Look, I'm generally with you, there isn't anything particularly bad about modern JavaSscript, and it's a perfectly fine language to write your tools in if that's what you like.

But your points are... Weird.

1. Familiar to people who know JavaScript, sure. But shell is the language you interact with the CLI with, so more people who use the CLI are familiar with a shell.

2. Most languages have a standard library.

3. I know many people who would disagree with you. Anyway, pip works fine in a Unix environment, which is what we're discussing here.

4. We're talking about glue code, right? Glue doesn't tend to be very parallel.

5. So, perfect Unix glue? Sure, you like it, and that's great for you! Keep on doing you. It's obviously capable of making nice tools, but so are most languages...

6. You luks-suspend example is weird because the Node.js code obviously does so much less than the other versions. I have no idea what the Go version is doing, but the shell version does things like remount the filesystem, handle udev, etc... A bash version of the JavaScript code would be shorter!

7. 0.1+0.2!=0.3 is a result of floating point arithmetic being floating point arithmetic. It's not specific to JavaScript.

1. Most people who interact with a shell know how to make it run one command with arguments; some eventually learn about IO redirection. Many such people who use Bash as a "command prompt" would find the way it does even simple things like variables and conditionals exceedingly arcane. Readability is important, and, for better or worse, a compact curly bracket syntax is seen as "the norm".

2. Different languages' standard libraries have different scopes. Node's is just right for the task at hand.

3. I know many people who disagree with me, too. That doesn't make me wrong (or right ;-)).

4. Why not? Doing a task and displaying a rich progress indicator is doing two tasks. Glue is a continuum - for me, the concept includes orchestration of external "heavy lifting" tasks according to the desired logic, as well as processing intermediary outputs in simple ways so the whole pipeline fits together.

5. Well, I've evaluated it and found it to my liking. I guess you grow to like the tools you use to accomplish things, and if everyone who went out of their way to publicly dismiss Node.js instead went out of their way to use it in place of Bash scripting, they might find it to their liking as well.

6. Yet ironically, it has been the most reliable in achieving its purpose of locking my encrypted partitions when my computer goes to sleep, not to mention unlocking them when I wake it up.

7. Exactly - but people keep pulling this and other quirks out of their ass as "proof" JavaScript is some horrible language. Show me a shell scripting language that even has floating point arithmetic! But maybe glue code can't ever need that, either?

> Readability is important, and, for better or worse, a compact curly bracket syntax is seen as "the norm".

Python seems pretty popular.

> I know many people who disagree with me, too. That doesn't make me wrong

The problem isn't that you're saying node is well suited. The problem was your suggestion that node was uniquely well suited when actually the CLI is a solved problem in most languages.

> Well, I've evaluated it and found it to my liking.

That's all that really matters when it comes to your own personal usage. Just don't expect everyone to have those same preferences nor argue with other developers that your preferences are any kind of "fact".

> I guess you grow to like the tools you use to accomplish things, and if everyone who went out of their way to publicly dismiss Node.js instead went out of their way to use it in place of Bash scripting, they might find it to their liking as well.

Personally I don't like node and find Bash easier however it's thankfully not an "either/or" argument where your only options are Javascript or Bash. There's a whole plethora of options out there.

> Yet ironically, it has been the most reliable in achieving its purpose of locking my encrypted partitions when my computer goes to sleep, not to mention unlocking them when I wake it up.

That's a pretty meaningless anecdote because it could be the result of a whole plethora of reasons outside of the language's control (including but not limited to user error) or even completely made up (like your point about floating point arithmetic (see below) or your other comment about one liners being more readable than multi-lined code).

> Show me a shell scripting language that even has floating point arithmetic!

Quite a few do: ksh93, yash, zsh and murex (disclaimer: I wrote that shell) are all interactive shells which support floating point arithmetic.

In terms of shell scripting languages, you can add the following to the above group: awk, bc, dc and perl.

Even those shells that don't support floating point arithmetic can still do pseudo floating point maths by using powers of 10 multipliers and printf to place the decimal point:

    printf %.10f\\n "$((1000000000 * 365/7))e-9"

Ok, I'll grant you that is rather ugly. Thankfully you can call other shell scripting languages from within your POSIX/C shell:

    bc <<< "scale=10; 365/7"

This isn't a hard thing to research either; which suggests to me you haven't done your homework before making your arguments.

> Python seems pretty popular.

One reason it's so popular: even though it skips the actual curly brackets (thus making itself more accessible), it doesn't stray too far from the general syntactic paradigm of "curly-bracket languages'" - even though I've also seen people being alienated by the idea of "significant whitespace", again for no other reason but that it "seems wrong" to them.

> The problem was your suggestion that node was uniquely well suited

No, just much better suited than people give it credit for. And this ecosystem brings new things to the table, too.

> when actually the CLI is a solved problem in most languages.

Being able to build a CLI at all? I guess, sure. Being able to quickly build an accessible CLI? Hardly so. Everything in the area of HCI is in flux, and gatekeeping the CLI domain for does more harm than good. Btw do you know anyone doing this sort of interface before Yeoman? https://github.com/synox/youtube-dl-interactive/raw/master/d.... before JS.

> There's a whole plethora of options out there.

Which is why you've got to take into account opportunity cost, choice paralysis, etc. All of the options out there are positively fascinating, but if I have the options to solve problems in multiple contexts with the same familiar tool, I tend to usually go for it.

> Quite a few do: ksh93, yash, zsh and murex (disclaimer: I wrote that shell) are all interactive shells which support floating point arithmetic.

None of which are POSIX, or common defaults?

> In terms of shell scripting languages, you can add the following to the above group: awk, bc, dc

None of which are shells - they're DSLs you can call into from the shell.

> perl

Again, not really a shell. Though I wouldn't be surprised if anyone's using it as one.

> bc <<< "scale=10; 365/7" > 365/7

Even though these tools have a good reason to be like this, having to _research_ how to _call out to an external executable_ to do something as trivial as floating point division (in the sense of "cmon man I can do this in my head, why does is it have to be so hard to tell a computer to do it") is hardly on anyone's list of priorities.

At this point, nearly anyone would rather reach for the scripting language they're already familiar with, rather than learn an unfamiliar one just because it has historically dominated the problem domain.

> one liners being more readable than multi-lined code

Admittedly, that was a pretty bullshit argument. Not any less nuanced than "[l]ine counts mean jack shit", either - I take it as a general rule that one would prefer to write less code.

And I'm not weighing any of the examples against "[c]ode golfing, minifified code, shell 1 liners, Perl, etc. I didn't really take into account the extra functionality of the Go version - it was just as unreliable as the Bash one so I quickly discarded it when I found something that worked.

There's a story behind this comparison. It was likely user error, in the sense I failed to prepare my initramfs correctly using mkinitcpio or dracut (incidentally, both of those being shell scripts - insufficiently documented and not very readable to my untrained eye, either) But the JavaScript version worked for me because it didn't even try to chroot into the initramfs - it just moves a tiny handful of required binaries into a ramdisk, and works from that.

Since that solution _works_, I am hard pressed not to commend its minimalism. Is it inconceivable that that the author was not more clever, but the tools that they used enabled them to come up with a simpler solution in the time that would've otherwise been spent googling `bc` or poring over the sources of initramfs builders?

> One reason it's so popular: even though it skips the actual curly brackets (thus making itself more accessible), it doesn't stray too far from the general syntactic paradigm of "curly-bracket languages'" - even though I've also seen people being alienated by the idea of "significant whitespace", again for no other reason but that it "seems wrong" to them.

All you're describing there is structured programming languages. Of course people like structured languages, that's why nearly every single programming language developed in the last 40 years (that isn't designed to interface directly with hardware) is structured. Even languages of entirely different paradigms such as Common Lisp, LOGO, Prolog, MATLAB and Forth all support structured control flows. Even BASIC - for all it's warts - supported structured logic (even if many implementations lacked proper functions).

So you're not making a convincing argument about curly braces now you've redefined the argument as structured control flows.

> No, just much better suited than people give it credit for. And this ecosystem brings new things to the table, too.

Maybe. My personal opinion is the unmanageable nest of unvetted dependencies makes it a worse solution than people give it credit for. As a pure dev tool for personal use, go for it. But I don't want to read 10k+ lines of code in the dependencies of a tool that's ostensibly just an interactive alias.

I get the whole "standing on the shoulders of giants" thing but it's gotten out of hand with node. Half the time even people writing code in node don't know what dependencies their dependencies are pulling in. All that trust placed in an ecosystem that is very easy to abuse (and has been abused too). So no thanks, I don't think that's the right tool for serious CLI work which often needs to be run on trusted machines.

> Being able to build a CLI at all? I guess, sure. Being able to quickly build an accessible CLI? Hardly so.

Actually yes. I've been writing software for 30 year in well over a dozen different languages. Much, not all but a reasonable amount of that, was writing software for servers.

> Everything in the area of HCI is in flux

Not in terms of the CLI. We're using the same standards that were defined in the 60s. If anything, the problem with the CLI is that it's engineered around design principles which are no longer relevant to modern terminals. For example:

  * we no longer need to inline formatting with escape sequences - we have better markups for that.
  * ASCII is no longer relevant in our international community with different writing languages and character sets.
  * Terminals are no longer teletypes - there's no reason why we shouldn't be able to inline multimedia content
  * Terminals are no longer teletypes - we shouldn't need to be defining TTYs/PTYs as an OS abstraction
  * Defining TTYs is OS specific and a fucking mess on even some of the better designed UNIXes
  * Capturing and sending signals can often be OS specific and a fucking mess to work with on even some of the better designed UNIXes
  * Not all OSs even support the same suite of signals
  * Signal hotkeys can't even be redefined - which means you can't always compensate for one OS missing a signal that another might have. The best you can do is enable or disable the capture of that hotkey on the TTY. Which again requires OS specific syscalls
  * POSIX pipes are just byte streams. Which makes working with structured data a fucking nightmare (hence why Powershell, jq and murex all exist)
  * stderr is a lousy way of capturing errors
  * in fact error handling on the whole is just terrible on POSIX shells (hence why murex exists)
  * POSIX shells don't support any kind of unit testing (again, hence why murex exists)
  * exit numbers are a half arsed solution
  * named pipes are a file system kludge
  * TTY modes can get broken if a fork()ed process changes the mode and crashes for forgets to reset it on exit. Meaning your shell then becomes non-functional without manually fork()ing `reset`
  * ANSI escape sequences are anything but standardised. With different terminal emulators often doing significantly different things

This is all just off the top of my head so I could likely think of more if I spent any time thinking about it. All of those problems are a result of the command line interface being largely the exact same design as it was in the 60s.

So no, the CLI definitely is not in a state of flux. The exact opposite of that in fact.

More likely is the Javascript libraries you're using for CLI work is in a constant state of flux. This was a problem I had when I used to write software in node and it's one of many personal reasons I decided against using it for anything I really cared about.

> Which is why you've got to take into account opportunity cost, choice paralysis, etc. All of the options out there are positively fascinating, but if I have the options to solve problems in multiple contexts with the same familiar tool, I tend to usually go for it.

That makes no sense. Last week you were saying it couldn't be done and now you're saying the problem is too much choice. Make your mind up.

> [ksh93, yash, zsh and murex] None of which are POSIX, or common defaults?

You didn't say "POSIX shells", you said "shell scripting languages"

Also node isn't a common default either.

> [Perl] Again, not really a shell. Though I wouldn't be surprised if anyone's using it as one.

Again you didn't say "POSIX shells", you said "shell scripting languages". You can very much use Perl as a shell scripting language (and that is in fact still a common use case for it).

> Even though these tools have a good reason to be like this, having to _research_ how to _call out to an external executable_ to do something as trivial as floating point division (in the sense of "cmon man I can do this in my head, why does is it have to be so hard to tell a computer to do it") is hardly on anyone's list of priorities.

Actually computers can't do proper floating point arithmetic (in the sense that humans do) either. That's why a lot of languages (and spreadsheets too) demonstrate seemingly broken logic when comparing floating point numbers.

> Admittedly, that was a pretty bullshit argument. Not any less nuanced than "[l]ine counts mean jack shit", either - I take it as a general rule that one would prefer to write less code.

Line counts literally do mean jack shit. There are plenty of examples of where more lines equal better code. As a developer you should already know this.

All of these features are things other languages have too (I personally disagree that npm's "download the world" model is better than Python's envs). A huge drawback Node has is that it's not installed by default on any Linux distribution I know of, so I'd take a Python script over a Node one any day. Compiled languages have an even better distribution story, and Rust's Cargo is nothing to laugh at.

I'm not sure what you hope your posted comparisons show, but the Go "monstrosity" seems to have many more features than the Node one.

Sure, the the node code uses fewer lines than the Go code. But what you’ve failed to notice is the Go code calls APIs directly rather than fork()ing and does so within its own code base (how many lines of code is in the countless imports of your node example?). Plus the Go code also has unit tests which the others don’t.

Line numbers are always a terrible metric for proving anything but in this case your especially abusing that data.

Well, compared to a shell script or a static binary...the Node example requires you to have Node (+ NPM) on your machine and, unless you wrap it all up for homebrew/apt/some other package repo. you're going to be using `npm install -g` to use it as opposed to shoving an executable into your PATH.

All of your stated benefits are nice for the maintainer but mean little to the end user. This gushing praise for Node reads like nobody has ever built elegant CLI apps until Ryan Dahl turned of age and put JS on a server.

> In absence of more accessible complexity measurements, line counts do matter somewhat. E.g. the briefer a program is, the more readable it is (last time I checked, brainpower and time is much more expensive than computing power and bandwidth.)

There are as many exceptions to that rule are there are examples when it’s true. Code golfing, minifified code, shell 1 liners, Perl, etc.

Line counts mean jack shit.

I would argue that even an esoteric, densely packed one-liner can still be more readable than the equivalent program spread out over a greater number of lines - simply because you can keep it all in sight at the same time and work through it without losing the bigger picture.

But of course, scary symbols is scary.

So you'd like to use APL, then? He didn't say symbols. Verbose code can be much easier, though stupidly verbose is stupid. Compromise is, as usual, the best solution.

Could you please stop creating accounts for every few comments you post? We ban accounts that do that. This is in the site guidelines: https://news.ycombinator.com/newsguidelines.html.

HN is a community. Users needn't use their real name, but should have some identity for others to relate to. Otherwise we may as well have no usernames and no community, and that would be a different kind of forum. https://hn.algolia.com/?sort=byDate&dateRange=all&type=comme...