* James Aylett <[EMAIL PROTECTED]> [2005-08-04 15:25]:
> On Thu, Aug 04, 2005 at 02:42:31PM +0200, Bjoern Hoehrmann wrote:
>
> > * Tim Bray wrote:
> > >"Implementors are advised that there is a common class of
> > >error in [...]
> >
> > Sorry but this is ridiculous; if we say X MUST Y even though
> > we know that many X won't Y we are abusing RFC 2119
> > terminology and make it much more difficult to evangelize
> > 100% compliance, since this allows people to argue that
> > compliance with this particular requirement is not relevant
> > in practise so they can worry less about compliance in
> > general.
>
> +1
>
> […]
>
> We don't spell out what Atom consumers should do if they
> encounter a Date construct with "Last Thursday" in, either.
> Pointing out the common error AS AN ERROR should be sufficient
> IMHO.
+1 to James and Bjoern.
There Is Only MAY, SHOULD And MUST.
I want to remind everyone that the spec’s job is to say exactly
what the bits mean, rather than what people should do with them.
It therefore strikes me as foolish to be coy about what
constitutes an error. Telling implementors that they KINDA MUST,
PROBABLY comply serves noone. An error is an error is an error.
Regards,
--
Aristotle Pagaltzis // <http://plasmasturm.org/>
