On Tue, 19 Sep 2023 23:36:42 GMT, Chris Plummer <cjplum...@openjdk.org> wrote:
> I think Alex has commented that this can be made more obvious with a check
> for AF_UNSPEC.
Thanks, yes, I have applied that suggestion.
> > For 748, it's handling a situation where IPv4 isn't specifically preferred.
> > The implementation comment notes 'if preferredAddressFamily is AF_INET6 or
> > not set'.
>
> What does "specifically preferred" mean? By default it will be set to
> AF_INET. If it is set to AF_INET, you can't tell if
> `java.net.preferIPv6Addresses=false` was specified (which implies AF_INET),
> or if nothing was specified and it just defaulted to AF_INET. The comment and
> the code read as if preferredAddressFamily will be something other than
> AF_INET unless `java.net.preferIPv6Addresses=false` is used. Maybe the code
> is correct and the comment needs some cleaning up. TBH I don't understand
> what this code is suppose to do.
I agree 'preferredAddressFamily is AF_INET6 or not set' is unclear, since
`preferredAddressFamily` was only ever `AF_INET` or `AF_INET6`. I wondered if
the use of `if (preferredAddressFamily != AF_INET) {` instead of `if
(preferredAddressFamily == AF_INET6) {` and the comment about 'not set' was
referring to the possibility of `AF_UNSPEC`. The history for this logic is in
* https://bugs.openjdk.org/browse/JDK-8250630
*
https://mail.openjdk.org/pipermail/serviceability-dev/2020-August/thread.html#32673
The bug mentions it fixed an issue on Alpine Linux, and I don't have convenient
access to a test environment for that.
I could update L750 to `if (preferredAddressFamily == AF_INET6) {` to prevent
this logic from re-ordering address when `java.net.preferIPv6Addresses=system`
is set.
What do you think?
-------------
PR Comment: https://git.openjdk.org/jdk/pull/15796#issuecomment-1726692404
