Electron considered harmful
drewdevault.com
drewdevault.com
But then the essay turns into a rant. It's 2016, people, the ship has sailed on pretending JS isn't a real programming language. Phrasing the argument the way he does just betrays his smug elitism.
The fact is, for many types of applications, Electron is a really good fit: for example, shipping a single-page webapp as a desktop app, a need that surfaces often in today's world, despite his derision of this concept.
An Electron-wrapped SPA is safer than going to the same website in a multi-tabbed browser and co-mingling your disparate cookies in the same jar, not to mention more convenient for the users, because they can treat it like any other desktop application -- with hardware access, and subject to the OS' native task management.
For other apps, Electron allows the world's most popular layout system (HTML+CSS) to be used with one of the world's most popular programming languages. This is a huge boon for re-using knowledge, and perhaps does lead to some amateurish projects. Similar criticisms were leveled towards Visual Basic, where drag-and-drop composing and a good IDE enabled people to quickly crank out a GUI app from an idea.
Just wanted to name the most obvious alternative: embedded browser component like QtWebKit, which allows to code most of the application logic on the language of your choice. Embedded browsers are the way people have been doing it since circa 2000 (by then, Windows was dominating platform and you could use embedded IE for rendering - the typical app size was much smaller than 200 Mb btw). I remember earning some money on a side project around that time - the CD catalog of bathroom equipment, which I implemented by borrowing huge part of my customer's website and embedding it in CBuilder app with few improvements for richer UX.
I hope you're being facetious here. If you develop any kind of desktop application and truly believe this, honestly this is disappointing to read.
I've spent a lot of time on team web-apps-suck, and in many ways I'm still a casual fan. But the truth is I can go from zero to good enough much faster with web apps than anything else, and unfortunately most people only want to pay for good enough.
Source?
Really, I'm not at all convinced that whipping up an Electron app is inherently quicker than whipping up a native app. Especially when we have robust, mature RAD tools for native toolkits, and things like Qt Quick, WPF, and Java FX let you go the whole "write your GUI in some markup language" route, if that's your bag.
Not that there aren't a lot of posts here defending Electron that basically say the same thing but don't you feel embarrassed in the slightest for saying that?
Yes, there's a lot of things we, as coders, can do to make our lives more convenient by treating bandwidth, RAM, and CPU wastage as an externality that can be shoved off onto the end-user to pay for. That doesn't mean we should.
Therefore all claims that electron bloats 200Mb are just temporary. Its still a relatively young technology with huge potential (which also means huge potential for misuse!)
I spent 5+ years writing native GUI apps in half a dozen frameworks, I hope to never do so again unless the project's requirements ask for it, which I don't see happening for most use-cases out there.
Besides, almost nobody writes true native applications anymore - its all C# now which is ironic as its also extremely bloated with huge pools of average developers at best.
All of them have at least 2Gb of virtual memory being mapped to them. Yet I still have 9Gb of free ram.
If you want to go against bloat, you might just as well uninstall every single software from your computer, including its OS.
Also, you're arguing in typical FYGM fashion. Not a good look.
No idea why VSCode is being sparing for you, but i've seen plenty of electron-based online messengers use that much just by running.
I call bullshit on that one.
That's fine, we can return to this topic when it will become a mature launcher app. The thing is that today it's not in that state yet.
>Besides, almost nobody writes true native applications anymore
True native applications are not just the applications written using some native API and platform-specific low level language. Native app is the app that comfortably exists within the operating system environment, follows the UI guidelines and integrates with native UX and performance enhancing functions of the OS. For example, some universal Windows app written on C# is native on Windows.
Using the date as an argument is also poor form. JS is a terrible programming language (I'm not denying that it is a programming language though) regardless of how popular it is (although I think I might be going against the grain here on HN, many people here seem to enjoy node.js).
> The fact is, for many types of applications, Electron is a really good fit: for example, shipping a single-page webapp as a desktop app, a need that surfaces often in today's world, despite his derision of this concept.
I feel it's unreasonable to turn a single web page into a desktop app by bundling an entire web browser with it.
> ...because they can treat it like any other desktop application -- with hardware access, and subject to the OS' native task management.
If you need direct hardware access it probably isn't well-suited to being a web app in the first place? Why use web technologies for it then?
> For other apps, Electron allows the world's most popular layout system (HTML+CSS) to be used with one of the world's most popular programming languages. This is a huge boon for re-using knowledge, and perhaps does lead to some amateurish projects. Similar criticisms were leveled towards Visual Basic, where drag-and-drop composing and a good IDE enabled people to quickly crank out a GUI app from an idea.
HTML/CSS can be nice to have in such a situation, although a refusal to learn other GUI frameworks probably makes for a poor (native) developer.
The resource bloat and (usually) extreme disparity with the native user interface rarely makes it worthwhile for me to use an electron app. It feels like forcing a square peg into a round hole because the developer is seemingly unwilling to consider more appropriate technologies outside of their comfort zone. Yes, there are well-made electron apps (like VS Code mentioned in the article), but I, personally, would hate for it to become the norm.
I can never see the point of comments like these. Not one person who reads that will have his mind changed.
HTML/CSS can be nice to have in such a situation, although a refusal to learn other GUI frameworks probably makes for a poor (native) developer.
No it doesn't. Either the developer benefits or doesn't benefit from "refusing". There's only so much time in the day. I'll refuse to to learn Qt or gtk+, unless there is some overwhelming benefit for me to learn it.
I wasn't expecting to change anyone's mind with that. There are more robust programming languages that don't have the same inconsistencies and idiosyncrasies and encourages better practices in general (even though it is true than you can write terrible code in any language).
> No it doesn't. Either the developer benefits or doesn't benefit from "refusing". There's only so much time in the day. I'll refuse to to learn Qt or gtk+, unless there is some overwhelming benefit for me to learn it.
That is true, but consider that the end product is not necessarily meant to be most convenient for the developer. Keeping the user in mind when making these decisions may make you consider more native and less resource-intensive alternatives. I have read in the past that Electron also posed some difficulties for screen readers, for example (although I'm not sure if that is still the case), which might not be something that impacts you as the developer, but which can make it harder for someone to make use of your software.
You can make an argument to say C++ is terrible, Java is, Ruby is, Python is, and it means absolutely nothing other than showing of your own insecurities. You can build great and mediocre things in any language and any platform. People do every day.
Also, if the developer is not one of the users of the software, you've already thrown a good chunk of quality out the window right there. You're simply not going to have the vision required to build the optimal user experience.
> but consider that the end product is not
> necessarily meant to be most convenient
> for the developer.
For that reason, your post along with the blog post comes off as a tantrum. And if users really cared about these things, then surely you'd be able to swoop in and capitalize on it. But we both know that's not happening.
Usually the choice for the user is between "have an electron app in a month" and "have native app in a year", most users prefer to not wait, and when native app comes in a year, electron app will have much more polish, and run acceptably on new hardware.
The issue is not about 'protecting poor users'.
Most electron apps are quick experiments, to test out an idea without spending too much resources on it.
If idea is great, and performance is an issue, and there are enough users to get a team of highly skilled developers to work on it, they'll create native app, replace electron version (and then embed js vm to support plugins:).
> I can never see the point of comments like these.
At minimum, I think the hope is that js-only programmers become aware that there are alternatives. And hear that the people who use those alternatives often find them better.
I have used a number of different programming languages before JS, a couple of them more extensively and for a longer time than JS. I like JS. I understand its flaws. I also understand that all programming languages are terrible in their own unique ways -- I have accepted JavaScript's terribleness and a lot of it is addressed by ES2015 and beyond. I prefer JS's terribleness over that of other languages.
I like to make fun of PHP programmers the same way, but I understand that there are PHP programmers who are able to see beauty in PHP or have found their own perspective of PHP which makes it less terrible than the alternatives. I'm not one of them but I can relate to them. It's called empathy. OP lacks it.
That said, yes, somebody ought to love JS in a not blind way. Big numbers work that way. If that's you, I don't think those comments are for you. All languages have different flaws, every flaw relevance varies by context, and contexts are incredibly subjective. Yet, some contexts are much more common than others, and many people preach in support languages that are harming them because they don't know any better.
Without context it sounds like you saw a bunch of nested callbacks in a Node.js example and judge the entire language based on that.
I've seen enough different use cases and teams and programming languages to understand there's no point in being arrogant about language choice: if a team already knows a given language, it is most likely the right tool for the task, unless the benefits of replacing it are worth the overhead of properly switching to and gaining competence in another language.
[Citation needed]
Sorry, JavaScript may have gotten lucky being the lingua franca of the web, and it might not be perfect or even great, but it's "good enough" for most purposes these days.
If it were "terrible" you would expect at least one compile-to-JS [1] to have gained a non-trivial amount of traction by now. The supersets of JavaScript like Babel, TypeScript, and Flow are more popular than "not terrible" languages like ClojureScript, Elm, etc.
1. https://github.com/jashkenas/coffeescript/wiki/list-of-langu...
There is more there than dynamic typing.
Ruby and Python do not fit it.
1 - Or, in the case of Python, mostly dictate.
It's fine to have some kind of option for data whose type isn't known until runtime. This stuff happens, and languages that insist there's no such thing make dealing with it more difficult. But this is the rarer case! Why optimise for it rather than for the more common one?
Dynamic typing isn't optimization; dynamic typing without any static analysis is the utter lack of optimization.
The very same can be said about people looking down on Electron. Most of them can't write decent native apps to begin with.
I used all of win32, Qt, GTK, wx and Cocoa to name but a few. They all suck in one way or another. Suggesting people are poor native developers for refusing to learn these frameworks completely disregards the fact that most people using these frameworks are bad developers to begin with.
For the average application going native will easily triple your development time and costs for absolutely no noticeable performance difference.
> If you need direct hardware access it probably isn't well-suited to being a web app in the first place? Why use web technologies for it then?
What if the only difference between the web and desktop apps are that the desktop app can open local files while the web app downloads them? Are you going to write two apps in that case? This would be a mediocre economical choice at best as you're more than doubling the production costs for a difference nobody is going to care about.
Sublime text still smokes Atom and VSCode in the performance arena.
That is not a useful indicator. There's a reason that 'editor-wars' is a running joke.
Because people don't know any better, shun anything that isn't their pet technology and generally sit on the wrong end of the Dunning-kruger effect.
Why isn't everyone using Spacemacs then?
Visual Studio is much, much slower than all of them yet its still an incredibly popular IDE for some reason. It has endless flaws, one of them being its still 32-bit and often crashes on larger projects. (The last game I ported to the PSVita would hang VS for minutes sometimes, often ending with out-of-memory errors, we'd reboot VS between every profiling session.)
For a good chunk of users, features and ease of use are more important than performance (especially when most of the time it really, really doesn't matter for an editor.)
This was the exact problem I needed to solve that led to my first Electron app.
Needed to recursively crawl a directory and generate a report on all pdfs found and the total pages in each pdf.
It seemed silly to learn a new language and ui kit when I could glue together some wen technology over a weekend and make it happen.
http://designbymobi.us/how-i-built-my-first-desktop-app-in-3...
> It feels like forcing a square peg into a round hole because the
> developer is seemingly unwilling to consider more appropriate
> technologies outside of their comfort zone
It just makes you feel good to dismiss other people as lazy or unwilling to venture outside their comfort zone, but that's more of a revelation about how you view others.
oh do us a favour, that's just YOUR opinion
Valid points are:
* Electron is often too much for the purpose
* it does not look native and the devs don't care
* often there are better tools
this is not valid. no need to limit our use case to devs, either. at worst, users don't care. at best, users prefer it. see Spotify, Discord, Skype, Sublime, Chrome, Steam... should i keep going?
whenever i open an app that uses entirely native controls i just expect the author didn't want to invest any time into making their app look good.
I don't know if there are any studies into what "users" want, but I'm with you. I prefer a UX that fits the app best, not the OS. Especially because in my experience I'm using the same app across multiple OSs and I'd like the same layout in each (within reason)
Well, if you'll accept a single data point, I dislike using them. If there's a native alternative I will always go for that. If there is not, then sometimes this tips the decision to not install.
I really don't think its such a great idea to reinvent the UI wheel, throwing the decades of research and refinement represented by native UI toolkits under the bus for the sake of branding and aesthetics, unless you have a good reason — a really good reason, like having to create brand new UI for a task that prior hadn't been performed on computers before.
I also see electron apps as stopgap measure, not a solution, and this feeling is reinforced every time I open up Activity Monitor and see Slack or Spotify sucking up upwards of half a gig of memory while sitting there doing nothing.
It's a romantic thought, but that boat sailed a long time ago. Users got used to the web that had a very inconsistent UI between websites and on the website over time; before bootstrap, you would be hardpressed to find 2 websites with the same button styling. The last time the web used unstyled OS-controls was the 90s (with that default grey background - blech).
However, on Linux and Mac it's a little different. On Mac you'll see many apps compromise their branded style to look more Mac-like in style. Meanwhile on Linux most people use apps entirely fitted for their desktop environment, and then it's branded apps that look out of place. Windows attempted to make it's desktop more this way and failed with Windows 8.
Yes an no. Windows is very versatile when it comes to design. However, if you just drag-n-drop elements in Visual Studio[1] or Qt, you get that "ugly" windows app. You'll notice plenty of Microsoft's own designs (as part of the OS, or as external apps) that look very well and fit in with the rest of the OS. There are also a few gems from 3rd parties.
> On Mac you'll see many apps compromise their branded style to look more Mac-like in style.
I agree with this, but I also believe it's an "optical illusion" in the sense that the GUI isn't very Mac itself, but it just borrows some design conventions used by the OS to make it easier for the user to understand the application.
> Meanwhile on Linux most people use apps entirely fitted for their desktop environment, and then it's branded apps that look out of place.
As a full-time (or at least 99%) Linux user, I tend to disagree with this. I know what you mean and you're sometimes right, but it's not really that black and white. a GTK app on Qt/Plasma won't look that nice unless you go out of your way to make it look nice (which is usually a lot of work), and open another GTK app and realise it looks broken, so you need to add more tweaks, and then you open the third app and give up. Same goes for KDE/Qt apps on Gnome.
So much of this! This is my exact sentiment, too! If you want your application to look different from the OS-provided GUI—which I think is a good thing—without investing much effort into implementing your own rendering/painting engine, the web is an excellent platform, not to mention there's much more you gain with it, too.
What bothers me is that the author focuses on a few bad points of electron, whilst there are also many benefits to using it.
"Electron is a really good fit: for example, shipping a
single-page webapp as a desktop app, a need that
surfaces often in today's world, despite his derision
of this concept."
It's been a while since I messed with it, but MVVM didn't work so well for some kinds of things, which seemed to be why for example selected items property is often not a dependency property of MS supplied widgets (so can't be participate in binding). For example see http://stackoverflow.com/questions/22868445/wpf-binding-sele... (from what I remember this accepted solution also has a subtle bug of double notification (or was it lack of notification on one side?)).
Will my Cocoa app run on Windows, Linux and the web?
(A well-designed web app built with Electron will).
Why should any company that wants their product to be widely available invest in Cocoa developers?
I prefer native Mac apps to cross-platform ones. Serious companies should build native apps and keep UX guidelines of each platform in mind.
Its no wonder all the best native applications ALL use proprietary toolkits. Adobe being a prime example. Such a toolkit is orders of magnitude easier to reproduce in HTML/CSS than it is in any of Qt/GTK/wx and others.
Thinking a web app can easily be expressed as a systems application grossly overlooks the complexity web apps have scaled to.
Forget for a second that the performance will be terrible because it's not native code. Forget for a second that a desktop application typically benefits from a more dense UI than a web application, and may benefit from such desktop-only UI features such as separate tooltip windows or OpenGL rendering (yes you can do that in a browser but come on it's not the same). The fact of the matter is that the beancounters typically care less about delivering the best possible UX and more about how their bottom line is going to be affected as they continue the corporate strategy of expending 20% of the effort to get 80% of the potential business benefit.
Electron was built for the beancounters. Not for the HN purists. Simple as that.
Is that necessarily a bad thing? I'll throw a great counter-example to you, Mattermost. It's a great open source project that uses Electron to build a client app for just about every single platform.
Have you maintained a cross-platform app for Windows, OSX, and all the various flavor of linux? I have, it's hell. You have compiler differences, ui semantic differences and a litany of per-platform quirks.
Does Electron solve every single use case? No. However in the cases where it makes sense it's a great tool.
Also for all the kids out there, Electron is the second coming of the MSHTML ActiveX engine.
I didn't like it then, surely don't like it now.
Really? I've always thought of it as the "blinders-on dev's XULRunner".
This is a status bar icon that notifies you when your build is broken. Cool idea, but what the heck does it need electron for?
http://catlight.helprace.com/i15-why-is-setup-so-big
Electron is useful for some cases, but this is not one of them
Electron was built to ship products as soon as possible. Most of the time, no one gives a flying fuck if the tech stack is the "right" choice and as efficient as it can be. I'd be willing to bet my net worth your users don't care at all how you built it, just that it does what they want it to do.
In the end, for most companies, it's more about shipping product that sells than spending 5x more time writing it in C# because "you're an engineer so you should act like an engineer." That's not how the world works. And until you figure that out, you'll constantly be left behind by others who made compromises to capitalize on the market.
Early optimization is the root of all evil.
That's exactly right, except "what I want it to do" is not use up all my RAM.
I think it's possible to build very well performing applications on a tool like Electron, and once the market has proven that your product is in heavy demand, then you can re-approach it and decide to go native or revamp it using some lower level tools.
First off, JS is a perfectly usable language. In some respects, it's even great: of the "big three" scripting languages (Python/Ruby/JS), it's the only one that got Lambdas right, and also got closures right (unlike, say, Python). It has sometimes-quirky syntax, and some odd semantics, but they don't bite as often as you'd think.
Second off, I would say that this article is right about the bulk of Electron: It's a pretty massive thing to lug around with your app, and not ideal in any sense. However, if you want to build a UI very fast, there is nothing better than HTML/CSS for it. Except maybe TK.
Speaking of TK, I would actually advocate TK over Electron: it's easy to use, has bindings to pretty much every language out there, and with TTK, it doesn't look like crap.
Finally, "considered harmful" essays rarely do any good, and frequently do harm.
So yes, while I usually like Drew's writing (at least a little: he's good, but far from the best), I wish he had spent the time he spent writing this doing more work on SCAS.
Seriously. If SCAS was actually done, and documented, and usable, that would be great. There is a freaking vacuum of good assemblers for the platform: another one would be nice.
There are many good languages out there, "when used as directed". Alas, out here in the flyover country of Sacramento, there is really only demand for 3 kinds of jobs: Java, C#.NET (falling off), and some Javascript, ignoring the long tail of one-of-a-kind shops. Given those choices, I'll take the Javascript. Java 8 lambdas are clumsy to use to assemble much in the way of higher order functions.
I mean, cuz, who doesn't want to put this as a parameter into a function/method header:
(..., BiFunction<Map<String,Object>,Map<String,Object>,Map<String,Object>>, ...)
E.g. - for the sake of argument, say I needed a callback to merge two associative arrays. Oh, you needed a third input argument? Write your own interface type, so you can use it to declare the callback parameter.
No, I'll take the JS, thank you very much.
(as we discussed a month or two ago, Java is a [somewhat] new Business Oriented Language, and it's pretty Common)
But where I my manners? Hi Robo. :-). It's been a while. Actually, it's been so long that I only hazily remember the contents of our discussion.
Another thing that JS is great at is the flexibility it provides: objects are very general and you can do almost anything with them.
Anyone would care to explain why does JS take so much hate?
First off, it gets a lot of (undeserved) hate from the Java-type people who think they know what OO is, but don't (because Java isn't OO, or so says Kay, and even if it is, there's a whole lot of rubbish mixed in. For that matter, I don't think JS qualifies either, by Kay's definition, but it's closer). These are the people who say things like, "How can it be OO if it doesn't have classes?" and, "where are my Refactoring tools?" or "Design Patterns!" and implement a bunch of java-esque "classes," and proceed to write IteratorStrategyFactoryManagerService (speaking of which, if you have any actual "classes" in JS with the word "Strategy" in the name, indicating they implement the GoF Strategy pattern, you should stop now. And feel bad about yourself. But that's another story).
With ES6, we have "classes," so these people can stop complaining, but it's still bad, because they're actually syntactic surgar over prototypes, so you still need to understand prototypes (to say nothing of how "this" works, a genunine mess in JS, but a necessary one), but it's no longer obvious that that's what's happening, so you might think that you don't.
The more significant complaints are about Javascript's famous weak typing (quick, what's []+""? What about ""+[]? And {}+[]?), ASI (which is why if you're returning an object/array, you can't put the openbrace or openbracket on the next line, and why semicolons only sometimes matter), and general quirkiness (the aforementioned this, and so on).
These are actually legitimate reasons to dislike JS, but in my experience, they're not as severe issues as they're made out to be. If you really can't take it, though, TS or Flow can help you.
From what I have read by/about Kay, it sounds like his vision of OOP was very much about the actor (message queue) model, rather than (just?) a Simula 67 interface/protocol method table structure.
As for the goofy "nonsense evaluation" in JS, it is a bit unfortunate, but I guess I must just write less nonsense than some others, as this doesn't seem to be a real problem for me :-)
In fairness, the JS I write is client side, and it usually doesn't take very long to get to a state to test new code, as opposed to a long batch process.
There are no abstract classes in JS: there doesn't need to be.
Anyways, that's not what I mean: I was talking about the Strategy design pattern, which is actually a way of implementing higher order functions in languages that don't have them: thus, implementing classes for it in JS would be very silly indeed.
As for Kay's vision of OO, it involves not only everything being an object, but extreme late binding. Java has neither.
I think one of the best points PG (of YC) ever made was about programmers thinking in a particular language. "If Blub doesn't do it, you don't need it" (paraphrasing). Convincing the C++/Java/C# fans they need other features sometimes is quite hard.
The answer doesn't really end up being static vs dynamic, but rather how dynamic languages often lack the other facilities that make programming in the large easier, and Eric Lippert uses js as an example.
Companies with a lot of money at stake have spent a lot of money and effort on things that transpile to js. The people making these decisions are almost always quite smart and very practical. They don't care if first class functions are sexy. They just want to write high quality code quickly. If they've spent so much time wrapping it up, you know it's because they had serious issues being productive at scale with js. They, and people on their team, probably got extremely frustrated along the way. Hence the hate.
>They don't care if first class functions are sexy.
It's not about being sexy (and believe me: first class functions aren't): it's about getting a very useful feature right, something many languages seem to have trouble with, despite the fact that Lisp (and ALGOL, IIRC) got it right in the 50s.
Otherwise, you certainly had first class functions, just not always with any external state captured.
https://en.wikipedia.org/wiki/Man_or_boy_test
I suspect TurboPascal was in the "boy" camp, in that a reference to a function (or procedure) would not have any external state bound to it.
I think your attitude is more that nothing is really bad with js relative to other languages, more so than actually trying to understand why it draws a lot of criticism, which is what you claimed and what I tried to link to.
(I'm ignoring the error handling thing, as I have not done enough server side JS to have a good feel for how the Node API handles things like I/O failures)
IDEs are starting to have better support for Javascript (or similar dynamic/functional languages). In the near future, the IDEs should effectively have type inference. Once you relegate (mutable) OOP to a tool of last resort, and embrace immutable value objects plus (higher order) functions as your primary methodology, you get the automated error checking and much less drivel to read through during maintenance.
OTOH, I suppose this means the dynamic/FP languages will remain a productivity lever for motivated "experts" only, sadly. At least until there is enough butt-kicking in the market that Java (and its ilk) will be displaced as was COBOL before it. COBOL was "safe" and fast, but we finally reached a tipping point were it was too painfully obvious that it simply wasted too much time writing and reading it.
Enter the microcomputer in the mid 80s, with its 16 bit memory model and single wimpy core. (ignoring 8 bit toys). Running a garbage collector on such meager machines wasn't practical, so malloc/new and free/destroy it is, then! (yes, the irony is precious when Java appears a decade later)
After the early 90s, the Simula programming model has been pretty much the only thing taught to developers - a few self-taught people perhaps aside - so it's the only way they think.
Trying to convince these people that large chunks of their logic should be immutable data, processed in a "one way path" (directed acyclic graph) through functions, some of which are built of or build other functions, and that input and output types (which can be traced backwards and often inferred rather easily) can be assembled "on the fly" makes no sense to them.
It's almost like standing on a box on the street corner and trying to preach at people in Greek. They don't understand a word you are saying. "Get these freaky cultists out of here!" :-)
By the way, there is nothing like Scheme + Emacs for interactive development. You can write functions, execute a buffer, and perform your data manipulations in real time, patching your functions as you go.
It's freaking magic.
Pros:
+ Fast startup speed
+ I can use Python (and its vast ecosystem)
+ Qt has good ready-made components for eg. displaying a list of files (which is exactly what I need)
Cons:
- Compiling Qt (which I have to do for various reasons) takes time and is a pain
- Bundling and deploying a Qt-based app (times 3 - OS X, Windows, Linux) is a pain
- I have to deal with auto-updating myself (which I believe Electron supports out of the box) - times 3.
- I would prefer to lay out the UI in well-understood HTML/CSS. Qt's components and its CSS-equivalent behave in very weird ways sometimes.
I still sometimes wish I could use Electron but as OP says, we're supposed to pick the right tools.
I explain my reasoning for picking PyQt over Electron (and other alternatives) in a blog post [2].
2: https://fman.io/blog/picking-technologies-for-a-desktop-app-...
For the very reason you outlined - startup times. The average use-cases of file managers are being opened and closed very often. Electron is rather aimed at long-running applications you don't usually restart, like Atom and VSCode.
Microsoft could've ported VS to other platforms, yet they chose Electron. I'm sure they spend quite a while analyzing the requirements and looking at alternatives as well before settling down.
The end result is very lightweight compared to VS and does almost everything users require. VS still starts a lot slower than VSCode for one.
Aren't they doing exactly this? [1]
That would explain why this is the first time I hear about it :)
It's not really a port though, just another piece of software labelled "Visual Studio something-or-other".
The main benefit of Electron is the community. Everybody knows HTML/CSS/JavaScript, so can write extensions / customisations. If I were Microsoft, that's how I would have justified Electron.
Still, I personally much prefer Sublime Text :-D
Emacs also has a client/server architecture so I can spawn the emacs daemon and then connect to it from anywhere.
We have a few people in VSCode at work, a few in Sublime and yet more in Atom. They're all very happy with their setup :)
My everyday work is in Java, JavaScript everyday, I wouldn't use Java in my free time for sure. But I do like JavaScript now with Babel/TypeScript. I did in the past worked with wxWidget and pyGTK but I found that building UI with React is more convenient and flexible.
UI performance are on par for this type of application, plus fetching file information from node is also pretty fast. But if I ever encounter performance issue, it's quite easy to redo the bottleneck in c/c++ with node. And pretty soon in the browser with webassembly.
Memory consumption is less than my Finder process (with all the opened helpers). But if I compare application like VS, VSCode and Intellij. Electron memory usage should be in between a C++ and Java apps.
The challenge would be to make it start/open really fast (now for the prototype it's like 1 second). But I guess it's possible to do the same thing as Nautilus/Windows File Manager/MacOS Finder and have a background process running.
Before the line between language and framework were more sharp. But I think with the evolution of the language and framework like electron, the line is getting a bit more blurry. Great apps could be written in electron as they can be written in a native framework. End users will be the real judges.
All of this takes time. If you want to launch an app on multiple platforms, Electron is a very time efficient way to do this. Obviously native apps for every platform would be ideal but not everyone has the resources for that.
> Let me be clear about this: JavaScript sucks. It’s not the worst, but it’s also not by any means good. ES6 is a really great step forward and I’m thrilled about how much easier it’s going to be to write JavaScript, but it’s still JavaScript underneath the syntactic sugar.
Use something like Typescript or Flow then. The above is like saying C++ sucks because assembly is a bad language to code in.
If writing software is about making users' lives easier, happier, or better, though, I think we ought to optimize for that and that alone.
I say let the market and communities of users decide which software they want to use. I use Nylas b/c I like the UX better than Ubuntu's Evolution, despite its flaws. I use Caret (electron markdown editor) because I wanted a cross-platform tool and I have > 4 gigs of ram. Hell, millions still use the Facebook Android app, despite the fact that it sucks down battery juice like your crazy uncle sucks down Miller Genuine Drafts before lecturing you on how terrible millennials are (happy Thanksgiving, btw).
Of course, it's important we know the trade-offs of the decisions we make, so we need critical pieces like this. But to be as prescriptive as this article is to miss the forest for the trees. Love it or not, JS is a point of entry for a hell of a lot more ppl than C++ is, so expanding its reach into new arenas is not only net-good, but I'd argue also inevitable.
That said, the article does skirt an issue I've thought about a little, and I'm interested to hear others' takes. If software collectively moves more towards higher-end systems, e.g., with SPAs and electron apps and all their cpu-intensive renders, are we further limiting a great web UX to those with the money to afford the latest laptops? We spend a lot of time thinking about accessibility standards for our products (A11Y, etc.), but when should we begin to think about accessibility from a socioeconomic perspective? Or are performant computers now so accessible that we needn't worry?
That's (imo) your flawed assumption. And no, just choosing electron won't make your users happier.
> Love it or not, JS is a point of entry for a hell of a lot more ppl than C++ is
C++ is not the only alternative. And considering the staggering amount of frameworks and tools one must learn to get started with JS other alternatives are actually easier. Unfortunately, not as cool.
> That said, the article does skirt an issue I've thought about a little, and I'm interested to hear others' takes.
Is this a joke? I had to google what skirt means: "attempt to ignore; avoid dealing with." That's the exact opposite of what the article does. Let me quote it for you:
About electron: "One of the slowest, least memory efficient [...]"
About terminal: "They’ve taken one of the most lightweight and essential tools on your computer and bloated it by orders of magnitude. "
About lossless-cut: "For the price of 200 extra MiB of disk space and an entire Chromium process in RAM and on your CPU [..]"
The conclusion that "people will require more money to buy more hardware to run bloated software" is not something only you thought about, it's only one step away from what the article says. And it's actually hilarious (read: infuriating) that you don't see the problem with that way of thinking. NASA put people on the moon with computers less powerful than a mobile phone.
The article criticizes people who write bloated software, but you argue it's fine, because users are happy. Then you go on wondering if it's a problem that users won't afford the latest laptops to run bloated software. WAT?
Ah thanks, I lost some clarity in a post-write edit. My point wasn't that choosing electron would do that, it was that we should look more at end-user satisfaction and less at often pedantic technical analyses of solutions.
Re: C++: You may be right, but we can't deny that popularity and cool factor has a significant effect on hard metrics--eg, more frameworks, more libs, more components == faster dev speed & iterations, which I think is a good thing. ex: Mac hasn't been "better for graphics" in over a decade, but the 'cool factor' has made it the de facto choice for designers, which has resulted in a bulk of the latest designer tools being Mac only (a problem which, coincidentally, electron & its ilk aim to solve).
Last: no, it wasn't a joke. I know what the article said. My point was that the point of socioeconomic accessibility was mentioned, as you note, in the last line, but for the most part, the supporting data were more asserted as intrinsically bad. If it was a key argument of the author's, I think the data would've been tied coherently to the conclusion instead of requiring a diligent reader such as yourself to farm & cobble them together into a patchwork argument.
So, if you'll forgive me for borrowing a little of your snark: next time you google a word, for instance, 'skirt', you should probably prefer the first definition ("go around, past the edge of") instead of the fourth definition, as you've done here.
To be clear, my thesis is, in tl;dr form: "Anything that makes reaching users in the environments they prefer easier is probably good, but the extent to which this trades off with accessibility is unclear to me."
That was the third definition. The first definition was "a woman's outer garment fastened around the waist and hanging down around the legs." /snark
> we should look more at end-user satisfaction and less at often pedantic technical analyses of solutions.
They aren't orthogonal. User satisfaction comes from a smooth experience. Regardless of how shiny something is, if it slows down the rest of the system (or is itself laggy) users will notice. I choose to run foobar instead of other music players because it's small and fast.
> are we further limiting a great web UX to those with the money to afford the latest laptops?
This is what infuriates me. People have been talking about how the "free lunch" is over [1] since 2009. About how CPUs have stopped getting faster, and how we should make the best out of the current hardware by learning to paralelize. Yet javascript is the most popular language and it's fundamentally single threaded. High quality software is not about having "a great web UX". VLC doesn't have it (great web UX). Blender doesn't have it. Intellij IDEA doesn't have it. And they're not only extremely complex and efficient softwares, they also have great UIs. Our computers are extremly powerful. The fact that someone would wonder if they're not enough to run pretty clipboard managers and fancy text editors is making me cringe.
Heck, I've got Rust talking to node.js in it. Elm for the front-end and Rust for the processing, it's a surprisingly pleasant development experience.
Rust -> Node -> Rust is pretty simple, just expose a c api and wrap it with node-ffi. Then I have the main thread piping messages over to the relevant windows straight into Elm ports.
Nice bit is you can pass callbacks into Rust that calls js via c function pointers. Saves a ton of time marshaling data and is more efficient.
> We use [JS] because we have no choice (people who know more than just JavaScript know this). The object model is whack and the loose typing is whack and the DOM is super whack.
Moot point. I use React + TypeScript. Good types and barely ever work with the DOM. If you don't like TypeScript, your favorite programming language probably compiles to JS anyways.
> The reason people choose Electron is because they are too lazy to learn the right tools for the job. This is the worst quality a developer can have.
I agree that bundling a copy of chromium for a todo list app or clipboard manager seems wasteful. But that's an edge case, most people build actual applications. If I'm building a non trivial cross platform app, I'll be orders of magnitude faster on the web stack. I'm not lazy, I just don't want to waste months of my time learning arcane APIs because anything else wouldn't line up with the author's point of view about what the "right" tool is.
Why is this garbage on the front page? Stop giving condescending people like this a platform.
Cool. I remember when Handbrake first came out, and 'make video go on iPod' went from some long winder research into ffmpeg flags to clicking a button that says 'output for iPod' [1]. This sounds similarly timesaving.
[1]. yes, iPod. Showing my age.
That was, what, 9 years ago? I realize it's an eternity in the average JS framework lifecycle timeframe, but come on.
I've had to work on PyQt5 at work, and was pleasantly surprised out how easy it was to hack up a full app.
I was using PyQt, and I found it nice (Qt's integration is good).
It goes out of its way to not rely on the STL, so that std::string and QString are separate, non-convertible classes. QList and std::list<> are the same, same for QMap and so on and so forth. So data communication is already a pain.
We then have further issues with function invocation. Slots and sockets are simple with python, because functions are first-class in python. In C++, they are very much not, so slot-socket binding looks like: connect(sender, SIGNAL (valueChanged(QString,QString)), receiver, SLOT (updateValue(QString)) );
This is less than ideal.
STL wasn't always here and still isn't in some embedded platforms.
TTK. That is all.
But yeah, there's no one clear winner, which means that Electron is not a uniquely bad choice. Sure, it might be a little heavyweight for a really small simple app, and if you're not aiming for something cross platform your native widget toolkit might be a better idea.
Sure it was easier to build, but your users suffer for it. Electron apps are pretty terrible performance wise, and you can immediately tell it's an Electron app.
- Atom
- Slack
- GitKraken
All of these apps have 'tells'.
Comparing Electron to AIR makes no sense, they're radically different runtimes with radically different ecosystems. AIR stands alone while Electron reuses the web stack already used by millions of developers.
I find it rather elitist to dismiss Electron apps as "easier" to build or having users suffer. Thats completely false and grossly overlooks the context in which these apps were built. This is the kind of comment you'd expect from a beginner without much experience wanting everything to be in his favourite language.
I am currently trying to make a GUI to compose and render Terraform assets in a similar manner to AWS CloudFormation's visual designer, and I've spent an eternity trying to devise the best way to do this (this is mainly driven out of boredom more than anything else...Terraform works just fine as is). As much as I don't want to use Javascript and the cancer that is it's dependency management, the web stack is proving to be the path of least resistance thus far. Although I still don't understand why you need to download half of the known internet to install gulp and babel locally.
I'm a GUI novice. Is there any cross-platform software that generates a true native UI on target platforms?
wxWidgets does
Electron tries to do just that on the desktop. It packages all the dependencies that the application needs together with the application and installs them in an isolated app folder (not affecting any other app). It enables easy automatic updates of the whole environment without worrying that the update will break other apps or will not work because some user happens not to have some system library.
Sure, this is much earlier stage than with backend services. Ideally Electron could ensure that identical dependencies that are shared between different apps are not installed multiple times, the way Docker does it. Even without such sharing I think that benefits of the Electron approach justify additional MBs on disk.
I haven't seen many examples of people using Electron to make applications that really should be web apps instead, but more for packaging web apps with native enhancements. Often I'd prefer a "proper" native app, pushing even more features in the browser IHMO is not the answer.
Ctrl+S -> Save whole webpage
alternatively get the Webpages source code.
There is. The major OSs provide APIs to access the platform browser. The downside is you need a platform specific shim to set this up using native code. So, developers need to step outside of their JS worldview for a few days to avoid that 200mb app size penalty.
If you are building a platform agnostic app then I understand the value of Electron to avoid that work across each platform. But, it seems several of those Electron apps aren't even platform agnostic, so why not build the platform specific shim and be a better citizen on my hardware.
edit: I'm daft
Now that internet is everywhere, couldn't Electron:
1) Setup a common bootstrap mechanism so it's version agnostic 2) Check for the declared version of the app 3) If online (and allowed by system policy) check if there are security issues, or updates for the needed version 4) Advise user where to download the update, or (again, if allowed by policy) allow the user the proceed with the installation?
With some magic to allow a 'side-by-side' distribution of the Electron runtime, on good-ol' durable media, wouldn't this go far to minimize the size of the distributions of these Electron apps? You could have a system with one (or more if nessecary) versions of the runtime, and multiple apps using that runtime (with isolated profiles, of course ...)
Oooh, we could give this core a snazzy name like "Flash" or "Silverlight" or something too.
Once again, history repeats itself.
But what are the alternatives? Getting started with Qt/GTK isn't nearly as easy, for example.
Maybe someone should write an intro to native GUI programming for JS devs.
Unfortunately, most users know Java from years ago and never understood the difference between an applet or desktop application, and in the minds of some users it's still falsely stereotyped as slow or insecure.
In most shops I know, Java is being extensively used on the back-end, but almost no custom Java apps with GUIs. Most shops (and people) who could do Java on the frontend have gone to learn Mono or WinForms or WPF instead, which gives you results that feel nicer to most users.
import sys
from PyQt4 import QtGui
def window():
app = QtGui.QApplication(sys.argv)
w = QtGui.QWidget()
b = QtGui.QLabel(w)
b.setText("Hello World!")
w.setGeometry(100,100,200,50)
b.move(50,20)
w.setWindowTitle(“PyQt”)
w.show()
sys.exit(app.exec_())
if __name__ == '__main__':
window()
b = QtGui.QLabel(w)
b.setText("Hello World!")
b = QtGui.QLabel(w, text = "Hello World!")I ran into this when teaching myself Swift last year. Learning a new language is easy. Learning the ecosystem, architecture, and some initial best practices is a sisyphean task.
I am on win10, on pretty hefty machines. No idea if it's because Electron or something else, but it's really annoying. I could not find any trace of issues in the event viewer etc.
Time is finite, and not all of it should be spent programming.
Either way, Electron sucks because it ships a whole goddamn browser to give you just portability; you have to still pile on a ton of JS and CSS hacks to get a decent-looking SPA UI out of it, which you'd get for free with any proper GUI runtime. All that waste because people want to write JavaScript (which I understand - I have my favourite language I'd love to write more too) and this was the most brute-force way to allow it.
Embed a system webview if you must, don't ship the goddamn browser!
There are perfectly valid use cases for Electron, and they have absolutely nothing to do with what you said. There are also insane amounts of cross-platform native code written on platform abstractions layers.
> Developers are so into code sharing, write once run everywhere, that they don't stop and ask if they should.
Isn't that the very purpose of Qt or GTK which the article is promoting instead of Electron?
If the trival app is too big for a trival app - who cares. The fact is there is now a serious contender in the cross-platform space that so far has been mainly occupied by Java. Java cross-platform is neither easy nor lightweight. And from what I have seen so far getting started with Electron looks a lot easier.
I also can't help but feel as if the author is a beginner developer lacking a huge amount of context and experience. Saying "Electron enables lazy developers to write garbage" displays a huge misconception about software development in the very first place - garbage will be written no matter the language or platform.
Alls the article did was remind me that those who can, do, and those who can't, blog.
Native is only an option if you want to rebuild your app UI from scratch a minimum of twice and never support Linux or other less common desktops.
UI development has always been in fragmentation hell.
I've seen native app with more ugly UIs and being laggy, etc... For many apps Electron is fine and will perform as well, or even better than native apps (just because, developers you know).
I really hope Rust or Go get great native GUI toolkits soon with nice declarative layout languages, but until that happens it is hard to get too angry at people using the wrong tool for the job because the right tools just aren't that great.
The idea that I might choose electron for ease/speed of development rather than use a leaner platform is being shit all over as a "business" decision. Not a "I don't have enough free time to learn two new languages and ui systems code three new apps" decision.
The JS performance issues should go away for the most part when wasm is standard. Also a lot of people write some compile to JS language, not JS, so I think that most of the points of this blog will be rendered moot in the near future.
Or is the cost of the license really that prohibitive for a well supported and performant piece of software?
[EDIT]: I found the pricing after a bit of creative googling: Startups is about $1,000 a year, non-startups are about $3,500 a year. Even without the discount, that's roughly equivalent to a month of one developer's time. IMO, not a terrible deal, for something you can freely license and use for cross platform development.
(not legal advice; I'm not completely sure about this)
> The LGPL allows you to keep the source code of your application private as long as it is “work that uses” the library. Dynamic linking is usually recommended here.
Using them could be a good solution for targeting OSX and Windows and reusing logic from your web React app without having to deal with the overhead of Electron.
As detaro says, it's for web apps, not desktop ones. Afaik its mostly boilerplate for node+react.
It's too easy to lambast Electron and "lazy developers" when you've never had business concerns trump technical ones because none of your work ever tried to run that gauntlet.
I had to chuckle to myself when I saw that your GitHub bio is:
> I make unoriginal projects with better code than
> the thing they rip off. I'm available for hire.To the parent: Electron is a framework for building cross-platform desktop apps with HTML CSS and JavaScript in an embedded chrome browsers/node.js environment.
More info here: http://electron.atom.io
Maybe he wants an answer that is curated and summarized by a human, not a litany of links where you don't know what to focus on. For example, when I asked what dependency injection is, some kind stranger responded "It's a million dollar term for a 5 cent concept. Pass objects as arguments to functions instead of instantiating them inside." Of course, this is a limited definition, but it was put into layman's terms and it answered my question. From that answer, I was able to really learn what it was instead of being stuck on phase 1 of understanding.
When you're online, it's much easier to come off as a dick than it is in person so what I try to do is imagine someone coming up to me earnestly and genuinely asking the same question and respond thusly. Accordingly, when I know I'm asking a simple question, I add self-deprecation to the beginning. Example: "I have the technical and programming acumen of a bag of rocks, so excuse my elementary question, but..."