On Tue, 26 Nov 2024 10:33:21 GMT, Daniel Fuchs <dfu...@openjdk.org> wrote:
>> test/jdk/java/net/Socket/CloseOnFailureTest.java line 345:
>>
>>> 343: SocketAddress serverSocketAddress =
>>> serverSocket().getLocalSocketAddress();
>>> 344: try (Socket socket =
>>> createConnectedSocket(serverSocketAddress)) {
>>> 345: assertReconnectFailure(() ->
>>> socket.connect(REFUSING_SOCKET_ADDRESS));
>>
>> I needed to delegate the assertion to the implementation because when I call
>> `connect()` on a `Socket` created using
>> `SocketChannel.open(address).socket()` and `new Socket(address, port)`, the
>> former throws an `AlreadyConnectedException`, whereas the latter throws a
>> `SocketException`. Is this discrepancy expected?
>
> There are some subtle behaviour discrepancies between plain Socket (created
> with new Socket()) and the SocketAdaptor (created from
> SocketChannel.socket()). SocketChannel::connect will throw
> AlreadyConnectedException, and so that's what the socket adaptor also does.
> Ideally we'd want to reduce the distance between plain Socket and the Socket
> adaptor - but changing behaviours in these APIs is always tricky and a bit
> risky (and can require a lot of efforts). FWIW - SocketChannel does the right
> thing - but we're stuck with SocketException in the Socket API due to
> compatibility reason.
Got it. Thanks for the explanation.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/22160#discussion_r1858251746
