mock

Mock backend

A fake backend for running the UI in a browser: just dev-web, then open http://localhost:1420/?mock. Every backend command is answered by a handler in handlers/, keyed by its Rust name; __tests__/coverage.test.ts fails when one is missing. See the ui-preview skill for driving it with Playwright.

Scenarios

The starting point, as ?mock=<scenario>. One at a time.

URL	Starts with
?mock	the saved profiles, nothing connected
?mock=empty	a first run: no saved profiles
?mock=connected	"Local dev" connected, for the editor, grid and schema tree
?mock=lost	connected, then the server goes: see below
?mock=servers	the default profiles plus four more, one per connect outcome

?mock=lost connects "Local dev" and, four seconds later, stops answering. The health check fails every three seconds and reports each failure, so the tab turns red and the status bar counts "Lost — retrying (n)". Queries fail with 2013 (QueryError from execute_query, text from the schema reads), a ping counts as one more failed check, and Disconnect still works.

?mock=servers adds:

Profile	Connecting to it
Production (read-write)	connects; production that takes writes, shared with agents in full
MariaDB 11	connects to MariaDB 11.4
Wrong password	rejected by the server with 1045, after the backend's credentials hint
Unreachable (times out)	no answer: fails after two 2s connect timeouts, with the backend's wording

Test Connection reports the same outcomes inline. A new profile in the dialog pointing at db.qa.example.internal or 10.20.30.40 behaves the same way.

Switches

Separate URL parameters, which combine with any scenario and with each other: ?mock=connected&fail=get_foreign_keys.

Parameter	Effect
fail=<cmd>,<cmd>	those commands reject with the error shape the real one has; get_* matches by prefix
update=available	the updater offers 1.5.0, which downloads in about five seconds
update=downloading	the same release, downloading over two minutes, to look at the progress
update=download-error	the download breaks halfway
update=error	the update check fails
keyring=missing	no OS credential store: "Passwords not saved"
startup=problems	the connection and history stores could not be opened at startup
agent=uninstalled	no agent CLI installed: the panel shows the install commands
agent=no-tools	sessions start without the database tools

The update chip appears after the automatic check; click it to download.

fail= takes the Rust command names from src/lib/bindings.ts (get_foreign_keys, save_connection_profile) or a plugin command (plugin:updater|check). execute_query, apply_grid_edits and sqlite_execute reject with a QueryError; every other command rejects with the string its CoreError renders as. The console names each injected failure.

Built-in behaviour

Some states need no switch, only the right click:

• Connect: "Staging" takes a second and a half; "MariaDB sandbox" is refused; "Production (read-only)" refuses writes.
• Ports: save_connection_profile and test_connection reject a port that is not a u16 (70000, -1, 3306.5) with the message Tauri gives when the argument fails to deserialize. Port 0 is a u16 and is accepted, as in the app; connecting to it is refused.
• Query errors: a statement containing syntax_error or starting with a misspelt verb fails with 1064; an unknown table with 1146; SELECT SLEEP(n) waits up to 5s and Cancel interrupts it.
• Schema writes: CREATE/ALTER/RENAME/DROP TABLE, CREATE/DROP INDEX and CREATE/DROP DATABASE change the fixture, so the refresh after the designer's Apply shows the change. Each is all-or-nothing, and the usual mistakes fail with the server's code: CREATE TABLE orders () with 1064, an existing table 1050, a duplicate column 1060, a duplicate key 1061. A reload starts from the fixture again.
• Data: 130 history entries (three pages); shop has routines, a trigger and events in each state, analytics a partitioned table; the SQLite fixture has a view.
• Users and roles: mysql.user, the role tables and SHOW GRANTS answer from a list of eleven accounts that CREATE/ALTER/RENAME/DROP USER, CREATE ROLE, GRANT and REVOKE change. On MariaDB the Users tab's full query fails with 1054 and it falls back, as on a real MariaDB.
• Agent turns: in an agent session, the message picks the turn. "delete" or "update" asks to change Local dev; "production" or "staging" asks on that environment; "schema", "alter", "index" or "column" asks for a schema change (with "production", on production); "both" sends two approvals at once, one from another client; "rewrite", "propose" or "improve" proposes an edit to the active tab; "draft" or "open" opens a draft tab. Anything else is a read. GitHub Copilot fails its first turn as not signed in, and an approval from an endpoint client arrives six seconds into the first session.
• Backup and restore run for about three seconds of real time. The first progress event arrives after 450 ms, as the backend's does after it has read the table list or the first chunk, so the dialog's state before any progress can be seen.