Timebomb

[klacke@REDACTED]

On Thu, Oct 28, 2004 at 11:35:30AM +0200, Carsten Schultz wrote:
> Hi Peter!
> 
> On Thu, Oct 28, 2004 at 11:26:23AM +0200, Peter L wrote:
> > So it seems indeed that it is the HTTP 1.0 YAWS implemen- tation
> > that needs improving...
> 


Not very likeley. We're using yaws in Nortel VPN products
shipping pages do gazzilions of ultra-picky clients.


> 
> Yaws itself is working well with HTTP/1.0.  (Otherwise none of my
> sites would show up in Google.)  As klacke pointed out, the reverse
> proxy code is probably having problems and is not recommended for use
> in real servers.
> 


It's chunked encoding in the reverse proxy that generates
one CRNL too little. The site is running yaws 1.42 which is
kinda semi old, 6 months and I checked the changes between now
and 1.42 and didn't see any obvious fixes in the revproxy code that 
could explain the missing CRNL.

I'm running revproxy myself (In my house) and  don't have those
CRNL problems ... I'm running the latest release though. As I said,
it needs debugging and the revproxy is (as I wrote) still experimental.


Besides, running different browsers to check the validity of
HTTP servers is generally a bad idea. The different browsers don't 
just render broken HTML, but also do a very good (or maybe one should
say bad) job to eat broken HTTP.

So just because one browser renders a page from a HTTP server, that 
doesn't mean anything. My prefered choice of tool here would be 
/bin/nc + possibly curl




/klacke




-- 
Claes Wikstrom                        -- Caps lock is nowhere and
http://www.hyber.org                  -- everything is under control