RFR: 8368249: HttpClient: Translate exceptions thrown by sendAsync [v3]

[Jaikiran Pai]

On Thu, 16 Oct 2025 11:37:16 GMT, Daniel Fuchs <dfuchs at openjdk.org> wrote:

>> Although it might look strange - we might want to wrap the UncheckedIOException instead of peeling it off.
>> If the UncheckedIOException is generated by us (HttpClient implementation) then we most likely want to peel it off. But if it originates from custom code, peeling it off might hide important stack trace information.
>> I am actually hesitating between the two possibilities.
>> 
>> We see here that peeling of the UncheckedIOException from within sendAsync forces you to modify one test, where the UncheckedIOException was originating from custom code. I am not 100% sure we want do that.
>> 
>> On the other hand - we don't want to add yet another version of toIOException.
>> 
>> Let me think about this a bit more.
>
> OK - I think what you have is fine. Let's peel of UncheckedIOException. In most cases I could see the UncheckedIOException is only used to smugle an IOException, and in most case it's created at the point where the IOException is created too. That let me think that the extra stack trace will not be needed in most of the cases. We could revisit possibly later in a followup if we have reason to...

I think, not propagating the original `UncheckedIOException` would be a bug. Imagine this application code:


final Future<HttpResponse<String>> future = client.sendAsync(req, new BodyHandler<String>() {
    @Override
    public BodySubscriber<String> apply(ResponseInfo ri) {
        throw new UncheckedIOException("application specific message", new IOException("some other message"));
    }
});
final HttpResponse<String> resp;
try {
    resp = future.get();
} catch (ExecutionException e) {
    final Throwable cause = e.getCause();
    // cause must be UncheckedIOException and its message must be "application specific message"
    assert ...
}

I think the application should receive the `UncheckedIOException` as the top level cause of the `ExecutionException` and that `UncheckedIOException` must contain the original cause and the original message that was there when the application raised it. Not passing along that to the application, I think, would be wrong.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/27787#discussion_r2451852130