Hacker News Reader

Comment by cosmotic | Hacker News Reader

HTML/CSS based native interfaces are some of the hardest to use, least reliable, least accessible, least robust interfaces I encounter; The demo interfaces on their website all fit into the unusable pot.

I admit most/all of the interfaces I've personal used are using electron/webkit/ie to render which are all quite bloated. Is this different? It's not entirely clear and would be an important differentiator.

I still don't understand the draw though. The best native apps look and behave natively; using a platform like this prevents that or just makes it harder to accomplish. Using such a platform seems like it's entirely developer-focused and not-at-all user focused, which is a shame.

View on HN

I had a similar discussion on reddit a long time ago, and posted this screenshot of various Linux music players compared to Spotify.

https://i.imgur.com/K44tJ2Y.png

Lots of native UIs (especially from the past) tend to all have the same "grey boxes" look. I don't see how that's universally better.

Another example was this part of Slack UI https://i.imgur.com/O71fdP5.png - how would you even do something like that with a typical native library? It would certainly take a lot more effort if you had to build that from a limited set of pre-existing components, instead of relying on the more free-form "rectangles and text" building blocks of the web.

> Lots of native UIs (especially from the past) tend to all have the same "grey boxes" look.

it's not "grey box", it's "uses the native theme of the user's desktop". e.g. I use strawberry music player, if I am on a computer with a dark theme it's dark.

Qt definitely allows slack-like look though if the developer opts-in to it - e.g. Telegram Desktop (https://cdn2.nextinpact.com/images/bd/news/163843.jpeg) is made with Qt Widgets.

Ripcord is another Qt app which is a discord / slack client made by a single guy, which goes more for the native look'n'feel : https://cancel.fm/ripcord/

(contrast this with slack needing dozens of developers and being at their third rewrite of the app because of performance issue :-))

Interesting, I just assumed Telegram Desktop is electron app since it behaves similarly to one.

In the end I don't care how it's done as long as it works fine (which it does).

"behaves similarly" How?

Besides Telegram, I have also electron based Rocket Chat and Slack (different teams, different apps :/) and all these apps look and feel very similar.

Telegram uses this non-native visual language which is very common in electron apps.

> free-form "rectangles and text" building blocks

As a backend dev, I find that the most annoying part. I don't want free-form; I want something I can put together in a few minutes without having to think about design and it still having it look acceptable on the platform I deploy it on. On the web I can do that with bootstrap (and it's 100.000s of themes), native I can do this with the native libraries but in the html5-for-desktop-apps space it seems I have to spend time on doing stuff I don't want to do or even care about to make it look anything different than vomit; that is simply not worth the time. I know that is very different from your use-case, so not saying you are wrong or anything, just that this is what annoys me about html/css. I think there would be a market for 'electron desktop themes' like (1) (basically unsupported now) but seems there is not (or no-one finds it necessary anyway; maybe i'm the only one with this 'issue', well, me and all my colleagues that is).

(1) http://photonkit.com/

A point I think is worth noting is that Linux software is far more likely to be FOSS, and FOSS is far less likely to have dedicated designers. LibreOffice wouldn't look out of place in 2002 for example, but it's not because of the platform/SDK it runs on. Whereas Spotify most definitely has a whole team of design staff.

That aside, if the rest of my OS has a "grey boxes" look, I would like my software to have a "grey boxes" look.

Sure, your design may be pretty, but if it looks and works completely differently to my other applications and the OS itself, I consider that a pretty huge point against it.

At least on those "grey boxes" buttons look like other buttons, there is good contrast, they blend in with the other things in the UI and don't stick out like a sore thumb.

One can preach about "brand identity" the whole day, but as a user, I couldn't care less about your brand. Integrate or die!

> Integrate or die!

While i otherwise agree with you, the "die" part requires users to value such integration and they do not seem to be doing that enough.

Nearly every native UI framework I’ve touched allows for a rich customized look of UI components. Often this means subclassing an abstract view thing that opens a toolbox of fine grained drawing commands.

Your examples of boring native UI is the result of developers that don’t bother to exploit these features for whatever reasons- probably because it can be extremely time consuming to build native custom UI and adds mental overhead unless one has/is a dedicated designer/artist.

I’m not sure which part of the Slack UI you believe can’t be done natively? Flat colors or the revealing menu?

I'm not saying it can't be done, I'm saying it would take a lot more effort (by a huge amount, plus more in prototyping phase of development when you're trying stuff out), and in the process you will likely lose many of the advantages of native UI that were mentioned (consistent behavior etc.).

Take for example that "Jump to" text box - if your native UI toolkit doesn't let you add icons into the placeholder text like that (does it even have placeholder text natively?), you'll have to deal with drawing offsets, caret position, or possibly nested components, overriding focus outline, in the worst case resizing the stuff when focused, and other stuff that makes the whole endeavor not worth it due to the added complexity (assuming that text box works how I think it works, I haven't used Slack in a while).

Huh, so is the “jump to” text box easy to do on the web? I’m not very familiar with html/css, so genuinely curious.

And I don’t think it’s quite as hard as you describe natively, at least on AppKit/UIKit, windows, and android.

No, you'd use Qt where this kind of text box would be straightforward to implement.

It would still look mostly native (even though it's clearly not a native control), be very performant and use a tiny amount of memory.

> I had a similar discussion on reddit a long time ago, and posted this screenshot of various Linux music players compared to Spotify.

In 90s this was solved by using "skins" which were actually drawn in graphics programs and then spliced and edited in some proprietary editor and/or positioned by hand crafted config file.

See https://winampheritage.com/skin for example. You can't go more original than these. :-D

I don't see a point why you would need web library for this.

Skins like that don’t scale. Fine if you’re still using a 1366*768 laptop, but even 1080p let alone 4K makes everything tiny or blown up and blurry.

Doesn’t mean you need electron to fix it, it can be done with a native UI. But bitmapped skins are generally not used anymore for a reason.

> it can be done with a native UI.

Yes, but with what price?

We (people outside of Microsoft, Apple and Google) cannot afford UIs nailed down to pixel grids and particular trends.

Years ago it was Skeumorphism, then it was flat UI as Metro, today people started speaking about Skeumorphism again. CSS is just a convenient tool to follow the trends.

And "yes" it "doesn’t mean you need electron" as there are other options :)

Take Sciter and forget about UI problems and "resource hogs" for years to come. HTML/CSS will be with us together with the Internet. And so are UI developers that can take care about it. And C++ developers in case you will need to fix something critical inside the Sciter.

You're taking a lot of crap here. I just wanted to chime in to say I thank you for trying to build something like Electron that's still lean and efficient. On top of it, I always use your licensing plans as a positive example in discussions about beneficial, but sustainable, licensing strategies. licensezero.com comes to mind. Keep up the good work.

Drab gray boxes may not be as 'engaging' or 'visually rich' but they are more accessible and usable. You know what's interactive, what components relate to what and how they affect the application.

The slack UI is a counter-example of a good UI. It's confusing, it's unpredictable, it's inaccessible. They made the application look like a fancy website instead of an application. That's a bad thing, not a good thing. There should be no attempt to replicate it. Using it as a benchmark to evaluate how flexible a UI framework is a non-starter.

I'm not 100% sure but I think the slack design can be achieved in wxWidgets. Yes the native apps look bland but why would I use an app that requires 500M to play music?

Why not? If it works ok and doesn't eat battery/CPU too much when in background, then who cares? For exapmle I like that I can plug my phone into the speakers (yay headphone jack), and control the music from a laptop on the other side of the room.

Also, it's more like 200 MB on my computer, most of which is predictably CEF. In comparison bundling Qt dlls would also add like 50MB if I remember correctly.

I'm talking about RAM not SSD. And no I don't have multiple computers in my house. A single laptop with 16G of RAM is all I have. I have to use it for work. We all know about music apps playing music while using less than 50M at any given time. If its in the background then even less. I know the dev experience is far greater in electron but its simply not performant. But if I were to launch a product on all platforms then I'll go for electron probably cause its the fastest to market

RAM's about the same, hovering slightly below 200MB (Windows shows a sum of all the different Chromium processes). Which for me is firmly in the "who cares" territory. Lightweight music apps don't have the online playlists, discovery mechanisms (Discover Weekly, etc.) and all that other stuff.

They could do though. All the smarts of Spotify comes from their APIs. There's been a bunch of third party low-resource clients / integrations for spotify; despotify, mopidy, recently this thing: https://github.com/hrkfdn/ncspot

Unfortunately iirc the neat dynamically generated playlist stuff isn't supported because it's not exposed properly through their APIs. Certainly when I tried to use Spotify's own libspotify its functionality was crippled to static playlist management, searching for stuff, and playback. despotify was in a similar state at the time.

Also - Spotify sucks HARD at doing all the stuff it does, and I blame this entirely on the UI implementation. The service, I love. The apps, I have become to loathe. They used to be nice and quick and responsive. Now? It can't even get shuffle right!! My mid tier smartphone takes 10 minutes to load search results or my own playlists sometimes; something i've encountered across multiple devices. The desktop app crawls on my 24gb 8-core workstation.

Conversely, I can have a 200gb library of music loaded into mpd or foobar2000 or vlc in all sorts of exotic (and higher bitrate) formats and from weird and wonderful local and remote storage and play it all on shuffle and have all the albumart showing and it organised neatly into a searchable library via the id3 tags and have volume normalized with replaygain and it'll do it in under 20mb and has been doing so for a decade.........

Sciter's author here.

> HTML/CSS based native interfaces are some of the hardest to use, least reliable, least accessible, least robust interfaces I encounter

This has nothing with HTML/CSS per se but rather questions to particular products. Same complain may apply to any UI.

> The best native apps look and behave natively;

"Native app" term these days is something almost non existent.

In these terms Microsoft Office is not a native app as they use custom UI framework for UI. Pretty much whole Window 10 UI now is not "native" in this sense. UWP is a custom UI layer that uses windowless DirectX based UI. There are some low level utilities left in dusted corners of Windows that can be classified as "native" but there are just few of them - no one cares.

Consider my https://html-notepad.com as an example (purely sciter app and so HTML/CSS UI), this application simply cannot use "native UI". No OS has needed components out of the box.

Yet pretty much any modern app must have one way to another to render and to produce HTML - we want our applications to be connected (read: interact with the Web).

UWP are an interesting example. They are the most 'native' (at least looking) kind of application on windows 10 but I realize they aren't using win32 (which I think is what we both mean when we refer to native). Regardless, UWP are terrible; they are very hard to use and look awful, if not flat and uncluttered.

Office is a bad example as well; that team is notorious for choosing new/different interface controls just to be different. They try to back it up with studies but the studies are flawed. Accordion and Ribbon come to mind, as does the silly fake shadow they had a few years ago.

Still to this day, the drab boring win32 applications are the best performing and easiest to use.

> drab boring win32 applications

For example?

It seems to me at this point that UI/UX developers just refuse to learn low level languages. Qt applications for example are extremely fast and cross platform.

It seems to me that wanting UI developers to learn systems languages is bizarre. We don’t expect Linux kernel developers to learn UI/UX, why would the opposite be the case? They are working on a vastly different abstraction level.

Also, yeah, Qt is indeed fast and cross platform. But somehow every single mainstream Qt application tends to look visually terrible and not any more native than Electron. E.g. Ripcord posted above, or Calibre. Meanwhile, mainstream Electron apps may be hogs in terms of performance, but they look polished.

Finally, Qt’s weird and deceiving licensing is a major reason I would never touch it with a 10 foot pole. I am not a copyright lawyer and have no desire to figure out how I can comply with LGPL, when Electron is MIT.

> Finally, Qt’s weird and deceiving licensing is a major reason I would never touch it with a 10 foot pole. I am not a copyright lawyer and have no desire to figure out how I can comply with LGPL, when Electron is MIT.

you know that Electron uses Blink which is LGPL right ? It also ships by default with ffmpeg which is used by chromium for video rendering, which is also LGPL (https://source.chromium.org/chromium/chromium/src/+/master:t...).

See https://lists.gnu.org/archive/html/directory-discuss/2017-12... - and some LGPL-prefixed files from Blink's source code if you don't believe :-)

https://chromium.googlesource.com/chromium/blink/+/master/So...

I'll let you note the large irony of WebKit & thus Blink & Electron originally coming from KHTML, that is, a KDE project originally built with Qt - Lars Knoll at the top of that copyright header is currently CTO of The Qt Company.

VSCode is by far the best Electron app... I mean by miles. It occupies a niche though where people tolerate and even want peculiar user interfaces that are highly optimized for their specific job-- in this case being a code editor and IDE. Electron and other HTML based UIs suck for more normal sorts of apps. They're bloated and look alien and clunky.

The 'native' 'modern' apps on Windows 10 are not known for being small, fast, or especially well designed either though.

I developed a number of WFP apps (starting with the earliest versions) and it was very very far from being easy compared to html. Debuggability, Stability, Performance were all sub-par.

Reply on news.ycombinator.com