fix: Resolve mismatch with Thrift compact protocol

The [Thrift compact
protocol](https://github.com/apache/thrift/blob/master/doc/specs/thrift-compact-protocol.md)
is used for Parquet file metadata.
[parquet-format-safe](https://github.com/jorgecarleitao/parquet-format-safe)
and other Rust implementations of the protocol eagerly read
string/binary fields as UTF-8.

However, based on the protocol which states that

> Strings are first encoded to UTF-8, and then send as binary

it cannot be known upfront, without using the schema to disambiguate the
field type, whether a field is a string or a binary. This means that
when the field is actually a binary field and contains invalid UTF-8,
Rust libraries error out when reading the field with `File out of
specification: Invalid thrift: bad data`.

To fix this, we patch the protocol implementation to correctly interpret
string/binary fields as binary.

Author: desmondcheongzx

src/parquet_format.rs

@@ -133,7 +133,7 @@ impl Error {
             name: "TApplicationException".to_owned(),
         })?;
 
-        let message_field = TFieldIdentifier::new("message", TType::String, 1);
+        let message_field = TFieldIdentifier::new("message", TType::Binary, 1);
         let type_field = TFieldIdentifier::new("type", TType::I32, 2);
 
         o.write_field_begin(&message_field)?;

src/thrift/errors.rs

@@ -324,7 +324,7 @@ pub(super) fn u8_to_type(b: u8) -> Result<TType> {
         0x05 => Ok(TType::I32),
         0x06 => Ok(TType::I64),
         0x07 => Ok(TType::Double),
-        0x08 => Ok(TType::String),
+        0x08 => Ok(TType::Binary),
         0x09 => Ok(TType::List),
         0x0A => Ok(TType::Set),
         0x0B => Ok(TType::Map),

src/thrift/protocol/compact.rs

@@ -263,7 +263,7 @@ pub(super) fn type_to_u8(field_type: TType) -> u8 {
         TType::I32 => 0x05,
         TType::I64 => 0x06,
         TType::Double => 0x07,
-        TType::String => 0x08,
+        TType::Binary => 0x08,
         TType::List => 0x09,
         TType::Set => 0x0A,
         TType::Map => 0x0B,

src/thrift/protocol/compact_write.rs

@@ -162,7 +162,7 @@ pub trait TInputProtocol: Sized {
             TType::I32 => self.read_i32().map(|_| ()),
             TType::I64 => self.read_i64().map(|_| ()),
             TType::Double => self.read_double().map(|_| ()),
-            TType::String => self.read_string().map(|_| ()),
+            TType::Binary => self.read_bytes().map(|_| ()),
             TType::Struct => {
                 self.read_struct_begin()?;
                 loop {
@@ -703,8 +703,8 @@ pub enum TType {
     I32,
     /// Signed 64-bit int.
     I64,
-    /// UTF-8 string.
-    String,
+    /// Binary or UTF-8 string. Based on the Thrift compact protocol: https://github.com/apache/thrift/blob/master/doc/specs/thrift-compact-protocol.md
+    Binary,
     /// UTF-7 string. *Unsupported*.
     Utf7,
     /// Thrift struct.
@@ -732,7 +732,7 @@ impl Display for TType {
             TType::I16 => write!(f, "i16"),
             TType::I32 => write!(f, "i32"),
             TType::I64 => write!(f, "i64"),
-            TType::String => write!(f, "string"),
+            TType::Binary => write!(f, "binary"),
             TType::Utf7 => write!(f, "UTF7"),
             TType::Struct => write!(f, "struct"),
             TType::Map => write!(f, "map"),

src/thrift/protocol/mod.rs

@@ -148,7 +148,7 @@ pub trait TInputStreamProtocol: Send + Sized {
             TType::I32 => self.read_i32().await.map(|_| ()),
             TType::I64 => self.read_i64().await.map(|_| ()),
             TType::Double => self.read_double().await.map(|_| ()),
-            TType::String => self.read_string().await.map(|_| ()),
+            TType::Binary => self.read_bytes().await.map(|_| ()),
             TType::Struct => {
                 self.read_struct_begin().await?;
                 loop {