Don't we need to check whether the PowerSource cluster is supported or not?

The lines here occur in info_changed. The device_init event will always run before this when a device is created or a driver starts up, and in device_init we have this logic:

if #device:get_endpoints(clusters.PowerSource.ID, {feature_bitmap = clusters.PowerSource.types.PowerSourceFeature.BATTERY}) == 0 then
    device:set_field(profiling_data.BATTERY_SUPPORT, battery_support.NO_BATTERY, {persist = true})
  end

doing that check. Therefore, the only way BATTERY_SUPPORT can be nil her is if the PowerSource cluster with the FeatureMap we care about is supported on the device.

The lines here occur in info_changed. The device_init event will always run before this when a device is created or a driver starts up, and in device_init we have this logic:

If so, why we need to call same logic below? the logic is already in the device_init

  if device:get_field(profiling_data.BATTERY_SUPPORT) == nil then
    device:add_subscribed_attribute(clusters.PowerSource.attributes.AttributeList)
  end

I think I added this as a kind of safety check, but you're right, it is not required in the current implementation. The AttributeList will always be added to the subscribed_attributes table used in subscribe in init if the nil condition is met by the time infoChanged would occur.