Fully delegate the file system management to the cache, since it's only required for that.
In principle, its usage might even be gated behind a feature, and we could use memory caching otherwise.
This would also allow to just run in a browser, where a file system is not available (or in other contexts where it might be available, but e.g. it's not reliable, or slow).
It will also pair with the configurable sources, since the data-fetching connection can be established on LAN, instead of through the web.
Currently, the main obstacles are:
tar archive unpacking, happening on file system, and
the lack of an actual memory cache, so the full set should be loaded every time a single member is requested
this would be fine in memory-only setup, since memory has to be consumed here or there
but it would be less compatible with the file system setup, where the full set is fetched upon individual member request, but only the single member requested is returned
Fully delegate the file system management to the cache, since it's only required for that.
In principle, its usage might even be gated behind a feature, and we could use memory caching otherwise.
This would also allow to just run in a browser, where a file system is not available (or in other contexts where it might be available, but e.g. it's not reliable, or slow). It will also pair with the configurable sources, since the data-fetching connection can be established on LAN, instead of through the web.
Currently, the main obstacles are: