Tango: post-sanitisation device-class-name collision when ids differ only in '-' vs '_'

[gilesknap]

Found while reviewing PR #360 commit 207d68d8 (slice #357).

validate_tango_id accepts [A-Za-z0-9_-]+ — both hyphens and underscores are allowed. tango_dev_class_name then collapses hyphens to underscores so the result is a valid Python identifier:

# transports/tango/util.py
def tango_dev_class_name(id: str) -> str:
    sanitized = id.replace("-", "_")
    if sanitized[0].isdigit():
        sanitized = "X" + sanitized
    return sanitized

The problem: two distinct, individually-valid ids "DEV-1" and "DEV_1" both sanitise to Python class name "DEV_1". If a user declares both in controllers:, TangoDSR._create_device is called twice, each type() call returns a distinct Python class object — but both share __name__ == "DEV_1". tango.server.run(tuple(self._devices), [FASTCS_TANGO_SERVER_NAME, ...]) then registers two Tango classes under the same name; whichever is created second silently overrides the first in any class-keyed lookup at the Tango DB.

The launch-framework's controllers: dict can't catch this — the dict keys "DEV-1" and "DEV_1" are distinct strings, so ruamel's duplicate-key detection doesn't fire. The collision only emerges after the Tango-specific sanitisation step, so it must be detected at the transport boundary.

The existing two-controller test in tests/test_multi_controller.py::test_tango_transport_builds_one_device_per_controller_with_id_in_name uses "ALPHA"/"BETA" which sanitise distinctly, so this case isn't covered.

Suggested fix

Detect the post-sanitisation clash inside TangoTransport.connect (or TangoDSR.__init__) and raise with the colliding inputs called out:

seen: dict[str, str] = {}
for api in controller_apis:
    id = api.path[0]
    validate_tango_id(id)
    cls_name = tango_dev_class_name(id)
    if cls_name in seen:
        raise ValueError(
            f"Controller ids {seen[cls_name]!r} and {id!r} both sanitise to "
            f"Tango device-class name {cls_name!r}; pick one variant "
            f"(hyphens and underscores are not distinguishable in Tango class names)"
        )
    seen[cls_name] = id

Add a regression test in tests/test_multi_controller.py that declares ["DEV-1", "DEV_1"] and asserts the fail-fast message.

Alternative

Tighten validate_tango_id to reject - outright. That makes Tango a slightly more restrictive lowest-common-denominator alongside GraphQL (which already disallows hyphens), and the per-transport charset table in multiple-transports.md would need updating. This is a more aggressive cut — the post-sanitisation check is preferred unless we want the docs/charset table to be cleaner.