htslib should implement a more complete URI scheme for `irods:`

[jrandall]

The current implementation of the irods URL scheme handler only supports basic path (it fills in missing information from the defaults in the user's irodsEnvFile.

The irods: URL scheme should:

• be consistent with RFC 3986
• allow all required variables from irodsEnvFile to be set from the URL (i.e. irodsHost, irodsPort, irodsZone, irodsUserName) such that a separate user env file is not required to access irods data
• allow 'relative' URLs that do require information from irodsEnvFile in order to function (i.e. if all of the above as well as irodsCwd are set, then a URL such as irods:foo might refer to the foo file in the current working directory.
• allow supported authentication methods to be configured (i.e. specify irodsAuthScheme, irodsServerDn, and irodsAuthFileName or possibly the actual password to be specified)
• allow irodsDefResource to be set (only applies when writing to iRODS)
• follow the pattern of established URL schemes and be backwards compatible with existing htslib implementation
• possibly allow for fragments
• eventually be registered with IANA under the irods scheme

[jrandall]

N.B. we should (obviously?) consult with and get buy-in from the iRODS consortium developers on the scheme

[dkj]

How about: irods://[irodsUserName#irodsZone@][irodsHost][:irodsPort]/collection_path/data_object ?
The "#" separating username and zone would fit with iRODS convention.

[jmarshall]

It would be nice, but that # (if not percent-encoded) wouldn't fit the rules for URIs. RFC 3986 §3.2:

The authority component is [...] terminated by the next slash ("/"), question mark ("?"), or number sign ("#") character, or by the end of the URI.

[dkj]

Doh. Thanks. How about we percent encode the "#" then?

irods://[irodsUserName%23irodsZone@][irodsHost][:irodsPort]/collection_path/data_object

A bit ick perhaps - but less ick that cooking up some other separator?

[jrandall]

If the username/zone is intended to be a fully-qualified way of specifying
the login account to use, then this seems fine, although it is missing the
password, which I think we should include in the same way as it can be
included in http and ftp URIs for completeness.

We might also need a separate zone specifier, but I think that is only
required when doing queries, so perhaps it isn't necessary if we consider
running a genQuery outside the scope of the URI scheme.

It should also be specified how to authenticate with a token rather than a
username/password.

On Wed, Sep 30, 2015 at 11:57 AM, dkj notifications@github.com wrote:

Doh. Thanks. How about we percent encode the "#" then?

irods://[irodsUserName%23irodsZone@][irodsHost][:irodsHost]/collection_path/data_object

A bit ick perhaps - but less ick that cooking up some other separator?

—
Reply to this email directly or view it on GitHub
#229 (comment).