Re: more on the simplified feed format

[Graham <dtcd@xxxxxxx>]

A few eTag-related points:
- The eTag needs to be included in the linked entry document, for robustness. The client obviously needs to recall the eTag of the version it has downloaded, and requiring it to infer that from the index strikes me as shoddy.

- Ideally the eTag should chnage at instaneously the same moment as the linked entry. This isn't always possible - eg If the feed is made of static files, they need to be generated in some order. There's a big problem for a client if it's possible to think an entry has been updated, but then end up downloading the same file. Therefore the spec must mandate that the new entry is already available for download when the eTag in the index changes, ie That the index is always rebuilt last. That way, if the index ETag doesn't match the AtomEntry documents ETag, the client can assume it has a newer version than when it downloaded the index.

- Shouldn't ETag be on the <link> element itself, it refers to the linked document, not to the SSFF <entry>?

- eTag as SHOULD rather than MUST is OK, but the spec needs to make clear that without it, a client will have little choice but to re-download every entry every time, and that even if that's no problem for the server, it could affect client performance dramatically.

- "an opaque string which MUST change each time the associated AtomEntry document is updated" - Does this mean "each time *any part* of the associated AtomEntry document"? If so, it should say that.


I think Searth has a point about aggregators ignoring attributes in the feed index. I know I will. Let's say someone puts in a title. Where am I going to put it in a list view sorted by date? If I just shove them at the top, the UI will resort horribly as each entry is loaded. And if you add dates to the index feed, what if the user has decided to sort by author? Or some other attribute? I don't see the point of adding complexity like this, especially when I'm going to have loaded the full entry half a second later.

(also SSFF entires aren't vaild Atom entries without titles are they? Again, I don't see the point of not telling parsers up front "This is not the full entry")

Graham