`NmtBase.state` should always return a `str`

[erlend-aasland]

Problem

NmtBase.state is documented and annotated as always returning a string, however there is a code path that may return an int:

https://github.com/christiansandberg/canopen/blob/62e9c1f851cb8721355d22c8571f2b41599c1b36/canopen/nmt.py#L74-L92

The code path in question is highly unlikely; the NmtBase._state member is set in the following places:

• ✅ NmtBase.__init__: explicitly set to 0
• ✅ NmtBase.on_command: set iff the new state is a valid/known state
• ✅ NmtBase.send_command: set iff the new state is a valid/known state
• ❓ NmtMaster.on_heartbeat: set to whatever was decoded from the heartbeat message, which should be fine for any compliant device

Possible solutions

In NmtMaster.on_heartbeat, only set _state if it is a valid state; if it's not, log.error(...) or raise an exception.

In NmtBase.state, either raise an exception if ._state is not a valid state, or return something a la f"UNKNOWN STATE {self._state}".

Draft diff

diff --git a/canopen/nmt.py b/canopen/nmt.py
index 401ad15..1316a5c 100644
--- a/canopen/nmt.py
+++ b/canopen/nmt.py
@@ -86,10 +86,8 @@ class NmtBase:
         - 'RESET'
         - 'RESET COMMUNICATION'
         """
-        if self._state in NMT_STATES:
-            return NMT_STATES[self._state]
-        else:
-            return self._state
+        assert self._state in NMT_STATES:
+        return NMT_STATES[self._state]
 
     @state.setter
     def state(self, new_state: str):
@@ -122,6 +120,12 @@ class NmtMaster(NmtBase):
             logger.debug("Received heartbeat can-id %d, state is %d", can_id, new_state)
             for callback in self._callbacks:
                 callback(new_state)
+            if new_state not in NMT_STATES:
+                log.error(
+                    "Received heartbeat can-id %d with invalid state %d",
+                    can_id, new_state
+                )
+                return
             if new_state == 0:
                 # Boot-up, will go to PRE-OPERATIONAL automatically
                 self._state = 127

[sveinse]

Awesome. I've found the same thing while type annotating, so this discussion is on my list 👍

Now, the core question is: Does the canopen standard allow for other/custom states than what's defined in the standard? I can't remember on the top of my mind if it does.

[erlend-aasland]

Now, the core question is: Does the canopen standard allow for other/custom states than what's defined in the standard? I can't remember on the top of my mind if it does.

Standard compliant devices are probably ok; but non-standard compliant devices do exist. I've got a few of them in the project I've inherited. The device in question does not heed NMT commands (apparently they dropped NMT support from the firmware, because the could not get it to work), it does not support SDO (including configuring PDOs), nor does it support the SYNC protocol; all it does is to continuously stream hard-coded PDOs from a hard-coded OPERATIONAL state.

Hence, I think the core question is: what to do with non-compliant and buggy devices?

[erlend-aasland]

To answer myself: I believe the most sane solution is to discard unknown states, similar to my PoC patch in #500 (comment)

[sveinse]

To answer myself: I believe the most sane solution is to discard unknown states, similar to my PoC patch in #500 (comment)

I'm conflicted about it. By discarding any unknown states, it can never be used to communicate with nodes that have some non-standard behavior (for whatever reason).

[erlend-aasland]

That's a valid concern. Draft for an alternative patch:

Details

diff --git a/canopen/nmt.py b/canopen/nmt.py
index 401ad15..d8951a7 100644
--- a/canopen/nmt.py
+++ b/canopen/nmt.py
@@ -89,7 +89,7 @@ class NmtBase:
         if self._state in NMT_STATES:
             return NMT_STATES[self._state]
         else:
-            return self._state
+            return str(self._state)
 
     @state.setter
     def state(self, new_state: str):
@@ -122,6 +122,11 @@ class NmtMaster(NmtBase):
             logger.debug("Received heartbeat can-id %d, state is %d", can_id, new_state)
             for callback in self._callbacks:
                 callback(new_state)
+            if new_state not in NMT_STATES:
+                log.warning(
+                    "Received heartbeat can-id %d with invalid state %d",
+                    can_id, new_state
+                )
             if new_state == 0:
                 # Boot-up, will go to PRE-OPERATIONAL automatically
                 self._state = 127

[erlend-aasland]

Related: the conditional return in NmtBase.state could probably be sped up by using a try...except clause, but I'm not sure it's worth it, unless .state is expected to be a part of a hot loop.

[acolomb]

The reason for passing out unkown values as integers is to maximize forward compatibility. The CANopen standard may not define more NMT states now, but what happens when another state is added in a later revision? Or by a buggy / noncompliant device.

I think it's perfectly fine to return something like "UNKNOWN STATE 123" in these cases. The log warning is too verbose, as that would lead to the logs being flooded with every heartbeat transmission. We already have the numeric values in a debug message, that's enough to troubleshoot.

As we don't really handle any behavioral changes based on NMT state, but just convert it to a string for the application to consume, I think we're fine returning these UNKNOWN strings.

Feel free to propose a patch including the try... except change. It is an exceptional condition, and we can avoid the duplicate effort of key in m and m[key]. The alternative m.get(k, default=...) is not any better because it would always calculate the default value expression. I don't expect any of this to be used in hot code-paths, but still more pythonic to handle an exception here IMHO.

[erlend-aasland]

I agree regarding the log warning. I'll create a PR with my draft diff from #500 (comment), with the optimisation applied.

[erlend-aasland]

Well, first, let me shave some yak; I'd prefer if NMT stuff was covered well by the test suite before changing the behaviour. Not to mention that some of the existing tests depend on hard-coded timings (I've observed flakiness while running the test suite locally).

[acolomb]

Hm, since this is only affecting devices that return a non-standardized state number, I think we're free to change whatever. But alright, let's look at tests first. I'm very grateful that you put such an emphasis on that part, as I personally could never really warm up with the whole unit testing topic.

[erlend-aasland]

FTR; I'll split the test improvements into several PRs, for the reviewers convenience.