replacing ContentsManager

[telamonian]

Instead of trying to hack notebook's existing ContentsManager, we should replace it with something more flexible, based upon a more general notion of what "content" is. There are a number of popular projects on github that all have the aim of creating a spec for a virtual filesystem-ish thing. We should use one of the extant python options and/or take inspiration from the following:

Python

• PyFilesystem2

  • More mature/popular
  • The minimal set of things you need to implement for a valid filesystem is well defined and wonderfully small: https://docs.pyfilesystem.org/en/latest/implementers.html#essential-methods

• filesystem_spec

  • pyfilesystem vs fsspec "discussion": https://github.com/intake/filesystem_spec/issues/5
  • More focused on performance and parallelization
  • May be more flexible

C++

• cppfs
  • pro: we could write all of our filesystem backends in C++
  • con: we probably shouldn't

PHP

• flysystem
  • Sadly, by far the best supported virtual filesystem library out there is 100% PHP, which is a bummer. Has a whole bunch of different filesystem implementations

[telamonian]

I'm not fully convinced that any of these do everything a jupyter contents layer needs, and I assume they all do a lot of stuff we don't need. I think the main thing to learn from these is what a truly minimal filesystem spec actually looks like

[timkpaine]

I'm not fully convinced that any of these do everything a jupyter contents layer needs, and I assume they all do a lot of stuff we don't need

Yep definitely. In this vein, some things that a filesystem doesn't necessarily do e.g. version control, you'd want a Jupyter filesystem to be aware of

[timkpaine]

@telamonian this should help us start to answer some questions. A very rough cut, but illustrates how much pyfilesystem offers on top: https://github.com/timkpaine/multicontentsmanager/blob/cleanup/multicontentsmanager/pyfilesystem_manager.py#L13