Interface with udev database for various fixes (but future work needed)

[abbaad]

Rather than attempting to parse /dev and /sys directly within smfc, we can outsource that task to udev (through pyudev). This allows us to simplify much of the logic and string parsing, making it less fragile and more robust.

This new strategy gives a convenient fix to #25.

Additionally: under linux (under the persistent storage naming rules present in most distributions AFAIK), the NVMe device by-id symlinks point to the block device (e.g., /dev/nvme0n1) rather than the controller (/dev/nvme0). However, the corresponding hwmon device is actually a child of the controller (thus a sibling to the block device in sysfs).

The original code has a bug where the symlink is assumed to be backwards (so it assumes the link points to /dev/nvme0 and concatenates the namespace within sysfs). This bug is also present in the unit tests in the test data created by the file test_00_data.py.

Properly parsing the string would require stripping the namespace (in 99% of cases, this is "n1") from the block device name. Instead, in this commit we use pyudev to query the udev database to find the sibling hwmon device. Unfortunately, the test file doesn't have such a simple fix, so someone will have to patch it more deeply.

Unfortunately, the unit tests will require a more extensive refactoring to be able to support pyudev, but these changes are not a part of this PR. If we want to keep the current strategy of creating a set of "test data" for the tests, then something like umockdev might be best. Otherwise, we could write Mocks for the expected function calls, but for that a wholesale restructuring of all the test data is probably required.

For now, this code continues to use the open and read paradigm on sysfs files, but a future refactor might instead use the pythonic interface from pyudev (e.g., hwmon_dev.attributes.get('temp1_input')).

[petersulyok]

Hi @abbaad, thanks for this pr, this is impressive. Let me have some time to understand how pyudev is working, and what this dependency means for the project.

[petersulyok]

Hi @abbaad, I think I got point behind pyudev and I have only one issue. This code seems not to work for me on a coretemp module (added in Fix: implement AMD CPU autoconfiguration):

for module in ['coretemp', 'k10temp']:
   hwmon_path = [self.get_tempinput(dev) for dev in self.udev_context.list_devices(DRIVER=module)]
   if hwmon_path:
      break
if not hwmon_path:
   self.log.msg(Log.LOG_ERROR, 'No explicit hwmon_path was configured, and automatic detection failed to find any devices using the coretemp or k10temp modules')

hwmon_path is always empty. Any idea on this?