RFR 6563286 6797318 8177648 - Undeclared IAE thrown from HttpURLConnection.connect for some URLs

[Michael McMahon]

Hi Jaikiran,

Thanks for looking at this issue. ProxySelector::select is not
a particularly well specified method. The spec is a little vague
on what is expected of the supplied URI parameter.

It seems to me however, that the intent is that the supplied URI should
contain a valid protocol (scheme) and host, even though that is
perhaps not stated clearly enough. Though it does refer to
the protocol and the destination address as the components
it is looking for.

The concern I have with the suggested fix is that it appears
to contradict that implicit requirement.

I wonder if another solution is possible where the IAE is caught
at the appropriate place(s) and converted to an IOException
which is what users are expecting.

Michael.

On 19/07/2019, 16:14, Jaikiran Pai wrote:
> Hello,
>
> Could I please get a review and a sponsor for a patch which fixes an
> issue in sun.net.spi.DefaultProxySelector? The patch is available as a
> webrev at
> http://cr.openjdk.java.net/~jpai/webrev/defaultproxyselector/1/webrev/
>
> The issue has been reported in multiple different JBS issues[1][2][3]
> all coming back to the same root cause. There probably are other similar
> JBS issues related to this. The root cause is that the
> DefaultProxySelector throws IllegalArgumentException when the passed URI
> is non-null and thus contradicts its javadoc which (only) states the
> IllegalArgumentException is thrown when the URI is null.
>
> The change in the patch, now returns a Proxy.NO_PROXY when either the
> protocol or the host of the passed URI is null. This is the same as
> what's suggested in JDK-6797318.
>
> In addition to the change to DefaultProxySelector, this patch also
> changes sun.net.www.protocol.http.HttpURLConnection and
> sun.net.www.protocol.ftp.FtpURLConnection. Just before invoking the
> ProxySelector.select(...), both these classes use
> sun.net.www.ParseUtil.toURI(...) which can return null. The changes to
> these classes now ensure that the null isn't passed to the selector (to
> avoid a IllegalArgumentException) and instead is handled properly.
>
> 2 new test classes have been added in the patch. These tests fail (as
> expected) without the change to the source classes above and pass with
> these changes. After these changes, locally, I have also run the
> existing jtreg tests under test/jdk/java/net/HttpURLConnection and
> test/jdk/sun/net/www/http/HttpURLConnection/ and they have passed
> without regressions.
>
> I haven't added anything new for the FTP testing, since I didn't find a
> easy way to do that (based on my very limited search of existing tests).
>
> [1] https://bugs.openjdk.java.net/browse/JDK-6797318
>
> [2] https://bugs.openjdk.java.net/browse/JDK-8177648
>
> [3] https://bugs.openjdk.java.net/browse/JDK-6563286
>
> -Jaikiran
>
>