On Mon, 2018-12-24 at 01:05 +0100, Michael Osipov wrote:
> Am 2018-12-23 um 23:21 schrieb Oleg Kalnichevski:
> > On Sun, 2018-12-23 at 22:07 +0100, Michael Osipov wrote:
> > > Folks,
> > >
> > > why does a minimal client:
> > > > HttpClientBuilder bulder =
> > > > HttpClientBuilder.create().disableContentCompression();
> > > > CloseableHttpClient client = bulder.build();
> > > > CloseableHttpResponse resp = client.execute(new HttpGet("
> > > > https://google.de";));
> > > > EntityUtils.consumeQuietly(resp.getEntity());
> > >
> > > still send the "Connection: keep-alive" request header? This
> > > seems
> > > like
> > > wasted bytes for HTTP/1.1. Is that really for backwards compat?
> > > Aren't
> > > we good to break (remove) this in 5.0?
> > >
> >
> > It is mostly for the sake of compatibility. I am fine with dropping
> > the
> > header for HTTP/1.1 connections, though I see no harm in keeping
> > it.
>
> Right, there is no harm, but also no benefit and just wasted bytes
> for
> HTTP/1.1.
Wasted? Why? What matters is not how long the message is but whether or
not it fits into a single IP frame.
As far as I can see, it is RequestConnControl. I have created a
> branch
> (drop-http11-keep-alive) with a first shot of changes. I am not yet
> happy with the remaining test method names because they do not
> reflect
> the truth now. I am also not really happy with "Proxy-Connection"
> because RFC 7231 says in A.1.2: "As a result, clients are encouraged
> not
> to send the Proxy-Connection header field in any requests." Is that
> good
> reason to drop this "Proxy-Connection" fiddling in client and core
> altogether?
>
We can drop 'Proxy-Connection' entirely.
Oleg
> Show I also apply this change to
> httpcore5/src/main/java/org/apache/hc/core5/http/protocol/RequestConn
> Control.java?
>
> Michael
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
