Re: #147: header-specific canonicalisation

Mark Nottingham wrote:
> Proposed replacement:
>      - canonicalising both headers' values in a way that is known to have identical
>         semantics, according to the headers' specification (e.g., re-ordering field values
>        when order is not significant; case-normalisation, where values are defined to be
>        case-insensitive)
>    
It would be good to provide recommended canonical forms for 
headers--recommended case, recommended order, recommended whitespace, 
recommended (non-)quoting/escaping, etc. Otherwise, it opens the door 
for non-interoperability in implementors understanding of "identical 
semantics." Plus, a recommended canonical form would allow 
implementations to work better even with caches that don't do any 
header-specific canonicalization. If the working group isn't able to 
agree on canonical forms for headers, then it probably wouldn't be 
possible for everybody to agree on whether two different forms have 
exactly identical semantics either.

Header canonicalization also has benefits for compression of the header 
too. That's (at least one reason) why SPDY requires case normalization.

- Brian

Received on Thursday, 4 March 2010 16:51:32 UTC