On Nov 5, 2004, at 3:55 PM, Tim Bray wrote:
While I grant the force of Roy's arguments, I have two points of disagreement. First, my experience of feed errors does not accord with his hypotheses; they are often in places that can be deterministically detected by a client and helpful feedback provided.
I wouldn't disagree with that -- my point was that sending it in an automated HTTP message means it will go to the wrong person.
Second, and this is crucial, the error reporting is completely at the discretion of the server. The server can, if and only if it chooses, provide a ServiceError URI. If so, it's presumably ready to get reports and will do something more useful than dump them into logs/error_log.
How is that configured? Is it an httpd config setting, a script setting,
or data within the feed heading? Is it somewhere that a webmaster
supporting several thousand feeds is going to be able to control?
How does that tie in to managing a separate resource per feed for
the purpose of accepting error notifications. How does that information
get to the right person who can fix the software? How is that consistent
with other folks claiming that the notification contains no data on
what the error was, let alone how to fix? How many web servers do you
know of that allow anyone on the Internet to write unknown amounts of
data to a location somewhere other than a controlled logs directory?
As opposed to, for example, just encouraging readers to send a friendly comment to the person whose writing is being syndicated.
Quite frankly, I find this whole discussion to be rather silly. Keep it simple and you will succeed -- bog it down with a bunch of questionable features and people like me won't even bother to read the specification (and I read quite a few more than the average software developer).
Someone who is concerned about the quality of their feed will test that feed with clients that can point out errors, assuming there exist client authors who don't have MSIE disease and are advised that showing errors is a good thing.
Someone who doesn't care about the quality of their feed will neither test it nor enable something like ServiceError.
The only thing that automated error reporting accomplishes is to make the protocol harder to completely implement, while encouraging a bunch of well-meaning authors to waste bits by sending error URIs for feeds that they already know are valid.
Anyway, I've said my bit and feel no more need to elaborate. I won't implement it.
....Roy
