Rooted flag still causing serialization oddities

[cdunklau]

I thought this was the same bug as #90, but @glyph's #91 didn't fix it, so I guess it's at least slightly distinct.

_________ TestURL.test_reproduce_my_rooted_oddity _________

self = <hyperlink.test.test_url.TestURL testMethod=test_reproduce_my_rooted_oddity>

    def test_reproduce_my_rooted_oddity(self):
        a = URL(scheme='udp', port=4900)
        b = URL.from_text('udp://:4900')
        assert str(a) == str(b)
        assert a.asText() == b.asText()
>       assert a == b
E       AssertionError: assert URL.from_text('udp://:4900') == URL.from_text('udp://:4900')
E         -URL.from_text('udp://:4900')
E         +URL.from_text('udp://:4900')

src/hyperlink/test/test_url.py:1101: AssertionError

That assertion error is entertainingly befuddling 😃

After a cursory inspection of a and b there, the only differences I found in the instance __dict__ were _rooted and _uses_netloc.

The same failure happens with URL(scheme='udp', host='', port=4900), but doing URL(scheme='udp', port=4900, rooted=True) (with or without host) makes the test pass.

(this was tested on 688233a)

[glyph]

This should probably be addressed the same way that the rooted-normalization happens with paths; if you have an empty hostname but a scheme like this, then you're implicitly rooted. (Is this kind of URL technically valid according to the spec?)

[cdunklau]

Is this kind of URL technically valid according to the spec?

Looks like it (selected parts of https://tools.ietf.org/html/rfc3986#appendix-A):

   URI           = scheme ":" hier-part [ "?" query ] [ "#" fragment ]
   hier-part     = "//" authority path-abempty
                 / path-absolute
                 / path-rootless
                 / path-empty

   authority     = [ userinfo "@" ] host [ ":" port ]
   host          = IP-literal / IPv4address / reg-name

   reg-name      = *( unreserved / pct-encoded / sub-delims )

   path-abempty  = *( "/" segment )

Zero-length strings are allowed by both path-abempty (so no / after port is valid) and reg-name (so no "host" is valid), so by my interpretation udp://:4900 is allowed.

But it's not as clear to me what such a form implies in terms of hyperlink's rooted concept.

[wsanchez]

Yeah. rooted seems specific to URLs with paths. I think the existence of this attribute on URL is unfortunate.