Talk:Flow/Comparison of threading models - Classic

From mediawiki.org

This page contains examples of real conversations, laid out following the classic threading model of talk pages in Wikipedia. All comments are copyrighted to their authors and licensed under the CC BY-SA 3.0 License and GFDL.

It's more difficult than it used to be to figure out who is responding to whom[edit]

Original conversation

I have this philosophy about taking time away from new extensions after I've played with them a bit and given feedback, because otherwise I get frustrated when things don't move at "my" pace. And usually, when I come back weeks or even a couple of months later, I can see improvements and am happy to try them again. (Heck, it worked for VE and MV - They both got noticeably better and more usable over time, and the improvements were more obvious when I stepped away then returned.)

This extension is almost the opposite. I've yet to return to it to find it improved as a communication tool. I read the sections below, and I honestly cannot tell who is talking to whom, who is responding to what, without really having to work at it. It is significantly more difficult for me to figure out this page than it ever has been in even the most convoluted discussion on a normal wiki talk page. It's one of the few times where I've looked at something and been completely unable to give any constructive criticism. I've been able to figure out dozens of discussion systems pretty quickly over the years (going all the way back to the bulletin board days, probably before some of you were even born)...this is the most difficult one I've run into for at least 10 years. I know lots of people are really working on this, but from what I'm seeing here, your objectives are so divergent from what I as a user expect from a discussion system that never the twain shall meet.

Incidentally, what brought me here today was an email message informing me that User:Flow_talk_page_manager had moved a page called Talk:Flow Portal/Archive2 (a page that as far as I knew wasn't on my watchlist), and when I followed the link....there was nothing there. I can't tell what got moved where, and there is nothing on that page to tell me what the heck this "user" is. (A bot of some kind? who is running that obvious role account?). Risker (talk)

Which thread(s) are you having problems following? Diego (talk)
E.g. the thread named "New side rail". It looks like DannyH (WMF) replys himself. (looks like copy paste has some side effects - at least in the editor - I am not going to figure which icon I should press).
You can not expect people to read the threads from a to z. I read the first message in a thread, if interesting I read the first reply, and if its not interesting I skip to the next reply to the original message and so on...
(This replys is a good example - I replied to Diego Moya, but it looks like its to Risker). Christian75 (talk)
Why do you think your comment looks like a reply to Risker? It's right below my comment, so it feels natural to associate it to the text that is closest to it (immediately above), and to read the thread as a sequence of posts at the same indentation level.
This is the standard layout of new posts at internet boards and comment sections in blogs (i.e. most of the Internet). Reading text in order is fairly natural, we've been accustomed to doing that in books and magazines for ages. Diego (talk)
User:Diego Moya, I cannot tell strictly by observation if Christian75 is replying to you, replying to me or simply making an additional comment to further the discussion; only his parenthetical comment tells me that he was replying to you. This formatting is aberrant; it is unlike any other discussion system I have seen in a very long time. The wiki system, for all its ease of screwing up indentation, gives visual hints as to who is responding to whom, or if they are just making an added comment to the topic area. Forum software allows (and generally encourages) quoting. Comment systems for most media indent the first response to a comment and (depending on the system) keep all the rest of the related responses (including responses to responses) at the same indent level, indicating that a "side conversation" is taking place, or allow indentation of responses to responses to a specified level (sometimes 2 or 3 indents). This system gives absolutely no visual clue as to who is responding to whom.
As an aside, I have this thread on my watchlist, yet I have received no notifications that there have been further responses. What's with that? Risker (talk)
The new threading model does provide visual cues for side conversations, it just excludes the first reply from them and considers it part of the main conversation. If Christian75 had replied to you instead of me, his comment would have been indented one level, right below yours; because his comment is not indented in such way, I can tell that it was posted as a reply to the main thread just after my first comment, and thus he was not replying directly to you.
The visual hints you talk about are not part of the wiki software, it's a convention adopted by its users using formating tools, not any specific feature of the software to support conversations; it's a by-product of its flexibility, not something designed for collaboration.
Flow merely uses a different convention; the current threading model decides to apply subthreading to all replies to a post except the first one as a way to separate the main sequence of ideas from isolated remarks about a particular post (see the new reply by He7d3r to your fist post).
At this previous topic there was a proposal (initiated by Hhhippo, tweaked by me) to adapt the model to something more similar to mediawiki talk pages, see if you like it better (it works as you describe, creating a single indent level for each new group of parallel posts including the first).
Aside - have you clicked the star at the topic title to add it to your watchlist? If you did, you should report the lack of notifications as a bug (I've received the notification for your replies). Diego (talk)
User:Diego Moya, I do understand the convention being used. I disagree that it is a sensible one, because right now there are two replies to my initial post, but only one of them is indented; thus, it looks like one person answered me and another person (your first post) was simply a comment to add to the discussion. I cannot tell visually or by any other indicator than the postscript of his post that Christian 75 was replying to you: his comment could be a standalone, a response to me or a response to you. He7d3r has commented directly above where I am writing now. I cannot tell if he is responding to your previous post, or to my earlier post but not clicking "reply" or if he is adding an entirely new comment. He also responded to my first post and his response is indented, but it comes above your response to me and is thus no longer in a logical time sequence. There is no sense of conversation, of discussion as opposed to a bunch of people adding comments on a topic. There is no logical progression. Indenting responses makes sense to differentiate them from comments that are not responses.
I do understand that the use of indentation and bullets on wiki pages is a convention as opposed to a forced software option; that's appropriate, since the pages are used for so many different things. However, we've also known for years that adding an "INDENT" button that automatically added the correct indentation for a section was entirely possible, and that it could even be designed to automatically outdent (and add the "outdent" markings) after a specific number of indentations. But no volunteer decided to write the software, and the person in charge of engineering and product for many years was always extremely clear about how he really hated talk pages, so there was no reasonable chance that anyone from the WMF side would be tasked to improve the talk page software. I certainly won't complain that nobody was interested in committing career suicide over talk pages. What applies to content - i.e., that we're seriously falling behind in maintaining and improving existing content because there's so much more reward in making new content - applies to software too. Everyone know's it's much more interesting and fun to create something new than to rebuild something that already exists, and it's usually also easier.
As to the watchlist issue, this thread (and actually the whole page) is on my watchlist, I have been getting email notices for a bunch of page moves and other activities that have been happening recently to other pages on my watchlist, I am just not getting anything for this page or this thread. I'll take your advice and submit a bug report later tonight. Risker (talk)
But most blogs (like Facebook) have a philosophy that the users should not discuss, but just leave a comment. And I think most people do not read the internet like a book. Stay on one page, read it all, before going to the next page. Christian75 (talk)
Yes, I agree with that, and that Wikipedia talk pages should be forum-like rather than blog-like. But this model is not that of Facebook, it's closer to forums and bulletin boards.
BTW, boards a la BBCode have a philosophy that users should discuss, and they've managed to support it quite well through the years - mostly by the use of quoting previous comments. The current threading model is an attempt to improve on that by reducing the amount of quoting required. Diego (talk)
See also the WONTFIXed phab:T93883. He7d3r (talk)
Oh wow, He76d3r. That is one of the saddest phabs I've ever seen. Seems to me that what they've failed to recognize is that they're not getting as much criticism because fewer and fewer people are actually trying things out. I see that Short Brigade Harvester Boris was making the same point as me here, about the visual cues being very important in reviewing Wikipedia talk pages; as he says, they're usually scanned as opposed to being read.
EDIT: Oi! And why, when I have VE not enabled on this wiki (deliberately - I use it elsewhere), is this page not properly handling wikitext? Geez....
EDITED AGAIN: It's only because I was certain there *had* to be a way to switch to wikitext that I tried the </> symbol, and that certainty only came with years of having been around. All the mucking about I've done with VE, I still haven't found the process for making piped links. Risker (talk)

Even Poor man's Liquid Threads indents comments correctly: wikt:User:Yair_rand/addcomment.js. He7d3r (talk)

That would depend on your definition of "correctly", wouldn't it? ;-) Diego (talk)

I'm confused about why this conversation is unreadable. This is a group conversation taking place between four people. If this is post #8, then I'm not directly responding to message #7. I'm responding to message #1-7, as people do when they're having a group conversation.

You can tell which sentences in the conversation that I'm responding to by the content of my response. At the moment, I'm disagreeing with Risker, and agreeing with Diego Moya. I don't think this would be any clearer if I had to specifically choose which of Risker's three and Diego M's three comments I'm replying to. DannyH (WMF) (talk)

By the way, Risker: you can tell that Christian75 was replying to Diego Moya's comment, because Diego asked a question, and Christian75 answered that question. That seems pretty basic to me. Similarly, you can tell that I'm replying to a sentence in your post right now, because I said your name, and I'm answering your question. DannyH (WMF) (talk)

Why would I have to name the person I'm responding to? Why isn't it visually obvious? Why do the replies to comments not appear in the order they were written, so that the first reply appears before the second reply? Is not the purpose of this entire exercise to make communication clearer rather than more obtuse? Risker (talk)
I'm sorry that you don't like that aspect of the feature. I think in general it seems to be working as it's intended, because people are having successful conversations using this system. In fact, we're having one right now, and I think it's working fine. But I get that you disagree, and that you don't like it.
I think the questions you're asking are rhetorical, so I don't need to actually answer them all, but let me know if I understood that wrong. DannyH (WMF) (talk)
I'm sorry, DannyH, but they're not rhetorical questions, they're serious ones, and I do think it is reasonable to expect answers to them. The current discussion system uses social convention for users to respond in specific ways (indenting or bulleting); while they don't always do it right, the software is not able to force the behaviour. Here we have the opportunity to actually design the software to force the behaviour, and I'm at a loss to understand why it is not only not being done, but that for some reason the behaviour is essentially being treated as dysfunctional. I'm having a hard time with five people in this discussion, and I cannot imagine the challenge in sorting out a 15-person discussion with a heavy mix of standalone comments, responses to comments, responses to responses, and a subtopic or two that should remain a subtopic and not its own topic (common in many discussion areas), or differentiating between "votes" and comments and replies when analysing, for example, a very active Article for deletion discussion or RFC. Risker (talk)
Have it occurred to you that your difficulty may arise from you being strongly accustomed to mediawiki's "pure threaded" model, which you're trying to apply here where it is not relevant? This certainly may be a problem for adoption of the new convention by experienced editors, but it's not inherent to the model.
Why do you think there is a relevant distinction between posting a general reply to the whole thread, and posting a direct reply to the last post added to the thread? In both cases the post is placed right below the previous one at the same indentation level and in the same order. Again these are not rhetorical questions, your answers may be useful for a possible redesign of the threading model.
(For comments that are replies to posts other than the last one, you get all those "responses to responses" and "subtopics of subtopics", so that aspect in particular is not different from the old model, only the existence of a "main" thread were new "first" comments are located by default, in order. The new software is enforcing a convention, just a different one than the convention used at WP:Talk).
As for a complex conversation with multiple editors and several levels of responses to responses, "New indentation and threading model" at this same board is one of the more complex discussions held with the new model, and it manages to remain at two levels flat despite its interleaved interactions at various times. Do you have problems following its sequence? Diego (talk)
Well, Diego Moya, I'll put my hand up in admitting I've probably read more threaded onwiki discussion than most people; easily over 100,000 talk pages and talk page archives, sometimes going diff by diff to catch subtle changes, while I was on Arbcom. And there were plenty of times in all those hours that I found the incorrect use of indenting/bulleting to be frustrating too. But even bad indentations were what made the pages readable, especially when going through very active talk pages or lengthy and convoluted discussions. I cannot fathom going through one of those 15-20 person ANI threads without some sort of visual variation.
I get that this system is meeting the expectations of the people who designed and implemented it - good on them for successfully meeting their own expectations, I know it can be a challenge sometimes. But this page looks more like a wall of text than ANI does, and that's saying something. This system is probably fine for WikiProject Hampshire, but not a busy page.
And why is it that in order to preview, I come out of Wikitext again and wind up back at VE? Risker (talk)
I have to admit I agree, even this specific thread was an enormous pain in the ass to read and figure out who was responding to who. The plain flat threading has been one of the primary reasons I avoid editing flow boards if I can possibly avoid it (to the point of just not saying things sometimes for example on officewiki where I resort to email if I can or just let the thread happen it's why I'm opposed to moving things on wiki from the staff mailing list). It was already pretty bad by default earlier to be honest but this is significantly more frustrating :-/ It's frustrating and confusing on youtube and it's frustrating and confusing here. Danny, you may consider this a successful conversation but I'm 100% honest when I said that reading and following this conversation was incredibly difficult. It is unlikely I will ever use flow if I can avoid it in this situation, it just isn't worth the effort :-/. Jamesofur (talk)
Jamesofur, Flow is not a flat forum (I also happen to hate those BTW), it's just disguised as one. The software still keeps track of what post is a reply to which one, and when a single post has several replies, it gives you all the visual clues needed to know who's replying to whom; it merely avoids moving comments to the right when that's not needed. It has nothing to do with Youtube's. In fact the rule to know the parent comment of any given post is simpler in Flow than in talk pages (look for the post immediately above left-aligned with the same depth level), and it has no exceptions.
I'm preparing for your benefit a comparison of this conversation using the various layout styles, I'll edit the comment to add it here when it's finished. Diego (talk)
"I have to admit I agree, even this specific thread was an enormous pain in the ass to read and figure out who was responding to who." +1 Сунприат (talk)
May I ask you and Jamesofur too what do you think of HHHippo's layout proposal, and my cleanup of it? Diego (talk)
I think (phab:T94924), we are talking about one and the same. We used to see branches in discus. And Flow breaks logic branch or branches in it not at all. We're trying to add a logic branch in Flow but DannyH seems think somehow different. Сунприат (talk)
The model by HHHippo does have branches, and it does not move the first comment to the last position, so it doesn't have the problem in phab:T94924; that model has not been tried out yet. BTW Danny said the current model is experimental, not definitive; it's still open to changes (such as adding branches which are easier to see than the one we're using right now). Diego (talk)
I have an example of a real discus, translated into Flow. Not "lorem ipsum...". [1] vs [2] ([3]) And that's bad. Try to make the image "Flow branch" for any real wikitext discussion from the archives of your wiki. Then we see whether it is better. Сунприат (talk)
To be fair, lengthy and convoluted conversations do produce lots of visual variation with the current Flow threading as well, it's just that we haven't found any such conversation yet. One thing I think we're learning at this talk page is that most discussions happen to have extremely simple structures, with almost all comments getting a single direct reply; it's just that the old talk page convention makes the layout in those cases quite complicated.
What do you think of the layout model suggested by HHHippo? That one is mathematically equivalent to the mediawiki talk page convention, it's always possible to convert between them (except for the usage of the "outdent" to rebase deep threads).
By the way - The problem with porting the classic convention to software is that it can't really be fully automated; there are rules which are left to the gut feelings of the editors, in particular for when to outdent threads that run too deep. The only other platforms I know that try a large scale threaded model are Slashdot - which doesn't really show all the comments at once, it hides the deepest levels by default - and LiquidThreads - which produces an unreadable mess by trying to strictly follow talk page conventions without their safety valves/manual overrides. Diego (talk)
Why do the replies to comments not appear in the order they were written, so that the first reply appears before the second reply? Is not the purpose of this entire exercise to make communication clearer rather than more obtuse?
That "posting everything in order" doesn't happen either at mediawiki talk pages, where later comments made in a deeper indentation level are placed above shallower posts made earlier. Every time you "outdent" one level you may or may not return to earlier comments; in Flow, it is guaranteed that the outer post was made before all the comments in the subthread, and thus can be seen as a return to the main conversation that was interrupted by the aside.
The problem with defining 'clear' is that it varies with each user's expectations. We have reports of newcomers not accustomed to WP's conventions for talk pages that find them utterly confusing, especially regarding how to indent their comments in order to reply to a specific post.
P.S. By the way, comments shown at the same indentation level are always shown in chronological order. If you ignore at any location everything with a higher depth level, you get a clean main sequence of ordered posts, equivalent to a classic flat forum. Diego (talk)

DannyH (WMF): Christian75 has pointed out what may be the single major benefit of threaded conversations, which is missing in the current threading model. When all replies to the same post are grouped together as a subthread, it's easy to skip the whole subthread if it's not interesting. You can't do that with the current "main flow of comments plus interleaved digressions": here you can't say from layout alone at what point the main track deviates from the original idea and delves into a new line of reasoning.

I think it's time to start exploring the dynamic model proposed by Hhhpo. I happen to like its final layout better, as all direct replies to the same post are shown in the order they were made, instead of the first reply being demoted after all the later ones.

It has a huge drawback though, as it heavily depends on people getting right the distinction between "reply to the last post" and "add a new (independent) comment to the thread".

Solving this would require exploring again the reply/comment distinction of the prototype, maybe with a more explicit message ("create a subthread" or similar), and adding stronger visual cues of what comments are part of the subthread (similar to Stackoverflow).


I agree with Risker and Jamesofur that we will need a way to split very long conversations into smaller components, maybe reintroducing the concept of subthreads. Surely a new formatting convention may develop to achieve that (using headers inside of comments, reordering posts...) but it would be better to have software support for that in the platform itself. Diego (talk)