On Fri, 1 Jul 2016, Dan Fandrich wrote:

>> Mentioned here to allow you to protest, agree or discuss bikeshed colors.
>
> This has the possibility of increasing the number of packets sent for each
> request, depending on how libcurl does its writes for various protocols. Has
> anyone looked at this possible effect? For example, if headers are written
> individually as "Header:","contents" then an HTTP request that's sent in one
> packet today could end up taking 20 with this patch. And it's not just HTTP
> but all protocols that could be affected similarly.

Correct, but libcurl sends all headers in one write so it should have minimal
impact on this scenario.

It will actually fix a related performance problem we've seen in the past when
doing several small requests over a persistent HTTP connection as then libcurl
would be slow on the subsequent HTTP requests simply because of not setting
TCP_NODELAY thus inducing short delays.

-- 
  / daniel.haxx.se
-------------------------------------------------------------------
List admin: https://cool.haxx.se/list/listinfo/curl-library
Etiquette:  https://curl.haxx.se/mail/etiquette.html