I am an deep learning researcher and I use PyCharm to manage many codes on remote servers.
Unlike a developer on web or other type of python projects, I need to open many different projects in a single window with different python interpreter.
So I create many projects and attach them in the same window. Then I configurate the path mapping in development settings for each project (each project have at least one path mapping).
Besides, I use Anaconda to create many different python environments with different packages. For example, I setup many environments for each version of TensorFlow and PyTorch. And each interpreter should be able to be used by different projects.
However, the "add new interpreter" must setup at least one path mapping (project root ---> remote path) and after the configuration, all path mapping I configurated in "development" will be overwrite by this project mapping.
I think the logic behind this is that each interpreter sould be only used by a single project, which is not shareable with other projects. But for machine learning and research purpose, the interpreter should be independ for each project so that different project can use identical interpreter.
And I also think that whether we add new python interpreter, the path mapping in development should not be touched at all.