Testing#

Run the checks below from the project root with the development environment active. Use isolated state, accounts and SSH services for tests. Each check qualifies its own boundary; a browser fixture does not establish that a remote operation works on an installed machine.

Source and unit checks#

bash tools/check.sh
python -m pip_audit -r requirements.lock --disable-pip --no-deps
python -m build --no-isolation

The source check detects common credential formats, private infrastructure text, missing licence identifiers and oversized source files. Its negative fixtures must fail when restricted content is present. A clean source scan is not proof that all possible secrets are absent.

Python tests cover the API, authentication, token revocation, batched requests, SSH trust and resource schema. Managed-job checks cover distinct 1-node and 64-node batches, durable deduplication, revoked grants, unknown outcomes, resource controls, cancellation, output bounds and CLI recovery. Launcher checks cover path quoting, service identity and readiness.

SSH integration#

Run actual SSH integration with a distribution OpenSSH server binary and tmux. The test server listens on a temporary loopback port and uses generated keys, a temporary home and a private configuration.

FICC_TEST_SSHD=/usr/sbin/sshd python -m pytest tests/integration

If the server binary is unavailable, the integration test reports a skip. Set FICC_REQUIRE_SSH=1 to make a missing prerequisite fail instead. A skipped SSH check is not evidence of a successful SSH connection.

Browser checks#

Run browser checks against an isolated service. The test obtains its login credential through the local CLI and does not retain it in the report.

ficc serve --demo --state-dir /tmp/ficc-browser-state --port 8171

In a second terminal:

FICC_STATE_DIR=/tmp/ficc-browser-state FICC_URL=http://127.0.0.1:8171 \
  npm test --prefix web

Keep captures outside the repository. Check narrow and wide layouts, browser zoom, keyboard operation, focus, stale data, errors, access limits and logout. Inspect the rendered result as well as the assertions. Job browser fixtures cover preview and confirmation, permission separation, output following, recovery and cancellation. These fixtures do not qualify a remote systemd implementation. Exercise installed helpers on supported machines with bounded CPU jobs before claiming job execution, effective limits or desktop startup support.

Installed packages#

Package checks install the wheel in a fresh environment. Verify that the node helper and web assets are present without access to the source checkout. Record source revision, commands, case counts, skips and exit status for each qualification. Performance limits require separate measured evidence.

Maintenance and operation checks#

Maintenance checks use distinct 1-record and 64-record histories. They cover exclusive state ownership, consistent credential-free backup, restore integrity, WAL recovery, interrupted archive publication and idempotent resume. Active or unknown work, unsafe paths and changed identities must prevent retirement. Archive integration also runs the owner CLI through an isolated SSH server and checks real systemd unit absence. Use separate synthetic state for these checks; do not retire an operator's history as a test fixture.

Observation checks exercise owned connection reuse, renewal before expiry, changed SSH configuration, pinned identity and process cleanup. Count controller and child CPU together for performance measurements. A dedicated Linux cgroup's cumulative CPU counter includes terminated child processes; sampling only live processes can miss their cost. Measure browser memory separately.

File checks cover root and entry identity, hostile names, permissions, previews, verified transfers, interruption, explicit cleanup and distinct 1-entry/64-entry selections. Terminal checks use real PTYs and an isolated SSH server for binary bytes, resize, tmux detach/reattach and exact stop. Permission, ticket and bounded acknowledgement checks also cover active revocation. A private tmux fixture installation can use FICC_TEST_EXTRA_PATH for the SSH server PATH; production does not read this test setting.

Exercise the complete installed browser workflow with synthetic files before claiming file transfer or interactive terminal support. Compare source and retrieved hashes independently, and test revocation while a terminal is open.

Explorer browser checks cover sorting, filtering, keyboard column resizing, opaque selection identities and obsolete listing replies. Terminal workspace checks use local xterm rendering for machine tabs, split geometry, focus, hidden output, explicit attachment, fullscreen and disposal. Distinct 1-session and 64-session routing checks reuse the bounded concurrent attachment capacity.

Agent and bus browser checks cover exact launch previews, explicit recipients, 1-agent and 64-agent selections, lost-response retries, permission loss, literal message rendering and receipt stages. Browser fixtures do not establish native agent protocol behavior. Test the installed adapter version separately, with its actual extension or Unix endpoint, and exercise an isolated node workflow. An accepted message is not evidence of completed agent work.