sql/pgwire: decode multi-byte "char" datums correctly

Historically, our `"char"` type has differed from Postgres. We allow
multi-byte UTF-8 characters, while in Postgres `"char"` values are
always a single byte. This inconsistency is difficult to change at this
point.

In #70942 we began truncating `"char"` values to a single byte when
decoding them in pgwire. This caused odd behavior with these values
because the rest of the system truncates `"char"` values to a single
UTF-8 character instead. This commit updates the pgwire decoding to make
it consistent with other SQL logic.

Fixes #149427

Release note (bug fix): Fixes a minor bug that caused inconsistent
behavior with the very rarely used `"char"` type (not to be confused
with `CHAR`).

Author: mgartner

pkg/sql/pgwire/pgwirebase/encoding.go

@@ -922,14 +922,20 @@ func DecodeDatum(
 		sv := strings.TrimRight(bs, " ")
 		return da.NewDString(tree.DString(sv)), nil
 	case oid.T_char:
-		sv := bs
-		// Always truncate to 1 byte, and handle the null byte specially.
-		if len(b) >= 1 {
-			if b[0] == 0 {
-				sv = ""
-			} else {
+		var sv string
+		if len(b) == 1 {
+			// Take a single byte as-is, even if it is not a valid UTF-8
+			// character. The null byte represents an empty string.
+			if b[0] != 0 {
 				sv = string(b[:1])
 			}
+		} else if len(b) > 1 {
+			// If there is more than one byte, decode the first UTF-8 character.
+			r, _ := utf8.DecodeRune(b)
+			if r == utf8.RuneError {
+				return nil, invalidUTF8Error
+			}
+			sv = string(r)
 		}
 		return da.NewDString(tree.DString(sv)), nil
 	case oid.T_name:

pkg/sql/pgwire/testdata/encodings.json

@@ -104,6 +104,13 @@
 		"TextAsBinary": [],
 		"Binary": [0]
 	},
+	{
+		"SQL": "'☃'::\"char\"",
+		"Oid": 18,
+		"Text": "☃",
+		"TextAsBinary": [226, 152, 131],
+		"Binary": [226, 152, 131]
+	},
 	{
 		"SQL": "''::text",
 		"Oid": 25,

pkg/sql/pgwire/testdata/pgtest/char

@@ -208,3 +208,70 @@ ReadyForQuery
 {"Type":"DataRow","Values":[{"binary":"00"}]}
 {"Type":"CommandComplete","CommandTag":"SELECT 1"}
 {"Type":"ReadyForQuery","TxStatus":"I"}
+
+# Regression test for #149427. "char" datums should be truncated to 1 UTF-8
+# character.
+send
+Query {"String": "CREATE TABLE t149427 (c \"char\" PRIMARY KEY)"}
+----
+
+until
+ReadyForQuery
+----
+{"Type":"CommandComplete","CommandTag":"CREATE TABLE"}
+{"Type":"ReadyForQuery","TxStatus":"I"}
+
+send
+Parse {"Query": "INSERT INTO t149427 VALUES ($1)", "Name": "i149427"}
+Describe {"Name": "i149427", "ObjectType": "S"}
+Bind {"ParameterFormatCodes": [0], "PreparedStatement": "i149427", "Parameters": [{"text":"☃"}]}
+Execute
+Sync
+----
+
+until
+ReadyForQuery
+----
+{"Type":"ParseComplete"}
+{"Type":"ParameterDescription","ParameterOIDs":[18]}
+{"Type":"NoData"}
+{"Type":"BindComplete"}
+{"Type":"CommandComplete","CommandTag":"INSERT 0 1"}
+{"Type":"ReadyForQuery","TxStatus":"I"}
+
+send
+Parse {"Query": "SELECT * FROM t149427", "Name": "s149427"}
+Describe {"Name": "s149427", "ObjectType": "S"}
+Bind {"PreparedStatement": "s149427", "ResultFormatCodes": [0]}
+Execute
+Sync
+----
+
+until
+ReadyForQuery
+----
+{"Type":"ParseComplete"}
+{"Type":"ParameterDescription","ParameterOIDs":null}
+{"Type":"RowDescription","Fields":[{"Name":"c","TableOID":105,"TableAttributeNumber":1,"DataTypeOID":18,"DataTypeSize":1,"TypeModifier":-1,"Format":0}]}
+{"Type":"BindComplete"}
+{"Type":"DataRow","Values":[{"text":"☃"}]}
+{"Type":"CommandComplete","CommandTag":"SELECT 1"}
+{"Type":"ReadyForQuery","TxStatus":"I"}
+
+send
+Parse {"Query": "DELETE FROM t149427 WHERE c = $1", "Name": "d149427"}
+Describe {"Name": "d149427", "ObjectType": "S"}
+Bind {"ParameterFormatCodes": [0], "PreparedStatement": "d149427", "Parameters": [{"text":"☃"}]}
+Execute
+Sync
+----
+
+until
+ReadyForQuery
+----
+{"Type":"ParseComplete"}
+{"Type":"ParameterDescription","ParameterOIDs":[25]}
+{"Type":"NoData"}
+{"Type":"BindComplete"}
+{"Type":"CommandComplete","CommandTag":"DELETE 1"}
+{"Type":"ReadyForQuery","TxStatus":"I"}