From 4aeb3982def067625e00a4086d23f74c793934f3 Mon Sep 17 00:00:00 2001
From: AshSgDe29071999 <AshSgDe29071999@users.noreply.github.com>
Date: Fri, 28 Aug 2026 19:31:01 +0530
Subject: [PATCH] Fix deallocation of failed digested prepares

When Prepare is called with name == sql (the stdlib path), the server-side
statement is stmt_<digest> but a Describe-phase failure stored the SQL text.
The next Prepare then Close'd a name that did not exist, leaked the digest
statement, and retried with 42P05.

Store the Parse name so cleanup targets the real server statement.

See #2640
---
 conn.go      |  3 ++-
 conn_test.go | 60 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 62 insertions(+), 1 deletion(-)

diff --git a/conn.go b/conn.go
index 229c380c5..48adb81ad 100644
--- a/conn.go
+++ b/conn.go
@@ -360,7 +360,8 @@ func (c *Conn) Prepare(ctx context.Context, name, sql string) (sd *pgconn.Statem
 	if err != nil {
 		var pErr *pgconn.PrepareError
 		if errors.As(err, &pErr) {
-			c.failedDescribeStatement = psKey
+			// psName is the identifier sent in Parse; psKey may be the SQL text.
+			c.failedDescribeStatement = psName
 		}
 		return nil, err
 	}
diff --git a/conn_test.go b/conn_test.go
index db3578b4b..b1025036a 100644
--- a/conn_test.go
+++ b/conn_test.go
@@ -532,6 +532,66 @@ func TestPrepareHandlesTimeoutBetweenParseAndDescribe(t *testing.T) {
 	require.NotNil(t, psd)
 }
 
+// https://github.com/jackc/pgx/issues/2640
+func TestPrepareDigestedNameDeallocatesFailedDescribe(t *testing.T) {
+	// Not parallel because it is a timing sensitive test.
+	//
+	// stdlib/database/sql calls Prepare(ctx, sql, sql), so the server-side
+	// name is stmt_<digest> while the client cache key is the SQL text.
+	// Cleanup after a Describe-phase failure must Close the digest name.
+
+	config, err := pgx.ParseConfig(os.Getenv("PGX_TEST_DATABASE"))
+	require.NoError(t, err)
+
+	var faultyConn *faultyconn.Conn
+	config.AfterNetConnect = func(ctx context.Context, config *pgconn.Config, conn net.Conn) (net.Conn, error) {
+		faultyConn = faultyconn.New(conn)
+		return faultyConn, nil
+	}
+
+	ctx := context.Background()
+	conn, err := pgx.ConnectConfig(ctx, config)
+	require.NoError(t, err)
+	defer closeConn(t, conn)
+	require.NotNil(t, faultyConn)
+
+	pgxtest.SkipCockroachDB(t, conn, "Induced error does not occur on CockroachDB")
+
+	_, err = conn.Exec(ctx, "set statement_timeout = '100ms'")
+	require.NoError(t, err)
+
+	faultyConn.HandleFrontendMessage = func(backendWriter io.Writer, msg pgproto3.FrontendMessage) error {
+		if _, ok := msg.(*pgproto3.Describe); ok {
+			time.Sleep(200 * time.Millisecond)
+		}
+		buf, err := msg.Encode(nil)
+		if err != nil {
+			return err
+		}
+		_, err = backendWriter.Write(buf)
+		return err
+	}
+
+	sql := "select $1::varchar"
+	psd, err := conn.Prepare(ctx, sql, sql)
+	var pgErr *pgconn.PgError
+	require.ErrorAs(t, err, &pgErr)
+	require.Equal(t, "57014", pgErr.Code)
+	require.Nil(t, psd)
+
+	faultyConn.HandleFrontendMessage = nil
+
+	_, err = conn.Exec(ctx, "set statement_timeout = default")
+	require.NoError(t, err)
+
+	// Re-preparing the same SQL must deallocate the leaked stmt_<digest>
+	// and succeed. Before the fix this returned 42P05 for the connection's
+	// remaining lifetime.
+	psd, err = conn.Prepare(ctx, sql, sql)
+	require.NoError(t, err)
+	require.NotNil(t, psd)
+}
+
 func TestPrepareBadSQLFailure(t *testing.T) {
 	t.Parallel()