SET search_path in PostgreSQL is not transactional — it survives rollback
and persists on the connection. Previously, connections returned to the pool
with a modified search_path, causing subsequent queries on the same
connection to run against the wrong schema. Now RESET search_path is always
called after schema-scoped queries and transactions.
- Use client.transaction() for safer transaction handling
- Add prepare_cached for all system queries to enable statement caching
- Use query_raw streaming to avoid buffering large result sets
- Add batch_execute support for bulk DDL scripts
- Add COPY protocol support (copy_in / copy_out) for fast data transfer
- Add pg_quote_ident to prevent SQL injection in SET search_path
- Increase connection pool max_size from 5 to 10
- Fix list_indexes bounds check for expression indexes
Extended is_connection_error to recognize Oracle-specific JDBC error
patterns (idle time exceeded, end-of-file on communication channel,
not connected) so the auto-reconnect mechanism triggers for stale agent
connections after long idle periods.
DuckDB was not recognized as a schema-aware database, so only the
default "main" schema was shown in the object tree and schema selector.
Added DuckDB to schema-aware type lists, implemented schema listing via
information_schema.schemata, and replaced hardcoded table_schema = 'main'
with parameterized queries.
Treat blank-line-separated statements as part of the previous statement when the cursor sits on the semicolon boundary.\nKeep same-line cursor navigation moving to the next statement.\nAdd regression coverage for the cursor boundary cases described in issue #427.
Add mysql_async with client_ed25519 support for MySQL and MariaDB connections.\nSwitch MySQL transaction control and schema metadata reads away from sqlx MySQL APIs.\nPreserve connection compatibility by stripping unsupported URL params and handling nullable metadata safely.
The recent fix to respect explicit JDBC URLs in the Oracle agents exposed a
default mismatch: the Rust side generated SID-format URLs when the config
field was missing, but the old Java fallback always used service_name format.
Switch the default and retry logic so existing connections without an explicit
oracle_connection_type keep working.
Object browser now defaults to showing tables when tables exist.
Table/view context menus match sidebar tree with structure editing,
export, import, diagram, data compare, truncate, empty, duplicate,
and copy name support.
Keep the server session timezone for standard MySQL connections unless the connection URL explicitly provides a timezone override. This avoids TIMESTAMP shifts during SQL import while preserving user-specified timezone parameters.
parse_delimiter_command used str slicing (line[..10]) which panics on
non-char boundaries when the buffer contains multi-byte UTF-8 characters.
Also fixed DELIMITER detection to check only the last line of the buffer
instead of the entire accumulated content.
- Add `ai_list_models` Tauri command and core support for fetching available models from OpenAI-compatible, Claude, and Ollama providers
- Replace manual model ID input with SearchableSelect in EditorSettingsDialog, including refresh button and loading state
- Add i18n translations for model list UI (en, es, zh-CN)
- Add tests for settings dialog layout and web driver management
BEGIN TRANSACTION / COMMIT / ROLLBACK must not go through sp_executesql
(client.execute) because SQL Server raises error 266 when @@TRANCOUNT
changes inside sp_executesql. Route them through simple_query instead.
Add a 512MB maximum response size limit to read_agent_line to prevent
unbounded memory allocation that could cause the process to SIGABRT
when the agent returns an unexpectedly large response.
Ref: #345
t8y2