Re: nosgml options

On 25 Jan 2003, Ville Skyttä wrote:

> 
> On Fri, 2003-01-24 at 01:07, Nigel J. Andrews wrote:
> 
> > My little application is in perl. It sends the http request to the validator
> > and then reads the response using the standard perl construct of
> > while (<fh>) ... The first returned line was processed after 2 seconds but the
> > loop only finished 16 seconds after that. Adding the header Connection: close
> > to the request caused the loop to terminate after only 2-3 seconds.
> 
> Your app is most likely requesting Keep-Alive.

Keep-Alive is default behaviour with HTTP/1.1, but not with HTTP/1.0.
What does your Client label its request as?

>	  The validator.w3.org
> setup supports it, but the problem is that it doesn't send a
> Content-Length header.  So, your client doesn't know where the response
> ends, and must wait until the connection terminates; that's when the
> while loop ends.  And this is determined by the Keep-Alive timeout,
> which happens to be 15 seconds on validator.w3.org.

All this is Apache's default behaviour.

>	  See [1] for sample
> headers.  Common browsers must have some black magic installed to
> determine the end of a response, since they don't exhibit the 15 sec
> delay.

They support Chunked encoding properly.  This delay is one reason you
shouldn't send a request claiming to be HTTP/1.1 unless your client
really groks it.

> Yes, validator should send the Content-Length header.  However,

With HTTP/1.0 that would have some merit.  With 1.1 it's irrelevant.

> [1] Here's a sample how v.w.o responded to my browser a few minutes ago:
> 
> HTTP/1.1 200 OK
> Date: Sat, 25 Jan 2003 11:40:24 GMT
> Server: Apache/1.3.27 (Unix)
> Content-Language: en
> Keep-Alive: timeout=15, max=100
> Connection: Keep-Alive
> Transfer-Encoding: chunked
> Content-Type: text/html; charset=utf-8
> [...]

That is chunked encoding, which means the response is sent in chunks
(size determined by Apache's output buffering).  Each chunk is labelled
so that HTTP/1.1 clients know when the response is complete.  This is a
good second-best when it is impractical to set content-length -
e.g. because (as with validator) it would be expensive to wait for all
the content to be generated before starting to send it. 


-- 
Nick Kew

Received on Saturday, 25 January 2003 12:48:13 UTC