Hacker News new | past | comments | ask | show | jobs | submit login
Smarter Link Underlines For Every Website (eager.io)
113 points by adamschwartz on Nov 10, 2014 | hide | past | web | favorite | 57 comments

It's a very nice idea — I always like smartening up the look of the Web — but the resulting links are ghastly. If your word starts or ends with a descending letter (e.g. "ghastly"), then the underline looks like it omits the first and last letters. It breaks the semantics of what the underline means — any bit that's underlined is supposed to be the link, and if it's not underlined it's not a link.

They also made the mistake of showing a link that includes the word "Typography", which has an underline so broken up and interrupted that it's distracting (which devindotcom also pointed out). It's linked, then it's not, then it's linked, then it's not!

So yeah, I have to give them credit for a good idea and a clever implementation, but it just doesn't work out in the end.

I appreciate your feedback. No doubt on some level this comes down to aesthetic preference. I’m curious though, on the before-and-after examples on the homepage [1] do you feel that all of the afters are worse?

You and devindotcom did indeed bring up some good points, particularly about descender-heavy links, something I plan to take a crack at [2]. But I’m curious if you feel the entire strategy is flawed or if there’s just more work to be done. On a related note: if you’ve used/seen it, how do you feel about links in iOS 8 Safari? How do you feel about this before-and-after [3] (screenshot of the first paragraph of the blog post)?

[1]: https://eager.io/showcase/SmartUnderline/

[2]: https://github.com/EagerIO/SmartUnderline/issues/1

[3]: http://postimg.org/image/rgfx5icq9/

Well, I realised what was bugging me and why I prefer the iOS implementation for now -- the tighter underline, both a pixel closer to the baseline and a hair closer to each descender, helps ensure the underline feels continuous even when it's broken this way. I do prefer this approach for headlines or animated effects, but perhaps (as was suggested elsewhere) changing the underline colour and/or tightening the spacing would help. A neat implementation though. Seeing links here, it's true, the "g"s just seem to get demolished by the underline. I wonder, if I linked to qithub.com, would you see the difference? ;-)

Oh and if there's a way to apply this without requiring a background color, I suspect this might be a popular add-on or browser setting for some folks.

See my comment above [1]. I think you’ll find that the iOS 8 and SmartUnderline implementations are closer than you think.

[1]: https://news.ycombinator.com/item?id=8588492

> They also made the mistake of showing a link that includes the word "Typography"

Including some aesthetics-stretching cases seems like a fairly honest/transparent thing to do for an open-source project that someone is blogging about.

> It breaks the semantics of what the underline means — any bit that's underlined is supposed to be the link, and if it's not underlined it's not a link.

That's a reasonable way to look at it, but humans like us can be pretty flexible with semantics, thanks to pragmatics. If I see a link that covers most of a set of letters stuck together without a space, I tend to assume that the whole thing is a click target. Even if my assumption is wrong, my clicking error rate probably won't be too high, since I think that people tend not to hit the edges of pointing task targets.

> If your word starts or ends with a descending letter (e.g. "ghastly"), then the underline looks like it omits the first and last letters.

That's how I was taught to type underlines back in the 1980s on actual physical typewriters, so it's not a new idea.

I still do it today!

If one could apply a double-underline then the second, lower, line would be uninterrupted and make for a pleasing effect.

To me, it actually adds visual complexity while eliminating the useful visual cue of a single line being a single link. Here, there are many lines, which to me indicate many links. In some cases, the line may present as dotted or dashed (the word "piggy" randomly came to mind) — things that signify something other than a link to me.

I would rather descenders weren't subsumed by the line either, but interrupting the element itself instead seems like as much as a problem as it is a solution.

I think this is a very reasonable criticism, particularly about words like “piggy”—no doubt typography is *ing hard.

Like is often the case with design, there are trade-offs. In this case I think the readability improvements to links with descenders outweighs the visual cue / scannability of links, particularly because I think links with more-descenders-than-non are not as common.

An interesting experiment might be to try to turn off the effect on words which fail some check (a high percentage of descender glyphs, for example). I created an issue on GitHub [1]—definitely something worth exploring. Thanks!

[1]: https://github.com/EagerIO/SmartUnderline/issues/1

I concur. I've been trained that seeing a break in the underline means I'm looking at a phrase or fragment broken into multiple links. Reading the whole page did nothing to break in my brain to the concept. I do not agree with Mr. Schwartz that this style will "[i]mprove the typography on your website."

As an aside, I love Butterick's link style, what with the triangle and the background-changing color hovers, but that's too complicated for a general audience--especially one reading on the newest touch-screen whatsit that can't show you the "◊Amazon" word is actually a link to Amazon until you accidentally tap it.

But that's a great style for reading long-form articles because your brain isn't constantly interrupting you to say, "Ah, please compulsively hover over that underlined text so I can see what website this insightful blogpost is referencing."

You can also see some dotting/dashing in the link on the page to the Butterick book.

I have a hunch using a faded color for the (possibly-then-thicker) underline would work better - the foreground descenders would retain distinction, without the noise of dotty/dashy line-ends.

Thanks for the suggestion. We’ll be taking a crack at descender-heavy links [1], so this may come in handy!

[1]: https://github.com/EagerIO/SmartUnderline/issues/1

In order to more accurately compare iOS 8 Safari’s text-decoration underline implementation to SmartUnderline, I created a test page. [1] It contains five links to the words gesture, quitting, discovery, piggy, and eager, set in Helvetica.

To produce the iOS 8 version, I opened the page in Safari on iOS 8.1 pinched zoomed to the maximum amount while keeping all five links within the frame, and took a screenshot.

To produce the SmartUnderline version, I visited the SmartUnderline install page [2] in Google Chrome on Mac OS X, entered the test page URL, and clicked “Preview in a separate window”. With the screenshot from my iPhone opened next to the browser, I used Command+[+/-] to zoom the browser until the font sizes were the same size, then took a screenshot.

Here are the results: [3].

To summarize, I believe iOS 8 has some advantages and some disadvantages, but to me there is no clear winner.

Some specific points in no particular order:

- iOS underline breaks are always vertical cuts. Since SmartUnderline uses text-shadow to “white-out” the underline, the cut-outs are in the shape of the glyph. To me, this looks better in the “g” in “eager”, but worse in the “q” in “quitting”.

- iOS 8 shows a line between a “g” and ”y” (as in ”piggy”) and SmartUnderline does not.

- SmartUnderline’s underline extends less far to the left of the first letter and right of the last letter. This looks better in “quitting” and potentially worse in ”discovery”.

- In general, SmartUnderline sets a little more space around each descender.

- Both implementations show no underline before or after a leading or trailing “g”.

- Both implementations show the underline very close to the left side of the “y” descender. I think improvements could be made here with both implementations.

[1]: http://jsfiddle.net/adamschwartz/hyyerbq4/show/light/

[2]: https://eager.io/app/eA9ULux0UOJP/install

[3]: http://postimg.org/image/nf9b2ahd1/ (@2x http://postimg.org/image/688ricy2v/)

My biggest problem with this is that you're not using ligatures defined by the font to perform the hinting required.

If you did use ligatures - derived from the typeface itself - then the hinting of the underline cuts can be better fitted to the face. But that assumes that the type designer themselves considered the case. So .. to me .. you're patching something upstream which might be better considered at the source: the font itself.

Fyi: Postimg.com gives me a popup with "watch free hd movies now" and redirects me to some gambling page. When going back, I can see your image between some half naked guys. You may want to use something a bit more serious in the future.

Imgur mirror of the 2x: http://i.imgur.com/KD7k0Rz.png

I recommend installing AdBlock in your browser.

This kind of underline painting behaviour is proposed in http://dev.w3.org/csswg/css-text-decor-3/#text-decoration-sk... as text-decoration-skip:ink. Not yet implemented anywhere though.

Neat, though I don't actually think I like the look of it. I think I'd rather have an underline further down, so that it's below the descenders.

No doubt, at a certain point it does come down to aesthetic preference.

Using a border-bottom to achieve the look you’re describing is totally a valid preference. One potential advantage to the border-bottom technique is the way it can be used to make a link stand out. In fact, I use this link styling on my personal website for that reason [1].

(Note: when using a border-bottom, I’d recommend using a slightly larger line-height so that the line still seems associated with the text above it and not the text below it. This is one of the pitfalls described in the original Medium post [2].)

Unfortunately, aside from iOS 8, normal text-decoration underline links have the descender crossing problem.

I absolutely took this into consideration when designing SmartUnderline. SmartUnderline literally only applies itself to links with text-decoration underline, nothing else. It even tries to ensure other conditions which make the transformation favorable (display inline, solid background color, etc.).

Ultimately I think the before-and-after on the SmartUnderline website tells the best story [3]. No doubt one may appreciate other, very different link styles. (For example, just using a different text color to denote _link_ is a perfectly valid and commonly used technique.) But if you’re using text-decoration underline already, this just makes those links look better.

I think blogs, news articles and wikis are some of the primary examples of where these links look best. Anywhere where text is heavy and the links are _not_ the primary focus.

[1]: http://adamschwartz.co

[2]: https://medium.com/designing-medium/crafting-link-underlines...

[3]: https://eager.io/showcase/SmartUnderline/

I think this is very clever and I appreciate the work you've done to make it so robust. I'm not a designer, but I like to think I can appreciate at least some of really good work. Your pages qualify there.

One note on your before and after, it's cool that you included arabic, but the right to left nature makes it look like the examples are reversed. Though I realize perhaps that was intentional.


Ha! Yeah, my co-founder and I debated that one... (he thought I should reverse it). I can see it both ways, but your comment certainly puts another vote in the reverse-it column. Thanks again :)

I was about to send a PR for the reverse; then I realised what language it is. Perhaps having the columns not centered but reading-direction-aligned would make it clearer. Not sure there. But I'd keep it that way, it's rare enough that people treat RTL languages properly when they don't know them ;)

I naturally upvoted, but just wanted to add thanks for the links.

> First we define a helper mixin textShadowToCropUnderline which draws 12 text-shadows, half to the left, half to the right, spaced every .03em.

This seems like a big waste of CPU and memory. I don't want to drain my phone's battery rendering these underlines.

In our tests (scrolling an older iPhone on a page with 300 links), we found that it’s actually not as much an issue as you might think, particularly because none of the shadows use a spread value. I’d definitely be interested to see if you have evidence to the contrary.

That being said, it’s definitely something worth considering, particularly since iOS 8 now implements descender-aware link underlines now. If there were a way to duck type for this feature, we’d definitely disable the JS anywhere we detect it working. Drawing an underlined “g” on a canvas and then reading the pixels at a certain location may work, but we haven’t tried it yet.

I had seen this effect about two years ago on the website of Roman Komarov and was impressed by it at the time: http://kizu.ru/en/fun/

I really like how he applies the styles on hover. I wanted to do something like that with SmartUnderline but unfortunately there is no way to call getComputedStyle with the :hover pseudo-class applied. (This is necessary because SmartUnderline only applies to links which have text-decoration underline.)

Traversing the enter style ruleset is possible but didn’t seem worth it for a library like this.

Another way is through a configuration option. Something like `SmartUnderline.init({ '.content a': 'hover' });` perhaps.

While I admire the attention to detail, it is all in vain for me. I much prefer the simplicity of a solid line to the awkward fragments dodging descenders. My brain continuously tries to make sense of the incomplete line, and some words no doubt will add to the confusion with long stretches of characters having minimal expression of the underline.

Case in point, the word giggly has what appear to be specs of dust underneath.

Why limit yourself to just underlining links? Why not some other convention? I played around with this a bit over a decade ago: http://www.conman.org/people/spc/writings/hypertext/fragment... (note: In the twelve years since I wrote that, most of the links offsite are now dead, which is another issue with links).

At first, the links in the "fragment" are invisible. You can toggle among that, single angle quotes (not used in English, which is why I picked that option), small bullet points on either end, and the default HTML underline.

FWIW, my experience with your link:

The lack of any indicators at all was frustrating. I had to take physical action to determine whether something was a link. In practice I think I would just stop using the links. I think some kind of indication is needed.

The default underline was the best for me. I don't rule out that this is because I am used to the underlines. Going in I thought the dots would be bets, but I found them distracting. With practice they might be the best.

The angle brackets were by far the worst. It could be because I have seen enough HTML for angle brackets to have singificance, but I think it has more to do with parantheses and how we use them in text. I just found them distracting and they really took me out of the work.

Of the cuff thought... small superscripted dot or something? I'm used to seeing footnote or citation indicators in the corners and they generally don't take me out of whatever I'm reading.

The dots are small to begin with, any smaller and they might not be seen. There might be other, visible but not underlined techniques.

This choice came down to typographical best practices. Butterick, a well-respected typographer, recommends in Practical Typography not using underlines for stylistic emphasis. [1]

[1]: http://practicaltypography.com/underlining.html

The library's showcase page has the key bits imo:

> - It makes an assumption about selection color, which is OS- and browser-dependent.

> - Each smart-underlined link has 12 text-shadows drawn underneath it, which could potentially have performance penalties on mobile or older devices.


(Imho, the two points introduce enough risks and reasons to not use it and wait for OS- or browser-level updates instead.)

This seems like something user agents should apply to all 'underline' text-decoration by default in order to improve readability. Is there any reason not to?

iOS8 Safari agrees with you. :) (As do I fwiw.)

Unfortunately I imagine this is easier-said-than-done and may have to do more with the underlying typographical engines of OS-s than browsers. (I’m just guessing though.)

Ah, an excellent point. The TrueType font format apparently does specify an underline position and height [1], though there's no mention of any allowance for this descender-dodging feature. The fact that font formats defer the actual drawing of the underline to the renderer gives me some hope. This seems to highlight the issue of web browsers having different font rendering needs from other apps, since underlining is so much more prevalent.

[edit] Obviously, it's a lot easier to fix this issue with a single font, as Apple has done in Messages, than to fix it well in every font.

With regards to this implementation, I'd rather not have to worry about the background, font and select background colors for every link. Applying text shadows over gradient-drawn underlines is an inspired hack that I'll have to remember, just in case, but I really wish there were a better way.

Maybe we should push for a 'text-mask-underline' CSS property that accepts radius, hardness and opacity values, in order to apply a fade to the underline around text descenders. [/edit]

[1] http://fontforge.github.io/fontinfo.html, http://scripts.sil.org/cms/scripts/page.php?site_id=nrsi&id=...

Thanks for that information re:TrueType! My knowledge of these things tends to end at the browser, but I too hold hope for better type in future OS-s.

The fact that Apple decided to tackle this on iOS, an OS used on handheld devices which are typically viewed at closer distances, tells me that it’s something they feel is worth doing when you have the display resolution to do it right.

SmartUnderline makes a similar assumption, not applying itself to links with a small-enough font-size computed style.

With respect to your [edit] block:

> Obviously, it's a lot easier to fix this issue with a single font, as Apple has done in Messages.

Apple has implemented this throughout iOS 8, not just in Messages. Here’s a screenshot of Safari [1].

> I'd rather not have to worry about the background, font and select background colors for every link.

That’s actually the beauty of SmartUnderline. It figures all of these things out for you, automatically [2]. Please let us know if you run into any trouble though. :)

[1]: http://postimg.org/image/4ncgrezxr/

[2]: http://git.io/9YNfgg

Re: [2], nicely done! For a dirty hack around browser limitations, it is well implemented. It's a good thing gradient backgrounds behind text are out of style. : P


If a background-image is specified, we abort [1]. (This would apply to CSS gradients as well.) Of course we’d like to support non-solid backgrounds at some point, but short of a true HTML-to-canvas solution, there’s not really that much better you can do without wreaking lots of havoc.

[1]: http://git.io/1WqsAA, http://git.io/n4tRrw

Is this only getting the first container's background? If so, what happens when a layer below that has an image or gradient bkg?

Nope, it traverses up the DOM until it finds an element which has any non-transparent background. If the first thing it finds has a background-image, it aborts. If it’s a solid color, it returns that container. [1]

This absolutely still is prone to bugs when `position` or other properties are used to move elements visually outside of their parents, but it’s about as good as you can do without a true HTML-to-canvas implementation.

[1]: http://git.io/ChvO3Q

I can think of one reason not to: This hinting belongs in the font file itself. I don't agree with hacks to the font being done outside the scope of the source.

A few issues with that; 1) I've only been able to find evidence that font formats specify underline position and height, and leave the drawing to the renderer, and 2) most fonts were created with a desktop publishing context in mind, and 3) the web context has significantly different needs. I think there's a good case for making this change for the font renderer used in web browsers.

The point is that if ligatures are used for this, no modifications to the font rendering front end is needed. Its already there - you just need to use the feature. The way to do that is with ligatures in your text..

That's not what ligatures are for; ligatures are for joining two adjacent letters. I don't believe font formats support this descender dodging feature for underlining, hence the need for the renderer to take charge.

Well, I would argue that using a ligature to establish the alpha gap around the lettering is a better use than hacking up a non-standard feature to the renderer. The gaps are attachments to the type, so ligatures are appropriate - since they would be used in this case to join/split the underline - which itself could be argued is a ligature, per-face, also.

This example shows once again that CSS is not a good declarative language: it provokes hacking. In this case, a hack has to be applied in every instance where this is needed.

Instead, if CSS were developed by actual language designers, it would have been possible to write this as a modularized thing, completely hiding the actual implementation.

Aside from the many other criticisms raised in the comments, this seems like something of a deal-breaker to me: "It requires that the text be on top of a solid background color."

It's a restriction, yes, but far from a deal-breaker, I'd say. There are very, very many websites out there that have text on solid-coloured backgrounds.

This seems like a hell of a lot of client-side processing just to enforce a trivial (and arguably bad) typography preference.

Web 1.0 called and they want their underlined links back.

I'm confused, is there some other way we're supposed to indicate something is a link now? I thought was considered bad practice to mess with link styling too much for usability reasons.

That's nice in theory, but it's also yet another example of Apple innovating and getting it right, and people implementing a quick copy without going through nearly as much necessary design thinking as Apple did. The end result is that only Apple's implementation is better than the default one. Of course I'm not blaming you personally for that, I mean, even Microsoft has fallen prey to this phenomenon (and on a vastly larger scale). But it would nice if people were more observant.

I appreciate your feedback. There are always ways of improving something, even in the case of Apple’s implementations.

Are there particular aspects of this implementation you feel make it worse than Apple’s?

Ultimately we see this as an incremental improvement to the default text-decoration underline styling provided by most operating systems and browsers. However, we’re eager to work with OS community to improve this even further. Thanks again!

Sorry somehow I thought that was obvious. Basically they cut the line very close to the letter, that way it doesn't break the line visually, which is what the other comments here on HN seem to criticize. It's a very simple and subtle thing but that's what makes the difference. My point was that sometimes the incremental steps between the default design and Apple's design are worst than if you just kept to the default design. You have to make a big step before perceiving benefits. Sorry for the pigeonholding though, that was more of a general rant.

See my comment above [1]. I think you’ll find that the iOS 8 and SmartUnderline implementations are closer than you think.

[1]: https://news.ycombinator.com/item?id=8588492

You keep repeating that comment but no, they are not. The idea is clever and the result is pretty good, but what the parent is saying – that in your implementation the line looks more broken up because it's more separated from the glyphs – is true.

You may be able to fix that by tweaking the shadows but as it is currently implemented it works pretty well for larger display text but not so much for smaller body text.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact