Indeed, the RFC-1738 does not mention the file:/path (single slash) as valid, although it does seem to work as expected. It is compatible with RFC-3986 uri syntax, though.

To sum up:

• if you use file:// (double slash), the next part is host name not the first element of absolute path.
• if you mean absolute path on local machine, use 3 slashes (as per RFC) or single slash (as it also seems to work)
• we pass it directly to hg command so all complaints should go to hg authors

I'm closing this isse as "invalid"

Slawek and Piotr,

I am not an expert on the subject of URLs, but I think there is a lot of confusion here.

From what I know, in URLs the first part is <protocol_name>://. Where <protocol_name> is e.g. http, https, ssh, and file. the protocol name should be separated from the rest of the URL with ://. So, I think of the "file://" part of the URL, as the "protocol part". The rest of the URL starts from there. Does this make sense?

So, file:///Users/amazkovoi/src/atlassian/fe-hg is a correct URL. Where "file://" is the protocol part, and "/Users/amazkovoi/src/atlassian/fe-hg" is the absolute path to the repository.

The URL is actually a short form of file://localhost/Users/amazkovoi/src/atlassian/fe-hg

The file://localhost/Users/amazkovoi/src/atlassian/fe-hg URL explicitly specifies the hostname, which is "localhost", and then it shows the absolute path, which is "/Users/amazkovoi/src/atlassian/fe-hg".

With the "file://" protocol, "localhost" is the default host. It looks like hg ignores the hostname even if it is part of the URL and assumes localhost.

So for me all of the following URLs worked:

1. file:///Users/amazkovoi/src/atlassian/fe-hg
2. file://localhost/Users/amazkovoi/src/atlassian/fe-hg
3. file://blah/Users/amazkovoi/src/atlassian/fe-hg

Of course, I have a Hg repo at the /Users/amazkovoi/src/atlassian/fe-hg path on my machine.

Also, the "hg help urls" command on Mac for Hg version:
Mercurial Distributed SCM (version 1.6.4+20101001)

Is correct. It only talks about the file://local/filesystem/path[#revision] URLs form, which works fine, and as far as I know is perfectly allowed by the URL RFC.

Please let me know if you think I am wrong about something here.

Cheers,
Anton

Actually, that's a mistake repeated so many times that it should be used as correct notation

If you use file://Users/dir the Users part is treated as host part (ignored with file: protocol) and the dir part is /dir
At least, this is how hg interpretes the url on Linux.
The correct url is with 1 slash (file:/absolute/path) or none (file:relative/path)
3 slashes work but don't make too much sense.

I dare to say: hg help urls lies about file:// url syntax
At least the 1.6.4 on Linux
By lies I mean it does not handle the file://absolute/path as /absolute/path despite what description suggests.

Questions:

• should we warn the user when we see file://path url that is almost surely wrong?
• should we transparently handle that and translate to file:/path ?