net/url: inconsistent Parse error with missing port #12200
Comments
|
CC @mikioh |
|
I'm not sure whether there's a good way to support both RFC 3986 and non-RFC 3986 compliant stuff such as database source names and connection strings. See #12023.
That's a good question, but I have no answer to it for now. Sorry. |
|
Just to be clear, my point is not that "http://example.com:/" should be supported, even though it was supported so far in Go 1.4 , and even though section 6.2.3. seems to imply it is a compliant case. My point is: if it is decided that the "http://example.com:/" form is supported for parsing, then all three of: https://192.168.0.2:/foo should probably be supported (as it was the case so far), and not just the two first forms. Or are you saying that the brackets around the host is exactly what makes that form non-RFC 3986 compliant? |
I think of simply accepting |
url.Parse in Go 1.5 now fails with the scheme://IPv6:/foo form (trailing colon). But it still works when the hostname is an actual hostname, or an IPv4, which seems inconsistent. This CL therefore not only accepts this new failure (which comes with Go 1.5) but also considers any host with a trailing colon as an invalid URL and refuses to rewrite it. There probably are other places in Camlistore where we should the same. See golang/go#12200 Change-Id: Ib9ea95a71013b5b53f2fd99415015ec916bb4d9d
|
I made this as well see https://play.golang.org/p/qm_YKwrvf3 and moby/moby#15703 |
|
For the record, https://2b01:e34:ef40:7730:8e70:5aff:fefe:edac:/foo is NOT a well-formed IPv6 URL. The [ ] around the IPv6 address are not optional. If browsers accept that, consider yourself lucky. For the most part, net/url makes no attempt to parse the host in the URL. The exception is IPv6 literals, although I am not sure why we bother to check them so carefully. Maybe because they are just one giant exception to all the rules. In any event, the grammar in https://tools.ietf.org/html/rfc3986#page-22 seems to suggest that port can be an empty string even when a colon is present, so I'll make that change in the IPv6 literal parsing. |
|
well, it's good to know that url.Parse can't be relied upon for checking the host part, thanks. I would think that specific bit should be mentioned somewhere in net/url, but I suppose there's a good reason for it not to be there. Maybe you don't want to commit to that kind of implementation "detail" ? |
|
CL https://golang.org/cl/17384 mentions this issue. |
go version devel +467a2cb Sun Aug 16 00:29:39 2015 +0000 linux/amd64
http://play.golang.org/p/FkTC5G-xhb
With Go 1.4, this runs without any error.
With the above go version, only the last form ([IPv6]:) fails to parse with the error:
invalid port ":" after host
I would expect either parsing not to fail at all, or to fail for all the URLs without a port after the colon, for consistency.
Is this the desired behaviour ?
The text was updated successfully, but these errors were encountered: