33 comments (33 new)

July 5, 2007 1:11 PM

Would it be possible to have more accurate counts of the number of new posts in a thread based on my most recent comments? I know currently it's generated based on a cookie, but that doesn't work if you read MeFi from multiple computers or if you don't look at the frontpage. It also doesn't take into account my comments like the Recent Activity page does. If there are 33 new comments, but comment #17 is mine, shouldn't the count reflect that? (33 comments (16 new)) instead of thinking all 33 are new?
posted by blue_beetle to Feature Requests at 1:11 PM (23 comments total)

Oh, and I heard a rumo(u)r that someone had something deleted somewhere. I demand it be restored immediately!
posted by blue_beetle at 1:13 PM on July 5, 2007

It's not really possible to have more accurate counts without a lot of system database strain. The system in place works pretty well for single computer use, or multiple computers, provided you're waiting for sessions to time out between computers.
posted by mathowie (staff) at 1:29 PM on July 5, 2007

They had to make a tradeoff between currency and server load. If they update too often the server comes to its knees -- and we've already had far too much trouble with that, thank you.
posted by Steven C. Den Beste at 1:52 PM on July 5, 2007

I've most ignored the "x new" messages up until now, because as far as I could tell they were never even close to accurate.

Now that I'm paying attention, I'm even more confused. If I click on a thread on the front page of Mefi, read all the comments, and then go back and reload the front page, it still tells me that all those comments are new. What do I have to do to make the site know the comments aren't new to me anymore?
posted by vytae at 2:16 PM on July 5, 2007

mostly
posted by vytae at 2:18 PM on July 5, 2007

I've never really figured out how the "x new" comments thing is supposed to work. It doesn't seem to anyway.
posted by puke & cry at 2:23 PM on July 5, 2007

vytae you have to not access the site for some predetermined amount of time for the counter to reset. It's not a very accurate measure, but it is a compromise. I know there have been lengthy MeTa threads in the past as to these "x new comments" things not meaning what people think they should, and the conclusion is always that doing it right is too hard.
posted by Rhomboid at 2:26 PM on July 5, 2007

Fair enough, it looks like the new comment count will continue to be a cruel mistress whose gaze titilates, yet burns me so.
posted by blue_beetle at 2:51 PM on July 5, 2007

So is it basically still working like it was in 2001?

Here's what's going on: you come to metafilter for the first time today, it says 3 new links and 24 new comments. You read the site, hitting the index page as many times as you want, and you'll never lose that 3 links and 24 comments status.

then you go away for a few hours, and you come back. it should say "1 new link and 13 new comments" (if that much changed between visits), and it should remain that way until you go away for over 15 minutes.
posted by vytae at 2:53 PM on July 5, 2007

wow, 2001. So, yeah. Pretty much.
posted by Rhomboid at 3:05 PM on July 5, 2007

I've never really figured out how the "x new" comments thing is supposed to work. It doesn't seem to anyway.

Same here.
posted by danb at 3:08 PM on July 5, 2007

Huh. OK. Thanks for the clarification. I'll continue to ignore those messages then. :)
posted by vytae at 3:08 PM on July 5, 2007

I just make a little chalk mark under the last comment if I know I am coming back to a thread.
posted by Iron Rat at 3:10 PM on July 5, 2007

Yeah, the counts work based on your first visit and don't get updated until you leave for 20 minutes and come back.

It's how a lot of sites do their "new since last visit" stuff, and it would take a ton of extra processing to record every page you have viewed to keep counts totally accurate like in vytae's example. I'd have to store a database entry for every page viewed by every member in order to keep the new comment counts up to the second accurate. The tradeoff of computing power required vs. usability gained doesn't really work so I go with this sorta-working method that isn't entirely accurate within a session viewing stuff but should be between sessions.
posted by mathowie (staff) at 3:16 PM on July 5, 2007

I used to find the (x new) thing frustrating, but I've found that I mostly am interested about up-to-the-minute comment counts in the threads I'm actually taking part in, so My Comments handles those pretty well. If it's a thread I'm not actively participating in, I'll often just check it out every few hours anyway, so the long timeout on the numbers works pretty well.

Don't know who well that translates to different reading/commenting habits, though.
posted by cortex (staff) at 3:26 PM on July 5, 2007

Matt, is there any way that you could expand the cookie info to keep the records of pages recently viewed on the user's computer, instead of in your database? (I'm not sure whether that would help that much, even if it is possible. I'm just tossing the idea out there.)

If you can't make the interface act the way people expect it to, it seems like the best solution would be to tweak the interface so people expect what it already does. It really is a pretty accurate count of "new since last visit" comments, at least, so maybe the real fix would be rephrasing the link. "33 comments, 16 new since last visit" still doesn't quite capture it (last visit to the thread? last visit to the site? well, kind of, but only if it's been a while).

There are plenty of smart people around here. Somebody should be able to come up with a pithy label for that link, right? Something short, but more clear?
posted by vytae at 3:36 PM on July 5, 2007

Vytae, you're missing the point. Your solution adds work for the server, which is overtaxed as it is.

It doesn't matter how it's done, or who writes the code. The problem is that it's more work for the server.
posted by Steven C. Den Beste at 3:42 PM on July 5, 2007

Yeah, I get that it's more work for the server. But if the info that says "the user who is loading the page last visited these threads at these times" was sitting on the user's machine instead of in the database, some of the load on the server would be reduced. Right?

I'm more in favor of rephrasing the link, personally, but it'll take somebody smarter than me to find that magic phrase.
posted by vytae at 3:45 PM on July 5, 2007

Works just dandy for me, as always.
posted by stavrosthewonderchicken at 4:06 PM on July 5, 2007

But if the info that says "the user who is loading the page last visited these threads at these times" was sitting on the user's machine instead of in the database, some of the load on the server would be reduced. Right?

Wrong. It would reduce storage slightly but increase compute costs.
posted by Steven C. Den Beste at 4:28 PM on July 5, 2007

It would reduce storage slightly but increase compute costs.

Increase compute costs compared to what? Compared to what's going on now? Obviously. But compared to the strategy Matt described ("a ton of extra processing to record every page you have viewed") it would reduce database hits for both storing and retrieving the "who visited what pages, when," since that stuff wouldn't be stored in the database at all.

But if that info isn't in the database, then you don't get to do any nice convenient joins in your queries. I suppose you'd end up doing a separate query for each thread to determine the number of new comments, based on the individual last-visited timestamp for that thread. Yuck.

OK, I'm done speculating. I was hoping there might be a way to do it with only a slightly greater hit to the servers compared to the status quo, but you guys have obviously been mulling this over for 6 years without any good solutions cropping up. I still think it'd be worthwhile to ask for community suggestions on a better way to phrase the link, and do it in a more visible location than the end of this rather esoteric MeTa thread.
posted by vytae at 5:29 PM on July 5, 2007

Works fine for me as it is (yeah, like anyone cares).

As you have already stated, there are lots of smart people here and they have been thinking about this for a long time without a workable solution, so you should consider that this is as good as it gets. Maybe it's your expectations that need adjustment?
posted by dg at 2:04 AM on July 6, 2007

<SNARK>If Google can do it, why can't we?</SNARK>
posted by blue_beetle at 7:07 AM on July 6, 2007