For some schemes, URLs with no authority are rendered incorrectly about uri HOT 4 CLOSED

Weigh in from him is that there's no such thing as (filesystem) relative paths, and that the file scheme doesn't accept them and won't handle them.

I.e., the file scheme only handles URI-style relativity, not system-style relativity.

from uri.

file: is a URL-like protocol (utilizing URL-like :// scheme separator resolution), meaning file://relativepath is declaring an authority of "relativepath", with no path. URI('file:///rootedpath').path → PurePosixPath('/rootedpath'). Because of it being URL-like, you're mostly right, the path can not easily start with two slashes. Or, rather, to begin with two slashes, you'd need four consecutive ones: URI('file:////foo').path → PurePosixPath('//foo') — similarly, use of just a colon to separate the scheme from remainder results in the remainder only being interpreted as path (e.g. as per isbn:); no possibility for authority or other URL structure. (Basically, doing that you're forcing it to be treated as a bare relative path, but notice the repr: URI('file:relativepath') → URI('file://relativepath'))

There's a lot of confusion around URI : schemes and URL :// schemes; I'll need to investigate how to properly document this. (And if any changes to the code are necessary…)

from uri.

You are 100% correct -- I actually thought I didn't submit this post, for just that reason.

The guy who did the spec seems pretty reasonable, and might go for file:foo/bar as relative, but I'm not sure. As of now according to the RFC, file:foo/bar is undefined, and relative file paths (as opposed to relative URL paths) aren't even discussed.

Informally, I've seen 'file:foo/bar' used, since it's conformant with the more general URI spec, and is otherwise an error in the file spec.

Edit: Sent him a message, comparing relative file:foo/bar.baz with his examples of proper usage that include file:c:/foo/bar.baz and file:/foo/bar.baz. ..we'll see if he has anything to say on the matter, if he notices the message :-).

from uri.

Until we have a weigh-in from him, and/or if we don't get one, this might be something that needs to either do:

• Strict RFC (i think):
  • reject file:foo URIs, but accept file:c:/foo and file:/foo URIs
• allow common usage
  • in ABNF: local-path = path-absolute / path-rootless instead of local-path = path-absolute.

..or, more simply, possibly just display rootless paths with no authority as file:rootless/path when making a str() of the URI.

from uri.