Simulator-backed testing

pybench tests can run against pyvisa-sim instead of real hardware. This is what CI uses to verify that every instrument class emits the SCPI it claims to.

How it works

bench.common.Common accepts a backend argument that is passed to pyvisa.ResourceManager. Setting backend="<yaml-path>@sim" swaps the live VISA layer for pyvisa-sim, which serves canned responses out of a YAML file.

from bench.keysight import N5182B

sg = N5182B(
    address="TCPIP0::sim-n5182b::inst0::INSTR",
    backend="tests/sim/keysight/n5182b.yaml@sim",
)
sg.connect()
sg.frequency = 2.4e9
assert sg.frequency == 2.4e9

Sim YAMLs live in tests/sim/keysight/. The CI workflow at .github/workflows/test.yml runs tests/test_sim_keysight.py against every YAML on push.

Recording from real or Command Expert-emulated hardware

The scripts/dump_to_sim.py helper captures real SCPI traffic and writes a pyvisa-sim YAML.

The intended workflow is:

1. Install Keysight Command Expert (Windows). Command Expert ships with simulator personalities for most current Keysight instruments and exposes them as VISA resources on localhost.

2. Configure the simulated instrument in Command Expert so it appears at a VISA address such as TCPIP0::127.0.0.1::INSTR.

3. Run the dump script pointing at that address:

   python scripts/dump_to_sim.py N5182B TCPIP0::127.0.0.1::INSTR \
       --set frequency=2.4e9 \
       --set power=-10 \
       --set output_enabled=1 \
       --out tests/sim/keysight/n5182b.yaml
   

   The script walks every readable property on N5182B (skipping reset), then exercises any --set name=value setters you provide. Each captured (request, response) pair becomes a dialogue in the emitted YAML.

4. Commit the YAML. From here on, CI replays it via pyvisa-sim — no instrument needed.

The same script works against real hardware (drop --backend and use the instrument’s actual VISA address), which is useful when Command Expert doesn’t carry a simulator for your model.

Limitations

• The emitted YAML uses a dialogues block, so query responses are frozen at capture time — writes don’t update what subsequent reads return. For round-trip simulation, hand-edit the YAML to use a properties block (see existing files in tests/sim/keysight/ for examples).

• Only properties defined directly on the instrument class are walked. Compound instruments with sub-objects (e.g., E36233A.ch1, E36233A.ch2) need to be dumped per sub-object or have setters supplied via --set.

• A passing sim test proves the SCPI string the class emits is correct, not that the instrument behaves as expected. Hardware-in-the-loop tests via the instruments_required pytest marker are still required for behavioral verification.