Wrap uninformative Java exceptions in ConnectionFailure

[rossabaker]

See http4s/blaze#373.

Java's UnresolvableAddressException does not carry any information about the host that failed to resolve. If a client fails to connect, we should provide context on which host we failed to connect to. We do so by introducing a wrapper exception, ConnectionFailure, in the client package.

Other clients should probably use this if they can. Fixed for blaze, as it was reported on blaze.

[tomasherman]

awesome, thanks!

[tomasherman]

is there any chance of having this backported into the old build? Or is there an ETA on next version of http4s going stable?

[rossabaker]

I'm working on 0.21-0-RC1 right now. It does subtly change the behavior, so I'm a little reluctant to backport it into 0.20, though it's difficult to imagine anyone is depending on catching the Java exception. I could be talked into it, because I'm also preparing an 0.20 backport release right now.

[tomasherman]

If it's not too much trouble, it would be helpful for us. BTW just a thought... Would it make sense to extend the original exception instead of ioexception to make this backwards compatible?

…

On Sat, Jan 18, 2020, 18:24 Ross A. Baker ***@***.***> wrote: I'm working on 0.21-0-RC1 right now. It does subtly change the behavior, so I'm a little reluctant to backport it into 0.20, though it's difficult to imagine anyone is depending on catching the Java exception. I could be talked into it, because I'm also preparing an 0.20 backport release right now. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#3078?email_source=notifications&email_token=AAAYMHXFBYROQLW7VRDYHQLQ6M3L5A5CNFSM4KIRFMC2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJJ5SFQ#issuecomment-575920406>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAYMHSZL3UBMCG5XSPB45TQ6M3L5ANCNFSM4KIRFMCQ> .

[rossabaker]

There are many Java exceptions that can be thrown here, and we can't dynamically mix a ConnectionFailure into all of them. But we could handle this one and throw a subtype that overrides the message in 0.20, and stick with the new wrapper in 0.21.