Fix parsing s3 accesspoint url

[wang2bo2]

Parsing S3 access point URI, like s3://arn:aws:s3:us-west-2:1234:accesspoint/abc/123.jpg, causes problems in infer_storage_options for the :accesspoint was parsed as port number. This PR addresses that.

[martindurant]

Whilst this appears to do the right thing, I wonder if you see any way that this can be isolated in s3fs rather than here?

[martindurant]

ping @wang2bo2

[wang2bo2]

Yes, we could remove the use of infer_storage_options and duplicate the extraction of url_query here
https://github.com/fsspec/s3fs/blob/f3f63cbfbfe71a4355abd63cafd8c678c4a5a0af/s3fs/core.py#L394

In the current infer_storage_options implementation, anything after the last : would be excluded from getting into options["host"]

filesystem_spec/fsspec/utils.py

Line 97 in 9a16171

options["host"] = parsed_path.netloc.rsplit("@", 1)[-1].rsplit(":", 1)[0]

even if the non-int port error is caught and handled. So I thought it's good to fix that too.

[martindurant]

I think it would make sense to look for "accesspoint" in s3fs, but still use to infer_storage_options in the common case. Perhaps the accesspoint branch could simply modify the URL into something that makes sense. An ARN description isn't very much like a URL anyway...

[martindurant]

Closing in favour of fsspec/s3fs#912