JavaScript is weird
jsisweird.com
jsisweird.com
It's surprising in almost every language that does it to any degree. JS's is only slightly stranger than, say, PHP. About the only exception is "empty values are falsy", and even that surprises people pretty frequently, and can be used to construct highly confusing snippets of behavior.
Sometimes it seems javascript has a mind of its own, and turns all my numbers into strings when I'm not looking, so I have to sprinkle x = x | 0 throughout the code, like hanging garlic.
It seems to me like those two things are orthogonal. (Not that JS wouldn't necessarily be nicer with a separate concatenation operator.)
const y = +x + 12;
Will be number addition even if x is a string.
const y = Number(x) + 12
1 + "five" is an error in PHP, not some random correct-looking value.
On older ones though, it's not an error: https://onlinephp.io?s=s7EvyCjg5UpNzshXMFTQVlBKyyxLVbIGAA%2C...
Even in the previous major version (7) it's only a warning.
I agree that philosophy is likely the main cause of decisions like this, but JS is hardly unique in having it.
Meaning of course "no errors reported, at any cost".
Programming language language has redefined "error" - to substantial cost to programmers.
It’s a philosophy that might be acceptable for a simple-ish frontend, but when if you try and build something significant it’s just awful.
Source: I inherited responsibility for a > 5m LOC system written in JavaScript. The problems aren’t all JavaScript’s fault, but it does play a major part in it.
+1
A case I'm struggling with recently is Fastmail mail search. Given a valid query which it cannot handle, rather than show an error report, it does its "best" and delivers a false result. With potentially great cost to the unwitting user. Is there even a warning on the doc page?? No.
Forget JavaScript, that's not even an error in Java.
But anyway, it's much less bad in Java because you can force it to err by declaring what type of thing you want it to return.
true + ("true" - 0) ?
"" && -0 ?
Like, can you tell me a language where it does make sense to "apply AND to an empty string and a (negative?) zero"?
If you write code like this, you have much bigger issues than whatever you choose to write your code in.
My code has bugs occasionally, because I'm a human. Which is why I would like my programming language to return errors when I make obvious mistakes.
> Like, can you tell me a language where it does make sense to "apply AND to an empty string and a (negative?) zero"?
No I cannot, which is why I can't imagine why JavaScript wouldn't return an error for this situation.
How does one enforce a strict subset with an easy setup? Why are people writing non-browser code in this horrible language, often without strict subsets or typescript?
Obviously there are solutions, but the solutions clearly aren't pervasive enough because the problem is still here (and growing, as more JS code is written), so we as an industry still need to keep talking about how problematic and poorly-designed JS is.
The old solution are the good books like the good part. The new solution would be typescript.
It's unlikely the API will get deprecated no matter how you talk about it. It's the windows of internet. Any sane shops would use typescript. U don't even need to write ts, and could just let the compiler do the hinting for you.
Deprecation isn't what I want. Equivalent alternatives are what I want.
I still can't do <script type='text/typescript'>. Why?
And look, Typescript is also hampered by some of JavaScripts bad choices. Actor-model concurrency would be a really good way to handle events and parallelism to boot, but we won't get that in any language that compiles to JS.
WASM isn't an alternative, because you can't use WASM without JS.
Because typescript is not javascript. Google already tried with dart, they failed. Typescript is still nice-to-have, it's not replacing javascript. MS is not trying to replace javascript anyway. If your goal is to having javascript replaced, that looks fine but I don't see how it succeed. If you u want to work on TS codebase in backend, there is deno
> Because typescript is not javascript. [...] Typescript is still nice-to-have, it's not replacing javascript.
+1. TypeScript compiles to JavaScript, so it can't replace something it compiles to. Besides, text/typescript isn't even a valid MIME type. The text type is meant for plain-text documents (.txt, .md, etc.), so TypeScript would have been in the application type, following JavaScript (application/javascript). Getting TypeScript in a web browser would be too much work. For starters, they would need to take the TypeScript compiler, slap it into a web browser, where it would compile the TypeScript code, pass the result onto the JavaScript compiler, which would finally pass the result to the browser. Whats wrong with this? The code has to undergo two compilation layers, and, most importantly, the TypeScript compiler in the web browser can be out-of-date with mainline TypeScript releases. Your best bet right now is using a web framework with TypeScript support.
it actually aids in writing faster (more JIT-friendly) code by alerting you to these implicit type mutations/coercions.
The point is that in lots of languages these statements would simply result in compile time errors. In JS, though, the type coercion rules are so loose (and in many cases extremely non-obvious) that you can see this head-scratching behavior.
It's the same underlying behavior that leads to very cool/very insane stuff like JSFuck: https://jsfuck.com/
This weak typing is unique to a few languages like JS and PHP. Python is just as dynamic, but will at least throw an error if you try to accidentally add a string and a number. Static languages catch this even earlier.
get_current_username: Returns a nonempty string iff user is logged in.
get_balance: Returns a user's "balance" of some sort of credit or currency. Negative balances are possible.
should_render_balance = bool(get_current_username() and get_balance())in js, Boolean('' && 0) will give you the same answer as bool('' and 0) in python.
if i code golf in python (like the article does in js), i'll also get weird-looking code and "unexpected" outputs.
valid python: 0*"string"or"other_string"
> Like, can you tell me a language where it does make sense to "apply AND to an empty string and a (negative?) zero"?
People do stuff like that in Python all the time. Personally, I don't do it in JS, though, because I find the conversion-to-bool rules much less predictable. In this case, though, I guess JS would behave the same.
typescript helps, but i have fired some of these footguns even with typescript.
This isn't just javascript though. It's harder to find a language for which this would be true than the other way around.
I'm not sure why it seems there hasn't been any progress made on making floating point types less error-prone in the last few decades.
Of course, after looking up what === is, the behavior is obvious.
I kinda expected text and code examples, on a blog. I forgot to check that this is a dedicated website about the weirdness of javascript.
var arr = [];
arr[2] = 2;
// now arr is [empty, empty, 2]
arr[-4] = -4;
// now arr is [empty, empty, 2, '-4': -4]
arr.length; // is 3
arr["foo"] = "bar";
// [empty, empty, 2, '-4': -4, foo: 'bar']
I'm not sure because I'm not at home, but did you verify this pseudocode? Pseudocode doesn't have to be valid syntax, that's the point pf course. But it should bring the point across regarding the behavior of actual programs, no?
AFAIK, unquoted colons inside of array literals in JS are illegal.
I'd understand it better if you had written sth like
{"0": undefined, "1": undefined, "2": 2, "-4": -4}
Array.prototype
- REPL context wishing to redeclare identifiers other than functions defined using the function keyword - very legacy browser Targets, < latest IE11
Sparse arrays are, just like var, a historic relic.
I have never encountered any code where sparse arrays were intentionally used or appeared at all.
Arrays are objects in JS, yes. You can construct a sparse array using
Array.from({length: 5})
And JS is honest that it's an interpreted language where "arrays" are not even required to have a uniform member data type - which of course is very advisable for reasons that go beyond code readability.
But let's be honest, is (i.e.) PHP more clear in this regard?
Personally I hate PHP's mishmash of arrays and maps ways more than the JS version. Both are syntactical sugar for memory-managed and heavily optimized "map" like objects where the keys are restricted to primitive types.
Personally, I like that JS at least has a usable notion of strict equality and makes it easy to understand when values are references (always, unless they are primitive).
PHP's copy-on-write behavior, reference operators and the mishmash of "associative" and "indexed" arrays are much harder to actually understand and bigger footguns IMO.
JS has massive quirks, but what you show is not surprising for anyone doing JS with a modicum of proficiency.
JS is a scripting language.
People tend to forget this because of fast JIT compilers and because of TypeScript, I guess.
Sparse arrays are a relic and the differences between your "empty" and the actual primitive value "undefined" in JS are not of any practical significance in my opinion.
It's really a wart, but not one of practical relevance.
At least this differentiates this from "problems" like
["1","2","3"].map(parseInt)
This isn't pseudo-code. It's what your browser will output as the value for arr
What will seem weird and unexpected in this is that you're essentially dealing with dictionaries rather than arrays. They just have a length attribute bolted on that counts the highest key number as a length.
So tired of those smug "hurr durr language X hehe" posters
parseInt(0.00005)
// 0
parseInt(0.000005)
// 0
parseInt(0.0000005)
// 5
parseInt(0.00000005)
// 5
0.00000005 -> "5e-8" -> 5
> The parseInt() function parses a string argument and returns an integer of the specified radix (the base in mathematical numeral systems).
Surprisingly, if you put garbage in, you get garbage out.
Why would you parse a float? Parsing is for strings. Converting a float to an int is done with stuff like `Math.floor`.
You wouldn't, which is why it should produce an error.
That would make it a different, entirely hypothetical, ill-defined language. A very clear strawman that isn't particularly worth fighting.
It would quite possibly be a better language, but it's not a thing you "just do" to solve footguns - it has pretty big effects on the language and ecosystem.
const arr = [1, 2, 3, 4, 5]
for (let i = arr.length - 1; 0 < i; i--) {
const j = Math.floor(Math.random() * (i + 1))
[arr[i], arr[j]] = [arr[j], arr[i]]
} Uncaught ReferenceError: can't access lexical declaration 'j' before initialization
While /, +, and - are rare in the beginning of the statement, ( and [ aren't.
I enjoy teaching this to junior developers as the "winky frown rule" because that is a silly name that sticks in your memory and does help you remember it. Lines that start with an emoticon frown must wink:
;( ;[ ;`['10', '10', '10', '10', '10'].map(parseInt)
["10","10","10","10","10"].map(x => parseInt(x))It’s a known good practice in JS to always wrap callbacks with a lambda in this situation. Even if you don’t, usually the type checking will catch it, this bug only works because the second argument is a number to both functions.
Defining functions with new names also make it easy to evolve an API over time without causing breakage.
Looking at, e.g. the Python standard library a large number of functions have arguments that were added after the function was first defined. If Python didn’t have optional/keyword arguments then the number of API calls would be substantially larger.
They do, but potentially falsely.
> Looking at, e.g. the Python standard library a large number of functions have arguments that were added after the function was first defined. If Python didn’t have optional/keyword arguments then the number of API calls would be substantially larger.
The inability of Haskell not to have optional arguments isn't too bad, but all in all, I have to concede that you have a fair point. I personally don't miss optional arguments but I can see how they're beneficial (and I would really like named arguments).
1. Functions can be called with the wrong number of arguments.
2. map() passes multiple arguments to the callback.
NaN === NaN -> false
In fact that’s exactly how isNaN(x) is implemented under the hood. It compares the argument to itself.
https://stackoverflow.com/questions/1565164/what-is-the-rati...
Because, conceptually, 0/0 is not equal to -inf + inf. NaN is not a semantic value (it is a lack of one), so you can’t do semantic comparisons with it.
The opposite is almost more surprising.
[0]: https://www.oreilly.com/library/view/javascript-the-good/978...
I'm working on a book called "How to not get your knickers in a twist because you neglected to learn from people who came before you."
My issue with JS (particularly but not exclusively server-side) is less that it has some weird aspects and more that its built-in functionality/standard library is incredibly barebones in many respects. IMO that’s the main reason why it’s considered normal in the JS world for a project to pull in hundreds/thousands of largely un-vettable dependencies.
Knowledge transfer, context switching, the stuff you already know. But there's also an interesting non-JS reason that's gotten forgotten because the serverside landscape has changed so much since the early 2010s. Node very heavily popularized single-threaded asynchronous request handling (ie, not starting up a thread for every request that came into the server).
Of course there is nothing about Javascript that means it specifically as a language is essential for that pattern, but I remember hearing talks from Netflix employees and industry advocates about how much Node had improved their performance; I remember industry people coming to my college and talking about single-threaded serverside code and saying "you all need to check Node out."
And there are downsides to asynchronous code (callback hell was the obvious downside that got brought up a lot before `async` or even promises had arrived in the language) -- but I remember among especially younger serverside developers this trend of people saying, "Node is just faster. That's why you use it, because it's just straight-up more performant than what you're currently doing." And I understand serverside devs from other languages can at this point jump in and say that it's not, that the single-threaded model is possible in other languages, that there are downsides to the single-threaded model, that Node doesn't do it well, whatever. I'm not arguing the point, I'm just describing the cultural shift I saw.
All of the conversations about sharing code happened too, but I remember having conversations where people talked to me about jumping to Node specifically because Express was single-threaded and because it could handle more concurrent requests than other popular server architectures that they were familiar with. And I suspect that it gave the ecosystem a lot of momentum.
Node's original purpose was in part an experiment to see if single-threaded async request handling would be able to handle large numbers of simultaneous requests. My somewhat-limited memory of the conversations that were taking place back then is that running the same language clientside and serverside was just the icing on the cake. The big reason people were excited was because Node was single-threaded.
When Node started to become a thing, Java and C# has already been doing multi-threaded web servers for about a decade.
Oh, and C# already had support for async programming too!
Node become popular in part because Silicon Valley just doesn't seem to be aware of anything that exists outside it.
Node was single-threaded async, not multi-threaded. The entire point was to not spin up multiple threads and to rely on the event loop instead.
> Oh, and C# already had support for async programming too! Node become popular in part because Silicon Valley just doesn't seem to be aware of anything that exists outside it.
> And I understand serverside devs from other languages can at this point jump in and say that it's not, that the single-threaded model is possible in other languages, that there are downsides to the single-threaded model, that Node doesn't do it well, whatever. I'm not arguing the point, I'm just describing the cultural shift I saw.
Right or wrong, you can go back and look at forum posts even only 6, 7 years ago and you'll see people asking about single-threaded performance and asking how Node manages to scale if it's single-threaded. A lot of developers at that time got introduced to single-threaded event-based asynchronous server programming through Node.
I'm not telling you that Node should have won, I'm telling you one of the reasons why it happened to win. Here's Dahl's original slides presenting Node in 2009: https://s3.amazonaws.com/four.livejournal/20091117/jsconf.pd...
Immediately the first thing he's talking about is not unity between the front-end and backend language or sharing code. It's "I/O needs to be done differently." That's what people were talking about back then.
What's interesting about this talk is that Ryan doesn't even mention Javascript at all (other than briefly to say that NodeJS is built on it) until nearly 13 minutes into the presentation. And it's clear that he's using Javascript as a means to an end -- he's going with Javascript largely just because Javascript by default has a single threaded event loop that every JS programmer had to learn to work with it.
So he brings up going with JS because JS programmers are used to callbacks already -- and then roughly 2 minutes later he immediately moves back to talking about I/O and request handling again. Very little of his presentation is about Javascript beyond an aside of "JS already does this and JS programmers are used to it, so we might as well go with that and try to copy JS conventions." He almost spends more time talking about common SQL libraries than he spends talking about JS.
I agree using plain JS anywhere is not a good idea, but Typescript (or js with ts annotations) is often ignored when it does a very good job of catching most of these issues, while offering a very flexible type system in a more mainstream paradigm
- Error handling is onerous; I write a lot of functions where most of the lines are just checking for and returning `err`
- No anonymous structs, and it’s really awkward to define structs inline
- Dealing with collections is really verbose; I guess since generics dropped I could handle this in userland, but it’s not idiomatic
etc etc. No language is perfect!
I make heavy use of map[string]any vs structs. I also find the right balance of checking err vs just making it _
JS has also had a lot of its "mistakes" fixed since 1995. It does so in a backwards compatible way and it isn't always obvious how much progress has been made, but it has never been a static unchanging, "unfixed" language.
(Personally, I'd much rather work with JS on the backend than golang. Golang looks to me like a throwback to the 1970s that misses good language design ideas from the 1990s. I understand some of why it has become popular, but there's so much bad design in that language that upsets my nausea.)
You choose against a language because of code that any sane person, which I hope includes yourself, would never write intentionally?
Weird flex, but ok.
Reusing code, tooling, training is meaningful.
---
Every language (Java, Ruby, Go, Bash, whatever) has quirks. JS may have more than others, but it's not alone.
I'll probably get flamed for saying this, but having all objects be hash maps that can be extended or edited at any point and be simultaneously dumpable and loadable into JSON with one line was an unfathomably big brained move. Being able to just use the data you get from various sources instead of having to create an ad-hoc struct/object that perfectly matches all types and values (and ofc crashes immediately if a change is made on the other end) is so underrated. That's not even mentioning the native async support.