It explains very well why is something being easy shouldn't be the main metric, when choosing some technology.
In this case you are advocating to pull in webpack (28MB, 348 packages from 210 contributors), @babel/core (12MB, 54 packages from 60 contributors), typescript (53MB, 11 packages, 4 contributors), to just mention the most obvious and necessary dependencies, not even mentioning the different plugins for integrating them with each other, test frameworks, linters, IDEs (syntax highlighting and auto-formatting) and of course the mountains of type definition boiler-plate.
100MB+ of dependencies which are all a responsibility to learn, teach, configure and upgrade.
It's like 6 pages of code on a 2K iMac monitor and none of the above...
I'm not saying that there are circumstances when you might want to put up with all that cost, but have you consciously considered, whether it really worth the overhead? What's missing from hyperapp, which is only present in the 100MB+ deps? Maybe `htm` (1MB, 11 packages, 4 contributors) is a better compromise between bloat and ease of development?
And if you are willing to take on the extra dependencies, wouldn't ClojureScript provide a lot lot lot more benefits? It's also based on the super mature Google Closure Compiler, which supports extremely sophisticated optimization techniques and DCE... Just to mention some recent optimization surprise: https://twitter.com/roman01la/status/1277148232276234240
> [...] have you consciously considered, whether [it's] really worth the overhead?
Yes, I have. The overhead of a React/TS stack is considerably less than the overhead of having to hand-build all the stuff that Hyperapp's 12 kilobytes aren't enough to provide, or that using ClojureScript means very few other engineers will want to work with me to build because I'm doing everything in a language and stack they've never heard of.
> What's missing from hyperapp, which is only present in the 100MB+ deps?
Eleven years! Hyperapp is written in pure ES5, which was finalized in 2009, around the same time PHP 5.3 came out. The world has moved on a great deal since then. (But hey, I haven't had to think about hoisting for like six years now, so that was a fun blast from the past!)
That said, those 12 kilobytes of ES5 code are very concise, for sure. They also implement a virtual DOM and a sugar syntax over element creation and nothing else. To compare it based purely on codebase size, with a stack that provides orders of magnitude more functionality, seems at best very incompletely thought through, and at worst deliberately tendentious - although I'm sure that can't be the case.
There's a lot of misinformation floating around about the modern Javascript world. A lot of that dates from years back, like the left-pad problem, or that builds aren't reproducible - both of which were true at one time, but have long since ceased to be. Or that using these tools makes it impossible to deliver compact, performant code to the frontend - which, again, hasn't been true in quite a while. It's not false to say that webpage bloat is worse than it ever has been, but it is false to blame the tools for that problem.
Of course, I freely grant that it's hard to see these kinds of nuances if you don't have much or any context on what modern Javascript is actually like. What I don't grant is that that ignorance is any kind of worthwhile place from which to pass sweeping comments on the whole field of endeavor.
If you are doing back compat for old browsers, Babel is necessary anyway.
Typescript is a choice independent from anything else.
There are alternatives to Webpack, none of which have gained traction, possibly because developers appreciate other features that webpack provides out of the box.
But, you are complaining about the developer tools, why? The source code size of emacs/vscode/vi is just as relevant, or the size of the linter, or the complexity of whatever build environment is used to deploy code to production.
All of those require upgrading and maintenance, and occasionally have bugs!
> But, you are complaining about the developer tools, why? The source code size of emacs/vscode/vi is just as relevant
I think you have a fair point that if it's not making the build artefacts larger in size it shouldn't be criticized for that.
Where the problem arrives is really about cognitive load: If my tools are emacs/vscode/vi or even visual studio or webstorm, I can use these tools, and come back to them in 18 months for a different project and use the new versions without giving it more than a few seconds of thought.
With Webpack, I am strongly encouraged to use create-react-app to allow for sane creation of new react apps with React/Redux/SASS. But: it's April 2018 and create-react-app doesn't yet support TypeScript, so I have to find somebody's forked version of it that does, but then that forked version doesn't support SASS, so I have to stitch support for that together into my own fork, and then understand it well enough to debug that if it doesn't immediately work.
When I want to make a new app 18 months from then, there is a new version of Webpack, CreateReactApp, etc which finally adds full typescript support, but the way these tools are configured is completely different which itself requires a lot of time and effort to adjust for.
The truth is that the experience of using a developer tool like Webpack is categorically very different than choosing your Code editor.
So...don't use CRA? You don't have to, although it does make a lot of things easier, and it's kind of a surprise to see the claim that Webpack encourages its use. I've never run across anything like that in the docs, and I have found those docs to do a good job of covering how to set up and use Webpack, with CRA nowhere involved.
Or - a much easier tradeoff - just don't use Sass. I mean, Sass isn't without its merits, but if you're already compiling something else down to CSS, you may as well go full CSS-in-JS and get the benefit of a real programming language instead of whatever sharply limited DSL Sass has, that I've totally forgotten about because I haven't needed to touch it in years and I'm honestly a lot happier for that. Yeah, you'll have to look at and vet some new tools, and you'll have to get used to thinking a little differently about how you write stylesheets. But on top of learning the ins and outs of an entire new tooling stack, how much extra cognitive effort is that, really?
There's always going to be tradeoffs. You make the ones that are right for you. And if you find yourself, while investing in a new paradigm, trying so hard to stick with an old familiar tool to the point where it's compromising your efforts to get productive with the new tools - that's a signal to start looking at whether it's time for the familiar old approach itself to be replaced with something that not only composes better with the new paradigm, but, as with CSS-in-JS versus older approaches like Sass or Less, might prove to be a lot better in its own right.
I mean, don't get me wrong, I see a lot of merit in sticking with proven, familiar technologies in which I'm quickly able to be productive. I don't expect I'll ever stop using Emacs, for example, and that's almost as old as I am. Platforms change considerably faster, though, especially in web dev. I started with Emacs eleven years ago, and at that time the frontend "framework" of choice was jQuery - even AngularJS was still a year in the offing - and the closest you could come to a standard backend stack was Rails. But you were more likely to be using something written in PHP, or a hoary old pile of Perl 5, or the like. We've come a very long way since then.
The pace of change has slowed considerably in recent years, I think, and I expect it will continue to do so. But it's not going to stop, and individual tools and stacks will continue to develop regardless. There's always going to be new features, new approaches, better ways of doing things, and some of them are going to be breaking changes. Keeping up with that kind of change, and knowing how to recognize and take advantage of the good, is one of the things it's an engineer's job to do.
Sure, I can get by fine without that, but then I've got to educate 2 other members of my team, both who can program in numerous languages but are quite inexperienced with web development, how to correctly set up each project using the React/Redux/Sass/Typescript/Jest. I can do it myself and hand them a project template, but then no-one else will learn how to do it. The reason for CRAs existence is (more than any argument I might make) an implicit acknowledgement that setting up new webapps is a far more painful experience than it should be.
> it's kind of a surprise to see the claim that Webpack encourages its use.
It was React that recommended CRA, not Webpack recommending CRA, apologies for my misleading description.
> Or - a much easier tradeoff - just don't use Sass.
I'm already quite familiar with plenty of reasons to not use -or to use- SASS. The point is that as a tech choice it's hardly a minor player, and yet the experience of setting up / migrating a project using easily recognisable tech (React/Redux/Sass/TypeScript) is a long way from a straightforward. In my example, I already had SASS on existing related projects. My team members were already familiar with it. Create-React-App was at the time was also actually already built to include support for SASS - the problem IIRC was that the fork of CRA that added Typescript support was forked from a CRA version before SASS was supported, so (at the time) it was a "one or the other" situation before I forked it and fixed it myself.
> Keeping up with that kind of change, and knowing how to recognize and take advantage of the good, is one of the things it's an engineer's job to do.
It's also part of our jobs to look at processes and procedures identify areas of inefficiency, and to ask questions about what a better way of doing things would look like.
Finally, the trajectory of this discussion is pretty absurd. I'm offering examples of what happens in web development to illustrate that there is a massive difference between treating developer tools like Webpack / CRA as though they're similar to text-editors that you don't have to invest a bunch of time to understand and keep up to date with (a point that if I'm understanding you right, you likely agree with). My comment above's central claim is: that while critiques on size of dependency chain of developer tool like CRA or WebPack might be misguided, they are hardly lightweight cognitively, and the choice to use these tools (or not use them) isn't something to just take lightly because we clarified that our deployed artefacts didn't balloon in size.
Okay, that's fair. I certainly can't disagree that setting up a new web app in the modern style, from scratch and without prior experience, is cognitively heavyweight, and that reducing that complexity wherever possible is valuable.
I would say that CRA's cognitive overhead, in a situation like that, isn't all that different from that of Rails, for example. They're both trying to do a very similar thing, albeit with very different approaches, and they're both very much of the convention-over-configuration school, which attempts to trade off not having to think very much at first about decisions you may not have enough information to make well, with making those decisions costlier later on once you do have enough knowledge to approach them. I have reservations about that entire school of thought, but despite them, my experiences with Rails and with CRA lead me to think that CRA does a generally much better job with those tradeoffs than Rails ever managed.
The _only_ reason I use webpack and friends are for JSX and Typescript. For me, those two tools are worth the compilation step. Qs soon as it's possible to stop using compilers while still enjoying the benefits of both, I'll dump them from all my codebases.
If you have to do that anyway, I think I could use JSX instead of "pure" JS.