[rfc][icedtea-web] disconnecting httpurlconnections

[Andrew Azores]

On 05/02/2014 10:25 AM, Omair Majid wrote:
> * Jiri Vanek <jvanek at redhat.com> [2014-05-02 09:34]:
>> I will elaborate a bit more, but if you may say few more words, it
>> would be nice:((
> Sorry, but I know next to nothing about this area. All I know is that
> Adam looked into it and decided that disconnect() is best avoided.
>
> Thanks,
> Omair
>

Maybe I'm just missing something but I really don't see how he came to 
the conclusion that disconnect() should be avoided... the initial 
problem seems to be that a new connection was being made and only the 
response code being read, nothing else. The proposed solution was to do 
this but then close the connection. The comment Adam refers to states 
that rather than simply disconnecting after reading the response code, 
the connection should actually do something - consume and discard it, or 
simply use a HEAD request in the first place - so as to not confuse 
these "mediocre" web servers (which presumably are expecting a GET 
request to actually want the response body, so they hang on sending it 
unless interrupted by a disconnect). Adam himself even says that it's 
difficult to actually reuse connections, which is AFAICT the reason they 
weren't disconnected to begin with. Why would it not be correct to both 
(1) use HEAD request with consume/discard fallback, (2) disconnect when 
finished, if the connection is not going to be reused later? As Jiri 
says, is the overhead incurred by creating new connections really that 
bad that we should deal with evidently flaky connection reuse?

Thanks,

-- 
Andrew A