I’ve had comments running for a few days here now (I prefer to say “contributions”, but whatever). People are irritated at me because an ongoing fragment shows up as unread in their feed-reader whenever a new comment comes in. I’m not sure what the right thing to do is. This piece outlines a few options and asks the community for discussion.
The Options · Here are some of the things that could be done when a new comment is contributed:
Change the feed entry’s
updatedtimestamp.Include the comment count in the entry’s
summary.Include the comment count in the entry’s
content.Include the comments themselves in the entry’s
content.Build a per-entry comments feed.
Build a global ongoing comment feed (this already exists, I just haven’t figured out where to point to it from).
Any others that I’ve missed? I’ll add plausible suggestions to the list above.
The Discussion · Have at it in the comments, but if you want to say something long and linky, you might want to do it in your own web-space and leave a pointer.
(Oh, by the way, comment authoring is apparently broken in IE6. I really must get around to fixing that, but for the moment go grab the nearest Firefox if you’re an IEnaut.)
My Feelings ·
When I update a fragment, I always bump the updated field and
put a notice in the summary so it’ll show up in the feed; I can suppress this,
but I have to push a special button; the default is to show the change.
To me, it seems really arrogant to do this for my own changes but not for
the contributions of others.
Contributions
· The World (148 fragments)
· · Life Online (273 more)
· Technology (90 fragments)
· · Publishing (160 more)
Comment feed for ongoing:
From: James Snell (Oct 06 2006, at 12:41)
Personally, I would not bump atom:updated whenever a new comment comes in. Provide a separates comment feed for folks who want to know whenever a new comment arrives. A single feed for all comments should be sufficient. Use RFC4685 and provide a comment count in your base feed but don't bump atom:updated whenever you update the counter.
[link]
From: rick gregory (Oct 06 2006, at 12:43)
A comment isn't an update to the content of the item, so new comments should not act as updates to the main item in the feed.
Ideally, your system would allow a feedreader to differentiate between new items, updated items and items that have not been updated. From a (human) reader's point of view, I like a link to the comments along with a count of them, (see techcrunch's feed), but I don't >need< that - I can use Cocomment to get notifications of new comments in most places. Or I can check back...
Interestingly, I don't see this issue in Google Reader.
[link]
From: Craig Andera (Oct 06 2006, at 12:49)
#5 and/or #6, please. Or leave them out of the feeds altogether - I don't generally care much about comments, and when I do, I'll come by and look at the web page. About the only time comments are interesting in a feed to me is when I've left one and I want to keep tabs on the responses.
[link]
From: Scott Johnson (Oct 06 2006, at 12:51)
When a comment is added, the entry itself isn't really updated--it merely has more comments. Well, that's how I see it anyway. It all depends on what you consider an entry to be. I see it as your text. You may see it as your text plus our comments. I like the global and per-entry comment feed idea, but then you end up with a gazillion feeds that are being polled every n minutes.
[link]
From: Dare Obasanjo (Oct 06 2006, at 13:01)
Why don't you support http://ietfreport.isoc.org/all-ids/draft-snell-atompub-feed-thread-12.txt like Sam's feed does? That seems to be the most correct option instead of all the other options you suggest.
[link]
From: Antone Roundy (Oct 06 2006, at 13:02)
I'm subscribed to what YOU have to say, not to what everyone else has to say about what you have to say. I'd prefer that the updated timestamp NOT be updated with every comment so new comments won't cause my feed reader to hilight an entry.
Per-fragment comment feeds would enable people to follow the discussion of particular fragments, but I wouldn't imagine many people will want to monitor EVERY fragment. But multiple comment feeds would create one problem--as people subscribed to more and more of them, your site would get hit with more and more traffic, much of which would be for feeds that, after the initial burst of discussion, never get updated. Last-Modified end ETag headers would mitigate the impact, but wouldn't eliminate it. And not all feed readers support those headers. I don't know whether you'll be closing comments on old fragments or not, but if you do, some method of marking a feed closed to tell feed readers to stop polling it would be useful. More of my thoughts on the topic can be found at http://antone.geckotribe.com/alpha-gecko/2005/08/02/marking-a-feed-closed/
[link]
From: Robert Sayre (Oct 06 2006, at 13:04)
I don't think it would be arrogant at all. I don't subscribe to Ongoing because I want constant updates on Smalltalk refugees arguing with Eclipse developers about the feasibility of features that Intellij is already shipping.
Do a comments feed.
[link]
From: Jacob Kaplan-Moss (Oct 06 2006, at 13:11)
I'd say I'm in the don't-update-the-feed-when-you-get-comments camp.
Honestly, I subscribe to feeds because I want to know what the authors have to say, not what their readers have to say. Often, I'm completely unintersted in any followup a peice might recieve.
For the times when I do care, I really like the per-entry comment threads; I've got a folder in my feedreader for threads I'm following, and it works really well. Much of time I don't bother reading comments unless I'm particularly interested in the thread.
[link]
From: Evan DiBiase (Oct 06 2006, at 13:12)
If you consider contributions to be part of each fragment, I think the content of the contributions belong in the feed's content for that fragment, with the corresponding updated-date bump. I would be curious to see how this approach would pan out. It doesn't seem like it would scale well for large numbers of comments spanning long periods of time, but, since ongoing entries seem to garner a few very good comments, it could be a good fit.
[link]
From: Sam Ruby (Oct 06 2006, at 13:14)
7. Include the comment count in either a slash:comments element or a thr:count attribute.
If you do go with changings the feed's updated count, consider appending the contributions themselves into your atom:content. To me the atom:updated element indicates when this atom:entry was updated, and atom:content is the content of the atom:entry.
Also, think for a moment about the meaning of atom:author and atom:rights.
[link]
From: Marcus (Oct 06 2006, at 13:20)
I will have to agree with what most everyone is saying. I subscribe to the feed because I'm curious of your writing, anc I assume most others do to. In general, I really don't read comments so I just find it a neusance when the fragment gets updated.
I'd go for a great big comment feed or anything else that is as non-obtrusive as possible.
[link]
From: Justin (Oct 06 2006, at 13:23)
I'd prefer not to have my feed-reader mark entries as unread everytime a comment is added to an entry. However, leaving atom:updated unchanged but modifying the atom:summary to reflect the comment count should produce better behavior. Compliant feed-readers won't mark the entry as unread, as atom:updated hasn't changed but the summary will reflect the new comment count should the entry be read.
[link]
From: Ian Brown (Oct 06 2006, at 13:27)
I don't think a massive comment feed would be useful. Ideally, if I was inclined to follow comments on a given post, I would prefer to subscribe to a feed of comments about that specific post, not across the entire blog.
[link]
From: Rob (Oct 06 2006, at 13:32)
The comments aren't part of the feed or its entries, so they shouldn't make entries appear as new.
As background, the main approaches to feeds for weblogs seem to be:
1. Just include entry summaries in the feed; actual entries are web pages, and you just include a pointer.
2. Include the beginning of an entry in the feed, so that readers can start reading and decide whether to visit the web page for the whole thing.
3. Include the whole entry in the feed, so that readers never need to leave their feed-reader.
You can make an argument that comments are an intrinsic part of the entry if you've chosen options 1 or 2, but ongoing is in category 3. I expect I'm in the majority in reading the feed and visiting the web pages (and reading comments) only when an entry particularly captures me. Annoying such users by remarking exactly the same feed content as "new" is just obnoxious.
Of course, if I visit the web page for an entry, then that page could (and IMO should) offer a feed for comments to the entry. I admit that current technology isn't particularly well-suited to managing hundreds or thousands of feeds, but given my reading habits (only a handful of entries a month where I'd bother watching the comment feeds) this system would be my preference.
[link]
From: Kevin Scaldeferri (Oct 06 2006, at 13:34)
I'm going with the crowd and cast another vote for 5, and maybe 6. This seems to be the most typical behavior on other blogs, so the principle of least surprise applies. The current behavior had me looking for updates to your content the first couple times, but has now trained me to pretty much just scan subject lines and ignore anything I remember seeing before. Consequently, I'm liable to miss it if you update the content of a fragment, and potentially I'll miss an entry entirely if I misremember the subject as seen-already.
[link]
From: Jesper (Oct 06 2006, at 13:42)
In my opinion - noone else's is mine to offer - the *count* of the comments is an irrelevant bit of information to include in the feed. Either you provide a static comment link or include the comments inline. Either they are contributions or they aren't.
Consider offering one feed with one static, never-changing comments link and one with comments inline.
[link]
From: Norman Walsh (Oct 06 2006, at 13:48)
I don't feel strongly about which of the solutions you adopt, except that I hope Bloglines will not keep showing me the same entries over and over again. I'll happily subscribe to a different feed if that helps.
Sometimes I'm going to care about the comments, but not always and only very rarely in real time.
[link]
From: Zellyn Hunter (Oct 06 2006, at 14:16)
Yours was one of the few feeds I subscribe to where I have bloglines set to show updated posts instead of ignoring them. The reason is that you're always clear about what the update is, and mark it in the first paragraph in a way that is easy to scan for. Until bloglines implements highlighted diffs, I'm going to have bloglines ignore changed posts on your blog - it just takes too long examining each post to figure out whether it's new, changed, or just commented on.
[link]
From: Justin Watt (Oct 06 2006, at 14:24)
Tangential observation: Other than administrivia like "Is this thing on" and "Fixed the broken link", I've noticed you haven't responded in the comments yet.
[link]
From: Paul Hoffman (Oct 06 2006, at 15:20)
#6. I could care less that someone has commented, even if it is me. If a posting was really important to me, I'll go back and read the comments later.
Feel free to upate a posting with "There was an interesting comment about ...". That's you adding content to your own blog.
[link]
From: Darren Barefoot (Oct 06 2006, at 16:59)
I see this phenomenon with several blog feeds to which I subscribe. I think you missed an option for solving the problem: ask the feed reader developers to solve this issue. In my case, Bloglines ought to be able to tell the difference between meta or trivial changes as opposed to significant content changes. Why is it up to the feed publisher?
[link]
From: David Megginson (Oct 06 2006, at 17:11)
One advantage of writing your own blogging software, as Tim has done, is that you have the chance to explore new approaches and techniques. I've never seen a blog that updates for every new comment, and it's been an interesting experience — basically, the article stays new in Bloglines as long as it's warm; once it goes cold and people stop commenting, it stops showing up as new.
This is information that's usually not published unless I explicitly subscribe to a comment feed for a single article; however, it also goes so strongly against people's expectations (as shown in the comments here) that I doubt Tim will decide to stick with it.
[link]
From: Manuzhai (Oct 06 2006, at 17:15)
Well, it seems it hardly have to add my point of view, since I agree with pretty much everyone that a new comment shouldn't update the article.
Luckily, my aggregator (FeedDemon) hasn't been showing updates for your articles whenever a new comment came in.
[link]
From: Chris (Oct 06 2006, at 18:17)
My vote is for #5 and #6. I don't really have much to add that hasn't been said already. I do like the idea of a big comment feed and a per-post comment feed for particularly lively comment discussions.
[link]
From: Postmodern Sass (Oct 06 2006, at 18:36)
No comment on the technical aspects of commenting, but rather a question: I'm curious to know why you decided to allow comments at all, having been blogging for -- how long, now? -- without them.
When I began blogging I consciously chose to turn off the comments feature, preferring instead to have email conversations with those of my readers who felt inspired to converse with me, and occasionally publishing as blog posts a collection of the more interesting conversations. Too often, comments left on blogs are of the, "Dude, you rock!" or "Dude, you suck!" nature; a waste of bytes.
I changed my mind last year and started to allow comments, but recently I've been considering reverting to my original stance.
[link]
From: Tim (Oct 06 2006, at 18:36)
I agree about not being notified about every comment update. My question is: do we need any kind of comments feed at all? Why not just a link back to the comments on the site? If I'm reading the feed and want to go peruse the comments I don't think it costs much to follow that link.
[link]
From: Aristotle Pagaltzis (Oct 06 2006, at 21:18)
I agree with everyone making the case for comments not being part of the entry and thus not entailing update semantics.
I also agree with everyone suggesting that you implement Atom Threading.
Whatever you decide to do, please do not get rid of the global comments feed. On some weblogs where contributions tend toward a high S/N ratio, I want to see them all, and such a feed is the only scalable solution for that. Per-entry feeds are too much work for me and will be too much work for the server too. Following the permalink and reading the comments on-site is not an option – if it were, then it would also be an option to read everyone’s weblogs by visiting them manually, which I don’t see anyone proposing.
(Note that even though the implications of offering some 1,500 feeds would worry me, I’m not arguing against per-entry comment feeds. Many people will prefer those. I’m just saying that I find a global feed at least as important.)
[link]
From: Avik Sengupta (Oct 06 2006, at 23:05)
I'll reiterate an earlier comment - Yours is the few feeds where I set Bloglines to show as new on an update, since you are very clear on what's updated in the content. With your current comments implementation, this is getting painful.
Please consider having separate comment feeds, and not updating the entry for new comments.
[link]
From: michael (Oct 07 2006, at 02:34)
If the comments showed up in my aggregator then, sure, flag the entry as updated. But they don't (I use bloglines), so the "entry" is not really updated.
My preference is for 6 - a global comment feed
[link]
From: Sam Ruby (Oct 07 2006, at 05:24)
> Why not just a link back to the comments on the site?
This is a topic that I'm interested in. However, as the base entry has not been updated (and I argue that it should not be marked as updated), the base entry is rapidly scrolling down my river of news. Because I am trying to actively follow this discussion, I looked for it, and clicked on it.
For many of your other posts, I will still be interested, just not enough to spend that effort. Furthermore, every once in a while something interesting will come in on an old post (trust me, it does on mine).
Another factor is that what I believe you want with comments is a conversation. Moderation delays and propagation delays detract from this.
Here are some stats: in the past week, my comments.atom feed was fetched from 123 different IP addresses. Of course, some people roam; others are behind a corporate firewall, etc. And some use agregators like bloglines which, in this case, reports that it is fetching the feed on behalf of 32 users.
Additionally, I render an alternate version of my comments feed, in xhtml. This was fetched by 209 unique IP addresses over the same period.
Clearly there is a demand. At least there is on my weblog.
[link]
From: Graham Parks (Oct 08 2006, at 07:31)
Don't change atom:updated. It's completely the wrong thing to do. Use slash:comments and feedreaders can do something clever if there's demand.
(also, your comment textarea has a space in it by default for some reason)
[link]
From: Bryan (Oct 08 2006, at 11:11)
Tim: By all means update the comment count, but not don't mark it as updated (so I'm with everyone else). I don't much care whether you build 5 or 6 ...
Cheers
Bryan
[link]
From: Sam Greenfield (Oct 08 2006, at 11:42)
Regardless of the technology decisions, it is annoying to have the same post display again and again in bloglines when a new comment is posted. It would be nice to have a per-post option to see changes or comments when they occur, but in general, I don't care if new comments are posted. In a way that is a consumer issue, but since you are the only publisher doing this I doubt that bloglines and others will change their behavior.
On the publisher side, I don't see why you would want to change the <em>content</em> of your post when a new <em>comment</em> is posted--what you are writing hasn't changed significantly.
[link]
From: Tim Bray (Oct 08 2006, at 23:03)
Thanks for the input, everyone. There's a followup which addresses a few of the individual contributions at http://www.tbray.org/ongoing/When/200x/2006/10/08/On-Comments
[link]