Hacker News Reader

Comment by atemerev | Hacker News Reader

Technically, I agree that limit/offset paradigm is flawed, however, the author just dismisses a valid criticism of keyset pagination with the following:

"keyset pagination has some limitations: most notably that you cannot directly navigate to arbitrary pages. However, this is not a problem when using infinite scrolling. Showing page number to click on is a poor navigation interface anyway—IMHO."

Sorry, but NO. Infinite scrolling sucks, because it fills the page (and therefore memory) with lots and lots of entries, and disallows me to quickly navigate to exact place in the view I need. If I know that what I seek was approximately on page 12 of last search result, I will go straight there (1 click) and navigate from here. I hate downloading the entire history and scrolling from the beginning.

Infinite scrolling is fine the first time, but on repeated visits is just not an option and destroys everybody's valuable time.

View on HN

I agree that infinite scrolling is terrible, but pagination based on nothing more than the rank isn't great either. Why do I need to guess that names starting with a 'G' could be on page 12 (no 14, no 13)? Just paginate by the first one or two letters then. When doing so, you can benefit from database indices again while improving usability.

The old Right Stuf website catalog [1] (old catalog not accessible) would list their pages with the first two letters of the first item's title on the page. I found it immensely helpful, and even "borrowed" the idea for a directory on a local govt website (also no longer accessible).

[1] http://web.archive.org/web/20071016095932/http://www.rightst...

Static Ice does this nicely. When you hover over the pagination links it tells you what price range is covered by the products on that page:

http://staticice.com.au/cgi-bin/search.cgi?q=laptop&spos=3

> it fills the page (and therefore memory) with lots and lots of entries

Yet observables are all the rage. Additionally, can't you just move prior entries out, so the "scroll and fetch" goes both ways?

> disallows me to quickly navigate to exact place in the view I need

I've seen infinite scrolling that implements paging as well

In general I'll agree infinite scrolling isn't implemented well, and is usually just the result of dropping in a module and forgetting about UX.

> I've seen infinite scrolling that implements paging as well

Then you're using offsets, which is the crux of the disagreement regarding the utility of offsets.

I'm sure he means infscroll using keyset pagination.

The UX, as noted, is ugly - have to somehow support scrolling in both directions with some sort of lazy-loading - but pagination in this context doesn't === offset-based pagination.

> If I know that what I seek was approximately on page 12 of last search result, I will go straight there (1 click) and navigate from here.

How do you know it's it's still on 12 and not 13 or 11 if the results changed? Wouldn't a better filter or other navigation be more useful?

> Infinite scrolling is fine the first time, but on repeated visits is just not an option and destroys everybody's valuable time.

To me this doesn't make sense. I can't think of a time where I've repeatedly gone to a specific page on results and where I wasn't able to filter for what I was looking for instead.

Twitter?

I'm not sure I understand this use case. I'm on twitter, I see a tweet I'm interested in. I leave, come back, and then try to find that same tweet by browsing for it.

Wouldn't it be simpler to save the tweet the first time around or if not, since I know the tweet I'm looking for, search for it where I'd get the exact result?

Isn't the issue there a lack of (usable) ability to have the computer remember what you were looking at in some reasonable manner?

I'm perfectly fine with my own memory for that. However, yes, bookmarking the place in the timeline would be a nice feature to have.

(Now if only Twitter, Facebook or Hacker News would have implemented this...)

Reply on news.ycombinator.com