From 2b8329f0f22f2fc1ec991d13e8798ad30606d5e7 Mon Sep 17 00:00:00 2001
From: zipg <geeky@foxmail.com>
Date: Fri, 25 Sep 2026 10:34:55 +0800
Subject: [PATCH] fix(export): declare utf8mb4 charset in MySQL export script

---
 crates/dbx-core/src/data/database_export.rs      | 16 ++++++++++++++++
 .../dbx-core/tests/live_mysql_database_export.rs | 10 ++++++++++
 2 files changed, 26 insertions(+)

diff --git a/crates/dbx-core/src/data/database_export.rs b/crates/dbx-core/src/data/database_export.rs
index 0a9374a60d..be98afb649 100644
--- a/crates/dbx-core/src/data/database_export.rs
+++ b/crates/dbx-core/src/data/database_export.rs
@@ -30,6 +30,15 @@ static EXPORT_CANCELLED: std::sync::LazyLock<RwLock<HashSet<String>>> =
 const EXPORT_CANCELLED_ERROR: &str = "Export cancelled";
 const EXPORT_CANCEL_POLL_INTERVAL: Duration = Duration::from_millis(50);
 
+/// MySQL scripts written by the exporter are always UTF-8: values are decoded off the
+/// wire and reformatted into Rust strings before they are written, so the file encoding
+/// never follows the connection charset. Declaring that encoding in the script keeps the
+/// importing client's default out of the decoding decision -- a `latin1` `mysql` CLI (a
+/// docker entrypoint initializing a container, for example) would otherwise re-encode
+/// every non-ASCII value into mojibake (#10242). `mysqldump` writes the same statement
+/// for the same reason.
+const MYSQL_EXPORT_CHARSET_STATEMENT: &str = "SET NAMES utf8mb4;";
+
 pub fn database_export_client_session_id(export_id: &str) -> String {
     task_client_session_id("database-export", export_id)
 }
@@ -3020,6 +3029,13 @@ async fn export_database_sql_core_inner(
     writeln!(file, "-- Generated by DBX").map_err(|e| format!("Failed to write file: {e}"))?;
     writeln!(file).map_err(|e| format!("Failed to write file: {e}"))?;
 
+    // Declare the script's encoding before the first statement that can carry non-ASCII
+    // text (the database name is written by the preamble right below, and table, column
+    // and DEFAULT values follow).
+    if matches!(db_type, DatabaseType::Mysql) {
+        writeln!(file, "{MYSQL_EXPORT_CHARSET_STATEMENT}").map_err(|e| format!("Failed to write file: {e}"))?;
+    }
+
     if let Some(preamble) = create_database_preamble {
         writeln!(file, "-- Database setup").map_err(|e| format!("Failed to write file: {e}"))?;
         writeln!(file, "{preamble}").map_err(|e| format!("Failed to write file: {e}"))?;
diff --git a/crates/dbx-core/tests/live_mysql_database_export.rs b/crates/dbx-core/tests/live_mysql_database_export.rs
index 3e33fe7db8..52a6dcc7e8 100644
--- a/crates/dbx-core/tests/live_mysql_database_export.rs
+++ b/crates/dbx-core/tests/live_mysql_database_export.rs
@@ -404,6 +404,16 @@ async fn live_mysql_database_export_creates_missing_destination_directory() {
     let exported = std::fs::read_to_string(&file_path).unwrap();
     assert!(exported.contains("'alpha'"), "exported SQL should contain the seeded row");
 
+    // #10242: the script must declare its own encoding before the first statement that can
+    // carry non-ASCII text, otherwise an importing client whose default charset is not
+    // utf8mb4 (a `latin1` mysql CLI in a docker entrypoint, for example) re-encodes every
+    // non-ASCII value the exporter wrote as UTF-8 into mojibake.
+    let charset = exported.find("SET NAMES utf8mb4;").expect("MySQL exports must declare their encoding");
+    let create_database = exported.find("CREATE DATABASE").expect("the CREATE DATABASE preamble should be exported");
+    let create_table = exported.find("CREATE TABLE").expect("the table DDL should be exported");
+    assert!(charset < create_database, "SET NAMES must precede the CREATE DATABASE preamble");
+    assert!(charset < create_table, "SET NAMES must precede the table DDL");
+
     std::fs::remove_dir_all(dir).unwrap();
 }