JDK 8 RFR 8010371: getaddrinfo can fail with EAI_SYSTEM/EAGAIN, causes UnknownHostException to be thrown

[Alan Bateman]

On 10/10/2013 21:24, Brian Burkhalter wrote:On 10/10/2013 21:24, Brian 
Burkhalter wrote:
> :
> Assuming that the general idea is correct, there are a few possible alternatives for a final version for Windows. I am not concerned in this discussion about picky details including naming which can be fine tuned at the end.
>
> 1) In net_util_mh.c/NET_ThrowUnknownHostExceptionWithGaiError replace gai_strerror() with FormatMessage().
>
> 2) In Inet4AddressImpl.c and Inet6AddressImpl.c replace NET_ThrowUnknownHostExceptionWithGaiError with NET_ThrowByNameWithLastError (see net_md_util.c).
>
> 3) In net_md_util.c change (line 97 in http://cr.openjdk.java.net/~bpb/8010371/webrev.3/ version)
>
>      { WSATRY_AGAIN,             0,      "Nonauthoritative host not found" },
>
> to
>
>      { WSATRY_AGAIN,             "UnknownHostException",      "Nonauthoritative host not found" },
>
> and in Inet4AddressImpl.c and Inet6AddressImpl.c replace NET_ThrowUnknownHostExceptionWithGaiError with NET_ThrowNew().
>
> 4) In NET_ThrowUnknownHostExceptionWithGaiError instead of using gai_strerror() use the hard-coded error message "Nonauthoritative host not found".
>
> The pros and/or cons for each, respectively, are
>
> 1) Most closely matches the Unix version and avoids hard-coded messages.
>
> 2) Simple and obviates the need for NET_ThrowUnknownHostExceptionWithGaiError(), but the exception message has a different format from the Unix version and contains the error code instead of an error string.
>
> 3) Simple and obviates the need for NET_ThrowUnknownHostExceptionWithGaiError(), but the exception message format is slightly different from the Unix version and the change could have the undesirable (and unpredictable) side effect of an UnknownHostException being thrown somewhere else where a SocketException used to be thrown.
>
> 4) Simplest option but has distasteful hard-coding (as do option 3 indirectly and NET_ThrowNew() for that matter).
>
> If the "con" of option 2 is acceptable then I think that would be the best way to go, otherwise option 1.
>
Option #2 seems reasonable, the exception messages for similar network 
conditions are rarely the same on Windows and Unix anyway. However I 
think it's important to have verified it with one or two errors to be 
confident that the errors translate as expected.

One other thing to add is that winsock_errors dates from early versions 
of Windows whether there wasn't a means to translate Windows Sockets 
errors. We should look at eliminating it (not for JDK 8 of course, it's 
too late) so that all errors are handle translated consistently.

-Alan