fsspec URIs with '+' instead of '::'

[hhoeflin]

Hi,

I wanted to ask about the fsspec convention of using '::' to chain URL handling protocols (e.g. simplecache with s3fs). Using '::' means the resulting string is not a URI (e.g. 'simplecache::s3fs://bucket_name/myfile') as the 'simplecache::s3fs' violates https://datatracker.ietf.org/doc/html/rfc3986#section-3.1

This is also related to pandas not being able to use these (see pandas-dev/pandas#59684) as it checks if the path given to pandas follows RFC3986. My question is if it would be possible to also allow specification of '+' to chain URL, which would then result in a valid URI.

This would not cover the more complicated chaining cases, but at least the simple one.

Thanks

[martindurant]

This was broken in pandas by pandas-dev/pandas#44619

I would say, that "+" is far more likely a valid string within a URL than "::" is.

fsspec URLs already are non-compliant with the RFC even if you make the replacement, for example something like "zip://path/to/file::s3://bucket/file.zip". The change should be made in pandas to make sure that valid fsspec paths are picked up, not here, which would break all other users of the current fsspec pattern.

(aside: I'm not really sure why pandas would accept a string in read_json, given that StringIO exists)

[martindurant]

Does using a URL like "simplecache://::s3://bucket_name/myfile" work?