Re: Feedback on content transformation guidelines ( LC-2066 LC-2067 LC-2068 LC-2069 LC-2070 LC-2071 LC-2073 LC-2072 LC-2074 LC-2075 LC-2076 LC-2077 LC-2078 LC-2079 LC-2080 LC-2081 LC-2082 LC-2083 LC-2084)

Luca Passani wrote:
> Francois Daoust wrote:
>> For clarification, the guidelines do not build on the assumption that 
>> GET is not safe.
>>
>> The mechanism described by Luca is actually recommended by the 
>> guidelines: send a GET with original headers, then send a request with 
>> modified headers if the first response is a "request unacceptable" 
>> response.
> 
> Francois, this is not what I meant. What I meant is "content tasting". 
> Proxies should send a GET with original headers and if they get a 
> response (which they probably will), they should smell the response and 
> figure out whether that content may be good enough for mobile (and err 
> on the side of assuming it is). If the content is likely to be OK for a 
> mobile device, no transcoding should take place at all.

Yes. That is precisely what I had understood.


> This is explicitly ruled out by 4.1.5.1:
> 
> http://www.w3.org/TR/ct-guidelines/#sec-content-tasting
> 
> 
> "The theoretical idempotency of GET requests is not always respected by 
> servers. In order, as far as possible, to avoid misoperation of such 
> content, proxies *should* avoid issuing duplicate requests and 
> specifically *should not* issue duplicate requests for comparison 
> purposes."

Yes. This is where the problem is and I personally agree that the 
document should be fixed (either by removing use of normative terms or 
by removing the paragraph). The intent was merely to alert transcoding 
proxies about practice vs. theory, and not to rule out content tasting 
that the guidelines explicitly encourage.

[...]

Thanks,
Francois.

Received on Tuesday, 24 November 2009 15:25:37 UTC