Add xml '...' TypedString support for PostgreSQL (apache#2299)

Co-authored-by: Donovan Schönknecht <don@tpyo.net>
Co-authored-by: Minjun Kim <48622976+funcpp@users.noreply.github.com>

Author: 3 people

src/dialect/generic.rs

@@ -300,4 +300,8 @@ impl Dialect for GenericDialect {
     fn supports_select_item_multi_column_alias(&self) -> bool {
         true
     }
+
+    fn supports_xml_expressions(&self) -> bool {
+        true
+    }
 }

src/dialect/mod.rs

@@ -1702,6 +1702,17 @@ pub trait Dialect: Debug + Any {
     fn supports_select_item_multi_column_alias(&self) -> bool {
         false
     }
+
+    /// Returns true if the dialect supports XML-related expressions
+    /// such as `xml '<foo/>'` typed strings, XML functions like
+    /// `XMLCONCAT`, `XMLELEMENT`, etc.
+    ///
+    /// When this returns false, `xml` is treated as a regular identifier.
+    ///
+    /// [PostgreSQL](https://www.postgresql.org/docs/current/functions-xml.html)
+    fn supports_xml_expressions(&self) -> bool {
+        false
+    }
 }
 
 /// Operators for which precedence must be defined.

src/dialect/postgresql.rs

@@ -310,4 +310,8 @@ impl Dialect for PostgreSqlDialect {
     fn supports_comma_separated_trim(&self) -> bool {
         true
     }
+
+    fn supports_xml_expressions(&self) -> bool {
+        true
+    }
 }

src/parser/mod.rs

@@ -1687,6 +1687,16 @@ impl<'a> Parser<'a> {
         }
     }
 
+    /// Returns true if the given [ObjectName] is a single unquoted
+    /// identifier matching `expected` (case-insensitive).
+    fn is_simple_unquoted_object_name(name: &ObjectName, expected: &str) -> bool {
+        if let [ObjectNamePart::Identifier(ident)] = name.0.as_slice() {
+            ident.quote_style.is_none() && ident.value.eq_ignore_ascii_case(expected)
+        } else {
+            false
+        }
+    }
+
     /// Parse an expression prefix.
     pub fn parse_prefix(&mut self) -> Result<Expr, ParserError> {
         // allow the dialect to override prefix parsing
@@ -1720,7 +1730,21 @@ impl<'a> Parser<'a> {
                 // so given `NOT 'a' LIKE 'b'`, we'd accept `NOT` as a possible custom data type
                 // name, resulting in `NOT 'a'` being recognized as a `TypedString` instead of
                 // an unary negation `NOT ('a' LIKE 'b')`. To solve this, we don't accept the
-                // `type 'string'` syntax for the custom data types at all.
+                // `type 'string'` syntax for the custom data types at all ...
+                //
+                // ... with the exception of `xml '...'` on dialects that support XML
+                // expressions, which is a valid PostgreSQL typed string literal.
+                DataType::Custom(ref name, ref modifiers)
+                    if modifiers.is_empty()
+                        && Self::is_simple_unquoted_object_name(name, "xml")
+                        && parser.dialect.supports_xml_expressions() =>
+                {
+                    Ok(Expr::TypedString(TypedString {
+                        data_type: DataType::Custom(name.clone(), modifiers.clone()),
+                        value: parser.parse_value()?,
+                        uses_odbc_syntax: false,
+                    }))
+                }
                 DataType::Custom(..) => parser_err!("dummy", loc),
                 // MySQL supports using the `BINARY` keyword as a cast to binary type.
                 DataType::Binary(..) if self.dialect.supports_binary_kw_as_cast() => {

tests/sqlparser_common.rs

@@ -18771,3 +18771,11 @@ fn parse_select_item_multi_column_alias() {
             .is_err()
     );
 }
+
+#[test]
+fn parse_non_pg_dialects_keep_xml_names_as_regular_identifiers() {
+    // On dialects that do NOT support XML expressions, bare `xml` should
+    // be treated as a regular column identifier, not a typed-string prefix.
+    let dialects = all_dialects_except(|d| d.supports_xml_expressions());
+    dialects.verified_only_select("SELECT xml FROM t");
+}

tests/sqlparser_postgres.rs

@@ -3750,6 +3750,25 @@ fn parse_on_commit() {
     pg_and_generic().verified_stmt("CREATE TEMPORARY TABLE table (COL INT) ON COMMIT DROP");
 }
 
+#[test]
+fn parse_xml_typed_string() {
+    // xml '...' should parse as a TypedString on PostgreSQL and Generic
+    let sql = "SELECT xml '<foo/>'";
+    let select = pg_and_generic().verified_only_select(sql);
+    match expr_from_projection(&select.projection[0]) {
+        Expr::TypedString(TypedString {
+            data_type: DataType::Custom(name, modifiers),
+            value,
+            uses_odbc_syntax: false,
+        }) => {
+            assert_eq!(name.to_string(), "xml");
+            assert!(modifiers.is_empty());
+            assert_eq!(value.value, Value::SingleQuotedString("<foo/>".to_string()));
+        }
+        other => panic!("Expected TypedString, got: {other:?}"),
+    }
+}
+
 fn pg() -> TestedDialects {
     TestedDialects::new(vec![Box::new(PostgreSqlDialect {})])
 }