The removal of the if() we can land immediately. I guess that will ease some stress before the bcon3, so then we can focus on making code to be more robust to errors.
Let's then limit this PR…
The only place where we currently construct default DeviceInfo() is in the default constructor of SessionParams, where the device is expected to be explicitly configured to some real information.…
@Sergey , so do you approve and I can merge it then?
Do we really need to anything about this DeviceInfo at this time?
Yes, because current code would cause assert failure at any attempt to compare object created by default constructor…
The rest I don't quite understand. This change proposes something that leads to a discrepancy between type, id, and, possibly semantic: the type is still set to DEVICE_CPU, but the id is set to…
I don't mind such a change, but I am concerned that it might decrease the readability of the code related to this variable in the session.cpp, especially the fragment (below), which currently…
@brecht, following up on our discussion from last Friday, I have completed the refactoring I mentioned and have retested the PR on various setups as a precaution. Everything is functioning well;…