(Sorry for the late response, I keep getting sidetracked by other stuff)
Is there a reason you're trying to deprecate the URL constructors?
And if we end up not using the URL constructors, then I still think it
would be nice to have a test library, since it's easy to forget to add the
"[]". Unless we fix up all the existing tests and you'll be sure to catch
any future bad uses of the loopback address in URLs in tests :)
On Thu, Mar 14, 2019 at 12:19 PM Chris Hegarty <chris.hega...@oracle.com>
wrote:
> Arthur,
>
> On 14 Mar 2019, at 17:44, Arthur Eubanks <aeuba...@google.com> wrote:
>
> Most of these eventually turn the String into a URL, so using a URL
> directly should work?
>
>
> URL constructors will work, but I would prefer to not add any more
> usages of these than we already have. I plan to start a discussion about
> the possibility of deprecating these constructors in the not too distant
> future.
>
> -Chris.
>
> jshell> URL u = new URL("http", "::1", 9002, "/")
> u ==> http://[::1]:9002/
> jshell> URL u = new URL("http",
> InetAddress.getLoopbackAddress().getHostAddress(), 9002, "/")
> u ==> http://127.0.0.1:9002/
> jshell> URL u = new URL("http",
> InetAddress.getLoopbackAddress().getHostName(), 9002, "/")
> u ==> http://localhost:9002/
>
> On Thu, Mar 14, 2019 at 9:40 AM Chris Hegarty <chris.hega...@oracle.com>
> wrote:
>
>> It's just a hack, but enough to get me started so I can asses some of
>> what is needed to support running the tests on an IPv6-only environment.
>>
>> http://cr.openjdk.java.net/~chegar/ipv6-only-hack/
>>
>> My /etc/hosts file has entries for `localhost` and the _host_, that
>> return `::1`.
>>
>> This is enough to get started, but it will run out of road quite soon.
>> And it will break some things. Let's work towards getting a better
>> solution.
>>
>> Also, we will need test library support for determining whether the
>> platform supports: 1) IPv4, or 2) IPv6, or 3) IPv4&IPv6. I filed 8220673
>> [1] to tracks this.
>>
>> ---
>>
>> Regarding what to do with URL's in test. I tried a few different
>> approaches, I am leaning towards:
>>
>> String host = loopbackAddress.getHostAddress();
>> if (host.contains(":"))
>> host = "[" + host + "]";
>>
>> Since it:
>> 1) is relatively short and concise, and
>> 2) ensures, in many cases, that the client and server use the same
>> IP address, and
>> 3) clearly shows the actual IP protocol in use, e.g.
>> http://[0:0:0:0:0:0:0:1]:46009/zero/xxyy
>>
>> A sample of some tests that have been updated as above:
>> http://cr.openjdk.java.net/~chegar/8220575/webrev.0.1/ (not too bad!)
>>
>> -Chris.
>>
>> P.S. I ran into an implementation bug and filed, 8220663: "Incorrect
>> handling of IPv6 addresses in Socket(Proxy.HTTP)" - the fix is straight
>> forward.
>>
>> [1] https://bugs.openjdk.java.net/browse/JDK-8220673
>>
>
