AVRO-4024: [Rust] support nan/inf/-inf as float/double default

Co-authored-by: Xiangjin <xiangjin@risingwave-labs.com>
Signed-off-by: xxchan <xxchan22f@gmail.com>

Author: xxchan

lang/rust/avro/src/types.rs

@@ -908,6 +908,10 @@ impl Value {
             Value::Long(n) => Ok(Value::Float(n as f32)),
             Value::Float(x) => Ok(Value::Float(x)),
             Value::Double(x) => Ok(Value::Float(x as f32)),
+            Value::String(x) => match Self::parse_special_float(&x) {
+                Some(f) => Ok(Value::Float(f)),
+                None => Err(Error::GetFloat(ValueKind::String)),
+            },
             other => Err(Error::GetFloat(other.into())),
         }
     }
@@ -918,10 +922,25 @@ impl Value {
             Value::Long(n) => Ok(Value::Double(n as f64)),
             Value::Float(x) => Ok(Value::Double(f64::from(x))),
             Value::Double(x) => Ok(Value::Double(x)),
+            Value::String(x) => match Self::parse_special_float(&x) {
+                Some(f) => Ok(Value::Double(f.into())),
+                None => Err(Error::GetDouble(ValueKind::String)),
+            },
             other => Err(Error::GetDouble(other.into())),
         }
     }
 
+    /// IEEE 754 NaN and infinities are not valid JSON numbers.
+    /// So they are represented in JSON as strings.
+    fn parse_special_float(s: &str) -> Option<f32> {
+        match s.trim().to_ascii_lowercase().as_str() {
+            "nan" | "+nan" | "-nan" => Some(f32::NAN),
+            "inf" | "+inf" | "infinity" | "+infinity" => Some(f32::INFINITY),
+            "-inf" | "-infinity" => Some(f32::NEG_INFINITY),
+            _ => None,
+        }
+    }
+
     fn resolve_bytes(self) -> Result<Self, Error> {
         match self {
             Value::Bytes(bytes) => Ok(Value::Bytes(bytes)),

lang/rust/avro/tests/io.rs

@@ -107,6 +107,14 @@ fn default_value_examples() -> &'static Vec<(&'static str, &'static str, Value)>
             (r#""long""#, "5", Value::Long(5)),
             (r#""float""#, "1.1", Value::Float(1.1)),
             (r#""double""#, "1.1", Value::Double(1.1)),
+            (r#""float""#, r#""  +inf ""#, Value::Float(f32::INFINITY)),
+            (
+                r#""double""#,
+                r#""-Infinity""#,
+                Value::Double(f64::NEG_INFINITY),
+            ),
+            (r#""float""#, r#""-NAN""#, Value::Float(f32::NAN)),
+            (r#""double""#, r#""-NAN""#, Value::Double(f64::NAN)),
             (
                 r#"{"type": "fixed", "name": "F", "size": 2}"#,
                 r#""a""#,
@@ -312,10 +320,13 @@ fn test_default_value() -> TestResult {
             &mut Cursor::new(encoded),
             Some(&reader_schema),
         )?;
+        // For float/double, NaN != NaN, so we compare the Debug representation here.
         assert_eq!(
-            datum_read, datum_to_read,
+            format!("{datum_read:?}"),
+            format!("{datum_to_read:?}"),
             "{} -> {}",
-            *field_type, *default_json
+            *field_type,
+            *default_json
         );
     }