Log: Bug Bash & Adversarial Testing

Commits: 9b563c4, ade9056, bd4e5bc, 1494700

Overview

Spent the session stress-testing the HAL from every angle — found 4 real bugs, fixed them all, and built a permanent adversarial regression suite to keep them from coming back.

Bugs Found & Fixed

1. MeasurementResult.__format__ Crash (Medium)

f"{result:.2f}" on a list-valued or None result hit format(str(...), format_spec) which raises ValueError: Unknown format code 'f' for object of type 'str'. The fix: return str(self.value) unconditionally when the value isn't a scalar — no format code to complain about.

2. Simulated PSU State Tracking (High)

SimulatedPowerSupply was a ghost town — set_voltage(5.0) printed the intent but get_voltage() handed back 0.0 every time. Same for set_output(True) / get_output(). Both now store their setpoints in instance variables.

3. is_sim_mode() Inconsistency (Medium)

factory.is_sim_mode() only checked for "SIM" while config.is_sim_mode() also supported "SIMULATED". Anyone setting INSTRUMATION_MODE=SIMULATED would silently get real hardware mode from the factory path. Factory now matches config.

Attack Surface

27 adversarial tests now probe edge cases: - MeasurementResult dunders with invalid types (lists, None, complex) - PSU state leakage between operations - Safety guardrail boundary conditions (negative frequency, over-power) - Factory AUTO with zero hardware available - is_sim_mode() env value consistency - Async method dispatch on non-existent methods - Repeated connect/disconnect cycling - Instrument isolation (shared state check)