RFR [14] 8231504: Update networking tests to avoid implicit dependency on the system proxy

[Daniel Fuchs]

Hi Chris,

Thanks for doing that. It's similar to what we've also been doing
in each tests that was modified to use the wildcard.

I wonder if at some point we should also try to add an additional
@run main/othervm -Djava.net.preferIPv6Addresses to each test that
doesn't have it - but that will be for another day ;-)

It might also be good to have (some) of the tests use
ProxySelector.setDefault(ProxySelector.of(null)) so that
URL.openConnection() continue to be tested - but we already have
some tests that do that so I'm not too worried.

The fix looks good to me!

best regards,

-- daniel

On 26/09/2019 14:04, Chris Hegarty wrote:
> A number of tests in the networking area invoke the no-args
> URL::openConnection. If support for system proxies is enabled, the
> default on macOS, then the built-in http(s) and ftp protocol handlers
> may attempt to make a connection to a proxy. In almost all cases, this
> is the wrong thing for the test to do. Typically, the test is
> exercising something in the protocol handler implementation, rather than
> proxy support ( there are other tests that exercise proxy support ).
> This is an accidental dependency.
> 
> In many cases the connection to the system configured proxy may be
> benign, but not always, it depends. Such accidental dependencies should
> be remove, since they serve little purpose and could negatively affect
> test stability.
> 
> Webrev:
>    http://cr.openjdk.java.net/~chegar/8231504/webrev.00/
> 
> This change may not address all such scenarios in the tests, but it does
> address a large number of them ( as observed when running on my local
> system with a promiscuous proxy configured ). We've done similar efforts
> in the past.
> 
> -Chris.
>