Chat2DB Files

Chat2DB Community 5.3.7 adds in-app desktop updates for the packaged macOS, Windows and Linux builds; basic configuration and preview for Excel, JSON and SQL imports, plus an ultra-fast CSV import mode; an Update Time column in the table list; Oracle EXPLAIN PLAN results; and 28 user-visible improvements and fixes across the query console, task center, dashboards, AI history and connection handling.

Added (5)

1. In-app desktop updates: the packaged macOS, Windows and Linux builds discover and install new versions, and a package is installed only after its signed manifest is verified. Settings → About gained a "Receive beta versions" switch. A downloaded update is kept when the application restarts or checks for updates again, so it installs without downloading again and also works offline; on macOS the update helper runs independently of the application so the update survives the application closing; on some machines the previous version is restored and reopened when an update does not start healthily. by @haizhi03 @openai0229 in [#2698] [#2917] [#2923] [#2931] [#2943] [#2947] [#2949]
2. Basic configuration for Excel, JSON and SQL imports: non-CSV imports previously had neither configuration nor preview. Excel/XLS now select the worksheet, header row, first and last data row and the column range, treat empty fields as NULL, and set the date, time and decimal formats for text cells; JSON selects the encoding, the data structure (array of objects, single object, JSON Lines) and the data node path, and can skip individual nested fields; SQL supports a file encoding (detected automatically or given explicitly, such as GBK). The preview and the import run use the same options, so the mapping shows what is imported. by @openai0229 in [#2921]
3. Ultra-fast CSV import mode: after the file preview an "Ultra-fast mode" can be enabled, which imports in parallel batches over a dedicated connection and adapts concurrency and batch size to the measured throughput; a failure or cancellation cancels the registered statements and waits for every worker to stop, and the standard mode stays the default. by @liushikuan63 in [#2908]
4. Update Time column in the table list: shown after Create Time, left empty when unknown, translated in all five languages. by @openai0229 in [#2915]
5. Oracle EXPLAIN PLAN shows the execution plan: running it in the Oracle SQL console now returns the plan instead of reporting success with no result: the statement runs under a generated statement id, the plan is read back from PLAN_TABLE through DBMS_XPLAN.DISPLAY and returned as a result set. A missing PLAN_TABLE reports an actionable message, EXPLAIN PLAN ... INTO <custom table> still goes to the database, and requesting the plan of a statement that already is an explain command no longer nests two EXPLAIN PLAN clauses. by @openai0229 in [#2945]

Optimization (3)

1. More reliable task center polling and paging: records are merged by id and update time, terminal states and already loaded history are kept, and responses that arrive after a task was deleted or the panel was closed are dropped; a failed load ends the loading state, reports an error and can be retried, and tasks cancelled abnormally no longer stay in RUNNING. by @HandSonic in [#2795]
2. MCP token reset is serialized: a reset shows a busy state, repeated clicks no longer create several tokens, and a stale callback no longer overwrites the newest token. by @HandSonic in [#2845]
3. Editable results for single-table subqueries: queries such as SELECT * FROM (SELECT * FROM APP.EVENT_LOG ORDER BY CREATE_TIME DESC) WHERE ROWNUM <= 100 now resolve the physical table key, so the result set can be edited directly. by @openai0229 in [#2916]

Fixed (20)

1. Redis key updates validate the type: a missing or unknown old or new key type reports an error instead of falling back to string commands that delete and recreate the key. by @Aias00 in [#2437]
2. CockroachDB editing without a primary key: updates and deletes without a visible primary key use the native LIMIT 1 form and keep the original row predicate, instead of failing with PostgreSQL's ctid. by @Aias00 in [#2484]
3. Redshift and CockroachDB database creation: each dialect generates a concise CREATE DATABASE, Redshift hides the unsupported database comment, and creating a database or schema stops after the first failing statement instead of overwriting an existing database comment with a duplicate-name error. by @Aias00 in [#2486]
4. Informix table operations: renaming a table, changing a column and reading the execution plan are generated by the Informix plug-in, owner references are correct, and character type aliases are no longer collapsed into a single character. by @Aias00 in [#2488]
5. DuckDB paging: pagination uses LIMIT … OFFSET instead of the comma syntax DuckDB does not support. by @Aias00 in [#2490]
6. JSON import null semantics: a missing field and an explicit null are distinguished, both are written as SQL NULL, and non-object records are rejected before an INSERT is generated. by @HandSonic in [#2809]
7. Clearer import failures: a failed CSV, XLSX, XLS, JSON or SQL import names the setting that caused it (encoding, data node, file row or column, duplicated source field, column range, row range) or reports the database error class with its SQLState and error code, instead of a generic failure. by @openai0229 in [#2921]
8. Exports no longer overwrite a file with the same name: an occupied target name falls back to a numbered available name, and a failed copy is rolled back and cleaned up on the next start. by @HandSonic in [#2800]
9. Dashboard list: an oversized page number returns an empty page instead of an index error, the dropdown keeps loading when a dashboard has more than 20 items, and a failed request shows a retryable error state while keeping the data that is already loaded. by @HandSonic in [#2812]
10. Out-of-range paging: console records and table data compute offsets with long arithmetic, so a very large page number returns an empty page instead of wrapping around or throwing IndexOutOfBoundsException. by @HandSonic in [#2815] [#2824]
11. Corrupted caches recover: a corrupted scalar JSON entry no longer makes the key unusable for the rest of its TTL; only the damaged key is removed and the cached value is rebuilt. by @HandSonic in [#2828]
12. A failed save no longer leaves memory and disk out of sync: data source order, namespace position, ER-diagram position and small data writes are persisted before memory is updated, so a failed write is not shown as saved and does not revert to the old state after a refresh. by @HandSonic @Aias00 in [#2480] [#2830] [#2840] [#2846]
13. More reliable AI conversation history: sessions and message history are written with a temporary file and an atomic replace, so an interrupted write no longer destroys the last usable history, and a failed message deletion rolls back the session index instead of losing the owning record. by @HandSonic in [#2843] [#2841]
14. Line endings when copying results on Windows: copied result rows use CRLF, so pasting into native programs such as SAP GUI no longer merges the rows or shows #. by @openai0229 in [#2909]
15. SSH connection test failures: a failed request or a desktop timeout reports a clear message instead of leaving the test loading forever. by @Aias00 in [#2548]
16. Cursor flickering in the Windows editor: moving the mouse inside the SQL editor no longer switches between the arrow, I-beam and zoom cursors, and macOS and Linux keep their previous behaviour. by @openai0229 in [#2922]
17. Duplicated connections keep their fields: duplicating a data source keeps the BigQuery and service name fields, deep-copies the SSH, driver and extension configuration, and resets runtime handles. by @HandSonic in [#2793]
18. Local data source filtering: alias search and type filters apply before paging and counting, PRIVATE and SHARED match case-insensitively, and a missing type in older data is treated as private. by @HandSonic in [#2820]
19. Clean process exit: the cookie cleanup thread is a daemon thread, so exiting or stopping the service no longer blocks process termination. by @Aias00 in [#2522]
20. Redacted IPC debug logs: passwords, SSH passphrases and JDBC connection strings in desktop request and response debug logs are no longer written in clear text. by @Aias00 in [#2473]

Full Changelog: https://github.com/OtterMind/Chat2DB/compare/v5.3.6...v5.3.7