetError::Schema(
+        return Err(RubyAdapterError::InvalidInput(
             "Schema must have 'type' or 'fields' key".to_string(),
         ));
     };
@@ -452,18 +473,18 @@ pub fn ruby_schema_to_parquet(schema_def: Value) -> Result<Schema> {
     parquet_core::SchemaBuilder::new()
         .with_root(root_node)
         .build()
-        .map_err(|e| ParquetError::Schema(e.to_string()))
+        .map_err(|e| RubyAdapterError::InvalidInput(e.to_string()))
 }
 
 /// Convert a Parquet schema back to Ruby representation
-pub fn parquet_schema_to_ruby(schema: &Schema) -> Result<Value> {
+pub fn parquet_schema_to_ruby(schema: &Schema) -> Result<Value, RubyAdapterError> {
     let ruby = Ruby::get()
         .map_err(|e| ParquetError::Conversion(format!("Failed to get Ruby runtime: {}", e)))?;
 
     schema_node_to_ruby(&schema.root, &ruby)
 }
 
-fn schema_node_to_ruby(node: &SchemaNode, _ruby: &Ruby) -> Result<Value> {
+fn schema_node_to_ruby(node: &SchemaNode, _ruby: &Ruby) -> Result<Value, RubyAdapterError> {
     let hash = RHash::new();
 
     match node {
@@ -552,6 +573,7 @@ fn schema_node_to_ruby(node: &SchemaNode, _ruby: &Ruby) -> Result<Value> {
                 PrimitiveType::TimestampNanos(_) => Symbol::new("timestamp_nanos"),
                 PrimitiveType::TimeMillis => Symbol::new("time_millis"),
                 PrimitiveType::TimeMicros => Symbol::new("time_micros"),
+                PrimitiveType::TimeNanos => Symbol::new("time_nanos"),
                 PrimitiveType::Decimal128(_, _) => Symbol::new("decimal128"),
                 PrimitiveType::Decimal256(_, _) => Symbol::new("decimal256"),
                 PrimitiveType::FixedLenByteArray(_) => Symbol::new("fixed_len_byte_array"),
@@ -597,7 +619,7 @@ fn schema_node_to_ruby(node: &SchemaNode, _ruby: &Ruby) -> Result<Value> {
 /// Convert old schema format to new format
 /// Old: [{ "column_name" => "type" }, ...]
 /// New: [{ name: "column_name", type: :type }, ...]
-pub fn convert_legacy_schema(ruby: &Ruby, schema: RArray) -> Result<RArray> {
+pub fn convert_legacy_schema(ruby: &Ruby, schema: RArray) -> Result<RArray, RubyAdapterError> {
     let new_schema = RArray::new();
 
     for item in schema.into_iter() {
@@ -630,7 +652,7 @@ pub fn convert_legacy_schema(ruby: &Ruby, schema: RArray) -> Result<RArray> {
         );
 
         if let Err(e) = process_result {
-            return Err(ParquetError::Schema(format!(
+            return Err(RubyAdapterError::InvalidInput(format!(
                 "Failed to process field: {}",
                 e
             )));
@@ -645,7 +667,7 @@ pub fn convert_legacy_schema(ruby: &Ruby, schema: RArray) -> Result<RArray> {
 }
 
 /// Check if schema is in new DSL format (hash with type: :struct)
-pub fn is_dsl_schema(ruby: &Ruby, schema_value: Value) -> Result<bool> {
+pub fn is_dsl_schema(ruby: &Ruby, schema_value: Value) -> Result<bool, RubyAdapterError> {
     if !schema_value.is_kind_of(ruby.class_hash()) {
         return Ok(false);
     }
@@ -678,7 +700,7 @@ pub fn process_schema_value(
     ruby: &Ruby,
     schema_value: Value,
     data_array: Option<&RArray>,
-) -> Result<Value> {
+) -> Result<Value, RubyAdapterError> {
     // Check if it's the new DSL format
     if is_dsl_schema(ruby, schema_value)? {
         // For DSL format, pass it directly to ruby_schema_to_parquet
@@ -716,7 +738,7 @@ pub fn process_schema_value(
                     convert_legacy_schema(ruby, array)?
                 }
             } else {
-                return Err(ParquetError::Schema(
+                return Err(RubyAdapterError::InvalidInput(
                     "schema array must contain hashes".to_string(),
                 ));
             }
@@ -733,13 +755,13 @@ pub fn process_schema_value(
                 ParquetError::Schema(format!("Failed to convert fields to array: {}", e))
             })?
         } else {
-            return Err(ParquetError::Schema(
+            return Err(RubyAdapterError::InvalidInput(
                 "schema hash must have 'fields' key or be in DSL format with 'type' key"
                     .to_string(),
             ));
         }
     } else {
-        return Err(ParquetError::Schema(
+        return Err(RubyAdapterError::InvalidInput(
             "schema must be nil, an array, or a hash".to_string(),
         ));
     };
@@ -748,7 +770,7 @@ pub fn process_schema_value(
     if schema_array.is_empty() {
         if let Some(data) = data_array {
             if data.is_empty() {
-                return Err(ParquetError::Schema(
+                return Err(RubyAdapterError::InvalidInput(
                     "Cannot infer schema from empty data".to_string(),
                 ));
             }
@@ -767,7 +789,7 @@ pub fn process_schema_value(
                     })?;
                 first_array.len()
             } else {
-                return Err(ParquetError::Schema(
+                return Err(RubyAdapterError::InvalidInput(
                     "First data item must be an array".to_string(),
                 ));
             };
@@ -793,7 +815,7 @@ pub fn process_schema_value(
 
             schema_array = new_schema;
         } else {
-            return Err(ParquetError::Schema(
+            return Err(RubyAdapterError::InvalidInput(
                 "Schema is required when data is not provided for inference".to_string(),
             ));
         }

data/ext/parquet-ruby-adapter/src/string_cache.rs

@@ -1,15 +1,15 @@
 use std::collections::HashMap;
-use std::sync::{Arc, Mutex};
+use std::sync::{Arc, LazyLock, Mutex};
 
 use magnus::RString;
 
+static STRING_CACHE: LazyLock<Mutex<HashMap<String, &'static str>>> =
+    LazyLock::new(|| Mutex::new(HashMap::with_capacity(100)));
+
 /// A cache for interning strings in the Ruby VM to reduce memory usage
 /// when there are many repeated strings
 #[derive(Debug)]
 pub struct StringCache {
-    /// The actual cache is shared behind an Arc<Mutex> to allow cloning
-    /// while maintaining a single global cache
-    cache: Arc<Mutex<HashMap<String, &'static str>>>,
     enabled: bool,
     hits: Arc<Mutex<usize>>,
     misses: Arc<Mutex<usize>>,
@@ -19,7 +19,6 @@ impl StringCache {
     /// Create a new string cache
     pub fn new(enabled: bool) -> Self {
         Self {
-            cache: Arc::new(Mutex::new(HashMap::new())),
             enabled,
             hits: Arc::new(Mutex::new(0)),
             misses: Arc::new(Mutex::new(0)),
@@ -36,9 +35,9 @@ impl StringCache {
 
         // Try to get or create the interned string
         let result = (|| -> Result<(), String> {
-            let mut cache = self.cache.lock().map_err(|e| e.to_string())?;
+            let mut cache = STRING_CACHE.lock().map_err(|e| e.to_string())?;
 
-            if cache.contains_key(&s) {
+            if cache.contains_key(s.as_str()) {
                 let mut hits = self.hits.lock().map_err(|e| e.to_string())?;
                 *hits += 1;
             } else {
@@ -65,7 +64,7 @@ impl StringCache {
 
     /// Get cache statistics
     pub fn stats(&self) -> CacheStats {
-        let cache_size = self.cache.lock().map(|c| c.len()).unwrap_or(0);
+        let cache_size = STRING_CACHE.lock().map(|c| c.len()).unwrap_or(0);
         let hits = self.hits.lock().map(|h| *h).unwrap_or(0);
         let misses = self.misses.lock().map(|m| *m).unwrap_or(0);
 
@@ -84,7 +83,7 @@ impl StringCache {
 
     /// Clear the cache
     pub fn clear(&mut self) {
-        if let Ok(mut cache) = self.cache.lock() {
+        if let Ok(mut cache) = STRING_CACHE.lock() {
             cache.clear();
         }
         if let Ok(mut hits) = self.hits.lock() {

data/ext/parquet-ruby-adapter/src/utils.rs

@@ -26,6 +26,7 @@ pub fn estimate_parquet_value_size(value: &ParquetValue) -> usize {
         ParquetValue::Float64(_) => 8,
         ParquetValue::String(s) => s.len() + 24, // String overhead
         ParquetValue::Bytes(b) => b.len() + 24,  // Vec overhead
+        ParquetValue::Uuid(_) => 16,
         ParquetValue::Date32(_) => 4,
         ParquetValue::Date64(_) => 8,
         ParquetValue::Decimal128(_, _) => 16 + 1, // value + scale
@@ -36,6 +37,7 @@ pub fn estimate_parquet_value_size(value: &ParquetValue) -> usize {
         ParquetValue::TimestampNanos(_, tz) => 8 + tz.as_ref().map_or(0, |s| s.len() + 24),
         ParquetValue::TimeMillis(_) => 4,
         ParquetValue::TimeMicros(_) => 8,
+        ParquetValue::TimeNanos(_) => 8,
         ParquetValue::List(items) => {
             24 + items.iter().map(estimate_parquet_value_size).sum::<usize>()
         }

data/lib/parquet/version.rb

@@ -1,3 +1,3 @@
 module Parquet
-  VERSION = "0.6.1"
+  VERSION = "0.6.2"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: parquet
 version: !ruby/object:Gem::Version
-  version: 0.6.1
+  version: 0.6.2
 platform: ruby
 authors:
 - Nathan Jaremko
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2025-07-03 00:00:00.000000000 Z
+date: 2025-07-04 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: rb_sys