# Testing

## Without hardware

`tests/conftest.py` serves recorded WS-Management answers over mocked HTTP
(`responses`): `Get` → `<Class>.get.xml`, `Pull` → `<Class>.pull.xml`, a method →
`<Class>.<Method>.xml`. A class without a file answers with the fault for an unknown
resource URI, as AMT does. A test changes one answer with `AmtServer.override` and a
recorded file with `patched(name, old, new)`, which fails when `old` is not in it.

| Fixtures | Source |
|---|---|
| `tests/fixtures/dmt-wsman-messages/` | Intel's go-wsman-messages (Apache-2.0), `pkg/wsman/wsmantesting/responses` at commit 00402b9: AMT 12.0.67 on an Intel NUC9, a few values replaced by placeholders. `fault_destination_unreachable.xml` is written by hand after DSP0226 |

There is no AMT simulator worth the name. Intel's Device Management Toolkit talks to
real AMT or to its own recordings, the ones used here.

## Recording a real AMT

`python -m napalm_amt.record HOST --password-file FILE --out DIR` makes every read the
driver makes and writes each raw answer under the name the mock serves it by, plus
`COMMANDS.txt` and `landing_page.txt` (the unauthenticated page with its headers, which
is what discovery sees). It never switches power and never writes to AMT. Pages after the
first become `<Class>.pull.2.xml`. A class AMT does not know is kept as
`<Class>.enumerate.xml` with its fault.

Name the directory `<vendor>-<model>-<amt version>`, for example
`lenovo-m910x-11.8.50.3399`.

## Running

```bash
.venv/bin/python -m pytest
.venv/bin/ruff check . && .venv/bin/mypy napalm_amt
```