[datlechin]

Summary

• Export Database in the main menu writes a whole database or one schema, structure and rows, to a .sql or .sql.gz file in the order pg_dump and mysqldump use, so the engine's own client loads it too.
• Each driver owns its part: Connection::read_dump_catalog reads the catalog (server text where the engine keeps one) and SqlDialect::dump frames the file (mysqldump's SET header and DELIMITER ;;, GO batches, pg_dump's header). tablepro_core::dump::write_dump orders it by what each object names and writes rows as multi-row INSERTs through stream_read.
• The export result lists what the file leaves out and why: encrypted or CLR routines, remote tables, rows a row-level policy hides, and the like.
• PostgreSQL user collations and SQL Server column collations go in the file, so a database with another default reads the same.
• Import SQL File in the main menu runs a .sql or .sql.gz file of any size on an editor session of its own, split as the engine's client splits it (ScriptReader: DELIMITER, GO n, routine bodies, dollar quotes), with Stop at First Error or continue, one transaction on PostgreSQL, SQL Server, SQLite and DuckDB, progress with Stop, and a summary of failures by line.
• pg_dump plain files load: COPY ... FROM stdin rows run as INSERTs, \restrict and \unrestrict are passed over, other psql commands are refused by line.
• Safe Mode Read-Only turns Import SQL File off; on Confirm Writes the import dialog is the one question for the file.
• Redis, MongoDB, Cassandra and ScyllaDB get neither menu item.
• Core ADR 0011 records why the core writes and reads the file in process instead of calling pg_dump or mysqldump.
• SQL Server's Show CREATE TABLE keeps a column's identity seed and increment instead of writing IDENTITY(1,1).

	Light	Dark
Menu, before
Menu, after
Export Database
Import SQL File

Deliberately not fixed here

• DuckDB macros and views made after SET schema keep unqualified names in their stored text, and DuckDB does not record which schema bound them. Such a body fails to load from a file. Rewriting the names needs the binder's resolution, which DuckDB does not expose.
• Rows of PostgreSQL extension configuration tables (pg_extension_config_dump, such as pg_partman's) are not written; the extension makes the tables again, empty. Their row filters are SQL text the browse reader cannot take.
• Object order comes from the names each definition holds, with the server's dependency catalogs unread. A cycle of names is broken at names read outside a relation or call position.
• SQL Server temporal tables load as ordinary tables with their period times, and an indexed view loads without its indexes. The export lists each one.

Test plan

• dump_loads_back_into_an_empty_database on PostgreSQL, MySQL, MariaDB, SQL Server and ClickHouse (Docker) and SQLite and DuckDB (unit suite): the sample database is dumped, loaded into an empty database through the import, and dumped again to the same file.
• native_dump_loads_through_the_import: files written by pg_dump 18, mysqldump 8.4, mariadb-dump 11.4 and sqlite3 3.46 load, as one transaction and as written.
• Unit tests for the reader cut into pieces of every size, long lines, COPY rows, gzip, object order across kinds, source database and catalog qualifiers, INSERT batching, and the import dialog.