I believe the issue here was that the library was a jQuery library? @humitos is this correct?

Yes, and it's because it's not maintained anymore. The last commit was 4 years ago: https://github.com/calebjacob/tooltipster

Load jQuery and library conditionally, via dynamic webpack import. We still use jQuery in this case, but only when the documentation project uses hoverxref

I would avoid this approach if possible because the underlying library is not well maintained anyways. We will either, replace the library now or in the future anyways. So, I wouldn't re-implement it using an old library at this point.

We replace the library with another implementation

This is the path I was following. I found popper.js which looked promising, and then found tippy.js which is another layer on top of popper.js that makes it easy to work (my POC lives at https://github.com/readthedocs/addons/blob/main/src/tooltips.js)

Today I found that popper.js was renamed to floating-ui (https://floating-ui.com/). So, it's worth to take another look at it. It's also worth to note that I didn't find any issue with these libraries and I had something that worked, but it wasn't just styled properly. Example:

We rely on upcoming HTML features that aren't yet fully standard

Hrm, it doesn't seem this is gonna happen soon, tho.

Hrm, it doesn't seem this is gonna happen soon, tho.

Most browsers already have this feature actually ¹ -- Firefox is the outlier here, it's only in nightly right now.

I feel like our time frame for this feature is still a little ways out, perhaps after we get a wider distribution on the current work in addons? So, time frames might even align too.

That is, we might not have to drop the work we have here already, and instead could use our current dependency until we swap to native only. Or do some polyfill if the native feature is missing and back down to the current library.

My main concern is reworking our dependency here twice. I think we should ultimately land on using native features if we can. Investing in a secondary intermediate dependency doesn't feel super useful for that.

If floating-ui has plans to act as a popover polyfill, that would be exactly what we need. I'm not familiar with the library though.

I've used popper.js/floating-ui before, and I have good experience with it. As far as I know it's a pretty well maintained library.

I haven't thought of the native implementation before, but I like the idea, so I'll take a look into that first (and probably check floating-ui if they've had any discussions/thoughts around that). My initial reaction is skeptical towards the native implementation, but since I haven't looked into it before, so I'm ready to be surprised :D

I'm guessing that none of these libraries are operating as a polyfill for popover, but maybe I'm wrong.

Since my last comment, and since 2024-04, popover is now in Firefox and Firefox Mobile, and has been in all other browsers since early last year.

So I'm inclined to lean into this feature, rather than settle into a library that will naturally self-deprecate due to the overlap. But if there are strong reasons to use a third party library here, that might influence the decision.

Ok so unless I'm misunderstanding this addon, a very strong argument against using native popover is positioning. The big advantage of using popper.js/floating-ui is that it deals with positioning really well. Native popover doesn't seem to do any of that, which makes me think that implementation using native popover will be much more time-consuming.
Looking closer at native popover, I think comparing popper.js and popover is not really fair. We should instead be comparing it to CSS anchor positioning which unfortunately is not at all ready to be used (not even close).

So my vote is definitely for popper.js/floating-ui for now. What do you think?

So my vote is definitely for popper.js/floating-ui for now. What do you think?

I'm fine going in that direction 👍

Floating ui is a bit more low-level than tippy js, which is currently used in tooltips.js. So the implementation will be slightly different from what we have now. The idea is:

1. User hovers over an a.internal link
2. The provided url is fetched with "/_/api/v3/embed/?" prefix
3. A div is added to the DOM as a sibling to the hovered a.internal
4. This div is positioned using floating ui
5. The div is populated with the fetched content using innerHTML

Is it ok to leave that div in the DOM, and just hide it. And then show it again once the link is hovered again?

User hovers over an a.internal link

I'm not sure about the a.internal CSS selector. I suppose that .internal class may not be used for all the doctools we support. Finding a selector more generic would be awesome here. Maybe [href~={{ window.location.hostname }}] or similar. Would that make sense?

In any case, the initial implementation could use a.internal for now since it's the Sphinx one 👍🏼

Is it ok to leave that div in the DOM, and just hide it. And then show it again once the link is hovered again?

I suppose there is no problem. Maybe hovering a lot of links could make the page to become slow, tho. We don't need to premature optimize it, so either is fine to me.

I'm not sure about the a.internal CSS selector. I suppose that .internal class may not be used for all the doctools we support. Finding a selector more generic would be awesome here. Maybe [href~={{ window.location.hostname }}] or similar. Would that make sense?

In any case, the initial implementation could use a.internal for now since it's the Sphinx one 👍🏼

I just proposed what's already in tooltips.js and hoverxref.js. If there's a better selector to use, that's even better. For now I can go with the simple a.internal and then we can improve later?

For now I can go with the simple a.internal and then we can improve later?

Yeah, that's fine.

@humitos how should styling be handled? Currently (the way sphinx-hoverxref is implemented), things are displayed using the styles that are already on the page (which might be completely different from the content in the link). I assume that's ok, because the intention is that the links are all under the same site, right?

things are displayed using the styles that are already on the page (which might be completely different from the content in the link). I assume that's ok, because the intention is that the links are all under the same site, right?

Yes, this is correct. Since we are using links from the same project, the style should match. However, in the current (Sphinx) implementation we are linking to external resources as well (e.g. using intersphinx) and if the external content is not using the exact same Sphinx theme, the look & feel is weird.

Unfortunately, we don't have a good answer for the external link use cases; so we can skip it for now. If you have ideas about how to implement this properly, feel free to share them here or open a new issue.