RetrieveHass freq parameter not normalised to pd.Timedelta causing AttributeError in Docker Standalone deployments

[rossco555]

Please forgive me, this is above my head and below was generated by Claude following investigations into my issues. I believe it is a bug, but I'm largely trusting Claude for that.

Describe the bug
_process_history_dataframe in retrieve_hass.py calls self.freq.seconds but self.freq is passed as an integer (the optimization_time_step value from config, e.g. 30) rather than a pd.Timedelta object. This causes AttributeError: 'int' object has no attribute 'seconds' which is silently caught upstream and misreported as "The retrieved JSON is empty, A sensor may have 0 days of history".

Additionally, when a pd.Timedelta object is passed, recent versions of pandas have removed the .nanos attribute, causing a further AttributeError: 'Timedelta' object has no attribute 'nanos'.

To Reproduce

1. Install EMHASS as Docker standalone container
2. Set optimization_time_step: 30 in config (integer, as documented)
3. Run dayahead-optim or forecast-model-fit
4. Observe misleading "empty JSON" error in logs

Expected behavior
RetrieveHass.__init__ should normalise the freq parameter to pd.Timedelta regardless of input type. Suggested fix:

if isinstance(freq, (int, float)):
    self.freq = pd.Timedelta(minutes=int(freq))
else:
    self.freq = freq

This ensures self.freq is always a pd.Timedelta throughout the class. Any subsequent use of .seconds will then work correctly without needing to handle multiple types downstream.

Home Assistant installation type

• Home Assistant OS (running in Docker on Synology NAS)

Your hardware

• OS: Linux (Synology DSM)
• Architecture: amd64

EMHASS installation type

• Docker Standalone (separate container, connecting to HA via host.docker.internal)

Additional context
The type hint on RetrieveHass.__init__ correctly specifies freq: pd.Timedelta but the caller in command_line.py passes retrieve_hass_conf["optimization_time_step"] which is a raw integer. The fix belongs in __init__ to normalise at the boundary rather than patching each usage site.

Note: pd.Timedelta.nanos has been removed in recent pandas versions. Use pd.Timedelta.seconds or int(pd.Timedelta(minutes=x).total_seconds()) for portable code.

[davidusb-geek]

Well it seems that this is just an issue when catching generic Exceptions.
An attempt to fix these here #735