Deno will stop using TypeScript
startfunction.com
startfunction.com
[1]: https://docs.google.com/document/d/1_WvwHl7BXUPmoiSeD8G83JmS...
First paragragh of that document:
> Update June 10 2020: I saw that this design doc was being discussed more widely. Most people don't have the context to understand this narrow technical document - it is only applicable to a very particular, very technical situation in the internals of Deno. This is not at all a reflection on the usefulness of TypeScript in general. It's not a discussion about any publicly visible interface in Deno. Deno, of course, will support TypeScript forever. A website or server written in TypeScript is a very very different type of program than Deno - maybe much more so than novice programmers can appreciate - little of Deno is written in TypeScript. The target audience is the 5 to 10 people who work on this particular internal system. Please don't draw any broader conclusions.
The original reasons listed by Deno for removing TS had the undertones of "senior engineer who hates certain tech because they didn't use it right". Naturally, their response was to say "you probably don't understand cause you're a novice".
But I don’t think I’d ever want to go back to plain Javascript.
Guess a combination of esbuild during development, and tsc during commit would be the best of both worlds.
> we're removing the types from internal code and making it pure JS. this reduces complexity and helps us ship a faster product
Yeah, because dropping static types from 10K lines of code is definitely not going to be a maintenance nightmare in the long run.
But perhaps this is something beyond the simpleton comprehension of us novice programmers
My personal experience is that there’s a size threshold where static tying becomes more useful. Below this threshold, the problems solved by static types are still mostly tractable by humans and things like linters. The threshold is different for every program and set of developers.
10K lines is small enough IMO.
> But perhaps this is something beyond the simpleton comprehension of us novice programmers
That’s just flamebait, the comment is better without it.
Here are some key words that many kibitzers here could benefit greatly from internalizing:
> > The target audience is the 5 to 10 people who work on this particular internal system.
> Yeah, because dropping static types from 10K lines of code is definitely not going to be a maintenance nightmare in the long run.
Are you one of the people responsible for maintaining those 10KLOC?
> But perhaps this is something beyond the simpleton comprehension of us novice programmers
I believe Ryan Dahl would admit to having made more mistakes in large system design than most of us will ever have a chance to. Who knows, he may have picked up a thing or two along the way that most of us haven't.
This could also just be an internal product decision that really doesn't affect anyone outside the core project team, while some of us here are projecting way too much onto it.
Okay, unsubscribe. This is not the tone you use to interact with the community. This is brash, elitist hogwash.
Object.defineProperty(HeaderImpl, "name", { value: "Header" });
export let Header = class Header { }
> Who knows if this kicks Header out of some optimization path in V8
It doesn't.
---
Forgive me for being suspicious of the whole thing because a couple of the rationalizations don't make sense, I'm sure there are some valid points raised there.
TS feels like something that was created to lure programmers who couldn’t wrap their heads around JS loose nature. Almost like it was created to convince Java and C# developers to use JS.
It have never felt about it like something that would make my code better or more organized. It definitely slows me down with very little benefits in return. (Again, can’t stress enough that this is a personal opinion based on personal use cases)
I appreciate that it forces me to be more intentional about organizing my code but I also get that when I use frameworks.
Of course I also appreciate when I can identify that an error is coming from a type mismatch, but as I said I learned to program in loosely typed languages first, so I never created that concept of defining types in my head. I’m always extra-aware of types mismatches in my JS code and is usually the first thing I check when something goes wrong. But I have never felt the need to have a way to explicitly define the types that I’m working with. I honestly fail to see the benefit when a large amount of code that you have to interact with (libraries and such) was written in classic JS.
Once again. Probably unpopular opinion, so I don’t mind the downvotes.
All of this is possible with jsdocs and flow but they are not nearly as powerful as ts and often times verbose. I don't like polluting the comments.
You don't need to use typescript everywhere although the popular zeitgeist will tell you to avoid any or ts-ignore. It's fine to write some parts in ts and others in js. Think of ts as eslint, jsdocs or flow. You don't need to be religious about it. Take its benefits wherever you can or do you think all the other tools are useless as well?
You can refactor your entire codebase with a few operations and don't have to worry about anything breaking.
Is your timestamp in seconds? Millis? Is it a duration? Does it have a time zone?
Have you ever written a method that returns more than one type of thing? Or had polymorphic inputs?
Where are your dynamically dispatched calls even used, and how do you know you modified them all?
One of the most common operations in building new features is plumbing new state through the system. Types are there to help you in such perilous times.
Types are refreshing and reassuring.
I never have the class of bugs in my projects that people seem to laud about typing to solve. It’s kind of ridiculous to me and cumbersome and makes certain dynamic programming concepts difficult or impossible. I don’t want to give up power just because other developers are sloppy.
I want my code to be readable, and succinct. A lot of times that requires creating functions dynamically at runtime or some other such thing that would void the benefits of a typing system anyways.
I recognize that in working with a team where people have other styles, varying attention to detail levels and skill, they can’t all be trusted to produce bug-free code without the constant nagging and overhead of a pedantic type system.
My solution is not to employ another piece of technology to solve this problem, rather it’s just not to work with those people.
If you can’t write dynamic code without a bunch of type errors at runtime, learn to be a better programmer first before blaming the tools.
That sounds fine, as long as no-one else will ever have to read or work on your code.
On these projects you often find yourself needing to work on higher levels of abstraction. Types help here by instantly telling you what the data is and providing hints.
Types take extra time to write and maintain, but they significantly reduce the difficulty of working on large projects.
How many times has six-months-hence you cursed you for this attitude?
Be honest.
So you could be leaps and bounds better than me. Or you could be--let's be generous and call it "optimistic to a fault"--and retrofitting to a narrative about Good Programmers Don't Need Guardrails. And, TBH, I see a lot of people who are the latter and very few who are the former--and most of that very few tend towards statically or at least optionally typed languages by default. So forgive me if I don't believe you.
Will you in two years? Will your coworkers? Will the contractor who has no context for the code understand it?
Like variable names, types convey information that are crucial to the understanding of your application.
Under the hood, there are "types", even if they are weak, so you'll need to know about implicit typing anyway. Making them explicit is both a way to test correctness and a way to document your code.
Can you say the same about all your dependencies?
Meanwhile: https://www.zdnet.com/article/microsoft-70-percent-of-all-se...
'Percentage of memory safety issues has been hovering at 70 percent for the past 12 years.'
And this is not just Microsoft–it's a pretty consistent finding across industry, that when programming in C/C++, memory safety issues make up around 65% to 70% of all issues.
What this tells us is that people in general can't guarantee the safety of their code, but an automated checker can.
And I suppose when those requirements never changed you never had to rewrite lots of these perfect functions or resort to “append only” programming in order to make sure regression was impossible instead of making large changes in place.
Forgive me, but it’s clear you have never worked on a program of any consequence, so it’s hard to read your comment and do anything but wish you the best while also completely ignoring its content.
It's like trying to convince experienced motorcyclists that you don't need a helmet because you're quite the driver. It's just defensive driving, people! That the helmet is some sort of crutch you've transcended, and maybe if everyone had a bit more skill -- like yourself -- they'd understand.
Yet it always seems to be half of the dynamically- vs statically-typed language debate.
The only perfect decision I made is to accept that I make imperfect decisions, I wish I could attain the consistent perfection that you describe here.
I and most people however are human and make mistakes sometimes. Worse, the bigger the project, the more you need to collaborate between people and sometimes teams. Types are almost like docs that check your code at compile time and reduce the chances of human error creating a production failure. Its not a panacea but its certainly nice to know that if I change a functions parameter, that the system breaks someone else's code depends on that function following a certain signature. At that point, I can either write a different function, message him to update his code or discuss how to make the changes together. All 3 cases are MASSIVE upgrades over fiding out shit broke in prod with some undefined behavior arising.
People often say this, and I don't get it. What JS tests are you writing that become unnecessary in TypeScript? I've used a fair amount of TypeScript and plain JS, and end up with similar amounts of tests for each. With JS, I almost never want to verify only that a value is of a specific type; I want to look at its contents, which means I'd need to write the same test in TypeScript.
We're pretty deep down in abstract discussions here and I have no idea what your code looks like or what it does, but I think it's helpful to point out that knowing the type means you already know the contents to some extent.
The way I see it, checking if a value is what you expect is always good, but if you additionally get an error automatically because your number is now suddenly an e-mail address, or the function you wrote that expects phone numbers suddenly gets a name instead, in my mind that's "built in testing". It doesn't make other testing unnecessary but it sure raises the bar for cleanliness -- which is especially useful when you're working with other components that aren't JS where types do really matter (like databases, for example).
If such (refactored) method is being called from code that's not covered by tests, the bug may not be discovered until it blows up in production. With static typing it would be caught at compile time.
Well there's your problem :p
I agree that if code has no tests, it's more likely to be correct if it's in TypeScript than JS. But I don't consider that an acceptable bar, so if I'm doing your code review I'm going to ask for tests in either case. And once you have tests that verify the behavior of the code, you're not getting much incremental benefit from type checking.
No unit tests needed to show that it is balanced :-)
function doSomething(thing) { ... }
You need integration-level tests around every method call to ensure that caller and callee agree what kind of 'thing' representation to use. Type systems can eliminate this class of bug entirely.
As for frameworks in the languages you listed...
Ruby and Python: https://github.com/HypothesisWorks/hypothesis
Elixir and Erlang: https://github.com/proper-testing/proper
Node and JS: https://jsverify.github.io/
As for real world use-cases, imagine you’re writing a program that accepts timestamps as input and has to implement branching, requirements-defined business logic based off of them. When you’re writing your unit tests you can use the requirements to select timestamps that are “known good” and “known bad”, but it’s hard to explore this state space on your own.
Same thing goes for handling unexpected inputs to certain functions. You probably don’t want to check _every_ type of input for _every_ dynamic function, but it might make sense to make sure that certain “entry points” to your program fail in the expected manner when they get poorly typed input.
1. The library makes an assumption on a type (in the case of Python, that's typically "there's only one type of strings", "there's only one type of streams", sometimes "there's only one type of numbers", etc.). Since the language is loosely-typed, there's no API barrier that checks this, it's all duck-typing.
2. Tons of code is written based on this assumption.
3. Assumption changes. In this example, this happened with the Python 2 => Python 3 migration, but it also happens once in a while with smaller version increments of either Python itself or Python libraries.
4. Where is the assumption used in your code? Well, if you're lucky, you're going to find out once a piece of your code throws an error because it's attempting to use a method that doesn't exist anymore. If you're not lucky, the behavior of your code has changed subtly.
In the loosely-typed world, the only way to avoid this is to have very strict boundary checks. Which means basically manually implementing subpar type checking at the borders and unit tests and/or fuzzing just to ensure that your mock-type-checking actually does its jobs.
Of course, you can often get away without doing this as long as your code is for internal use only. But if you're writing a library and if you want your users to be able to troubleshoot problems without too many difficulties, you're doing manually what the compiler is doing for you in a statically-typed language.
Metric or imperial integers? Your unit tests failed to prevent this $300 million blunder:
https://www.wired.com/2010/11/1110mars-climate-observer-repo...
But more commonly what is going to happen is that someone (two years from now) is going to change a 'person' parameter from legacy SSN to database id and some users are going to get "you don't exist" when they show up at the hospital to get medical service.
You're right though, most folks in the JS/Ruby/Python ecosystems don't do this kind of testing. It's a recurring joke:
https://www.google.com/search?q=2+unit+tests+0+integration+t...
Sure, but you need those tests anyway to verify that your code actually works. I agree that type systems reduce the occurrences of some classes of bugs; I'm only disagreeing with the claim that they reduce the amount of tests you need to write.
I recently ported some (quite complex) code I wrote from JavaScript to Typescript. The code has about a 2:1 test to code ratio, and a fuzzer for correctness. While porting, I ended up adding a couple “useless” assert(typeof x === ...) calls to quieten the compiler, which felt useless because my code was correct. Lo and behold, the assertion tripped in my test suite - apparently I was sometimes treating a string as an object and didn’t notice. Which was a serious issue; and could end up being a security problem for some people. My fuzzer didn’t find it because it never occurred to me to add string method names in my random data generator.
Generally I find that the bugs that are easy to find with tests and the bugs that are easy to find with static types are different. You can eventually find all bugs with a sufficiently large test suite; and with enough PhDs you can apparently formally prove everything. But you get the best bang for buck with a little of each. A few tests is much better than no tests. But in the same spirit, I find no matter how big my test suite, there’s a good chance static types will improve my code.
Typescript is far from perfect, but I sleep better at night with a type checker checking my code.
I don't think we disagree much. As your experience shows, trying to replicate type checking via unit tests is almost always impractical. Which means the tradeoff of not having static typing is not that you're writing more tests, but that there's an increased chance of bugs.
And I agree that once a JS project is large enough to have a build process, the benefits of TypeScript almost always outweigh the costs. I'd even like to see JS interpreters allow and ignore TypeScript types, so you could use it without a build step.
Why would you design a function in such a way that one of the arguments can represent so many different things? The problem here has nothing to do with testing. The problem is that the function itself is poorly designed.
If anything, the difficulty of writing a test for such a function would in itself be an indication that the function needs a refactoring.
Idk about you, but when I’m sussing out the behaviors of a complex application it can take numerous iterations over different possible types and shapes of the data in order to come up with something both robust and performant.
The single greatest benefit of languages that afford static analysis is that, when I change the shape of some piece of data where it is defined, my editor lights up like a Christmas tree and informs me of every place in my code that was just broken! This is extremely useful for being able to quickly iterate on features. I don’t have to remember all of the calls sites dependent on some API. It allows me to confidently make (sometimes large) changes to my domain and know for sure which pieces of code might also need to be refactored.
I’ve worked in codebases where the confidence I’m describing above does not exist, and what happens is a lot of defensive programming and wasted effort (and time!) dancing around new implementation because no one wants to change anything. Changes tend to become “append only” (we can only add to the interface) because it’s hard to know what’s going to break if you actually change it. This can be okay most of the time, but sometimes new requirements... well... require new approaches.
The above is related to “testing” insofar as I don’t need to test where pieces of code depend on one another in order to know when their contract breaks. But the benefit is not really about testing at all. It’s about work flow. It’s about velocity. And it’s about freedom.
Usually what I do is that once a project has reached a point of maturity and the requirements merit it I move parts of it to a typed language (go). That’s for things that need to be both performant and maintenable long term without much effort.
I actually benefit from that rewrite in that I have yet another chance to review assumptions and address higher order
I don't understand what that means. How is TypeScript removing friction between JavaScript and tooling? That doesn't make any sense. TypeScript is a typechecker and a language that compiles to JavaScript.
All it takes is once and you'll become a statically typed language evangelist.
If you eliminate a class of errors by doing so, isn't that still less errors overall?
A value of the wrong type passed forward because it's almost the right type can cause errors that make their way into production.
I hear you say "so unit test your code!", and to that I reply, we're a c# shop that has only recently started pushing some of our code to the client. If you can convince management that we should shift our paradigm to allow for the time to write tests that don't add much to the sprint we're in, I can send you their email address.
The fact is, a large portion of unit tests solve the same problems that strong types solve.
1.) The fact that I even got involved meant the issues were particularly hard to track down
2.) These errors seem to be more of a time sync within a larger team/project than many realize
Typically these were caused by either scope issues and/or some context object that's been riding dirty all day. In the case of the latter hunting down the code that did the naughty can be particularly soul crushing.
I have seen my share of "innovations" come and go, but I too fail to see the empirical evidence that most bugs/errors are caused by type mismatch. In fact, I would even go as far as to argue that in my experience the benefits of typed languages (and the class of bugs they prevent) do not outweigh the usually larger code size (and often even additional complexity).
Maybe I've just tried the wrong languages, but I was never convinced of the great improvements usually promised. At the same time, I have seen other people twist and jump through hoops, while vigorously holding on to their conviction how all that is an improvement.
Sure, some errors will move from run time to compile time with typed languages. But if those are a systemic problem in your code, you probably have even bigger things to worry about (the ones you already keenly pointed out).
From a cynical point of view, I can understand why corporate entities would love typed languages (for the increased code size and complexity), but that's another story.
For the (pretty obvious) logical error of claiming that you have to believe that most errors are caused by type mismatch to consider stating typing beneficial.
By having types, you open the door to formally describing the building blocks of your system. This lets you create very specific, well understood, pure pieces that then are composed together to bring about your desired functionality.
Why rely on documentation that a compiler cannot statically verify, as opposed to a type (annotation or otherwise) that can?
I think you're correct regarding your "most errors are caused by" comment, and no programming paradigm will ever get rid of these - you can always find a way to write bad code. And you are also correct that there are type zealots out there who create beautiful prisons they call abstractions. But at the end of the day, types are a tool, and while I'd highly recommend learning them, I also recognize some projects and teams will never make use of them.
Static types are documentation. Better yet they're documentation that can't go stale without a computer yelling at you that it's wrong.
> Static types are documentation. Better yet they're documentation that can't go stale without a computer yelling at you that it's wrong.
Absolutely, and I'd go further:
"Most errors, in my experience, are caused by bad architecture, poor documentation, and poor communications strategies between MVC, etc"
Good static types (a la Hindley-Milner) encourage good architecture, serve is documentation that cannot go stale, and encourage good communication strategies between different parts of a codebase.
TypeScript adds additional type checking at transpile time to code. The assumption that the people who can write code with these more rigorous checks can't write the same code with it all removed ... I would examine that a little more closely.
On the contrary, as somebody who learned programming with a mix of static and dynamic types, I felt like TypeScript showed me what a good type system was capable of in giving me the best of both worlds.
Instead of being pedantry, I felt like the type system was actually on my side for once and helping me to write correct code with fewer unaccounted-for corner cases and unexpected runtime crashes. And it does that while avoiding the feeling I got from previous statically-typed languages I used where the type system felt like a straight jacket incapable of expressing non-trivial concepts and relationships.
For small projects, dynamically typed languages are fine. but as the project grows bigger, the static typing will help and will make thing simpler.
As the project grows, refactoring will be an issue. Static typing will make it simpler to refactor.
Tests are smaller.
Reading the code will be easier since you know the contract of functions.
Fewer runtime errors since most errors will be caught at compile time. Also, the stronger the type system it is, the more it will be caught at compile time.
I don't think it slows me down that much. It's a little more typing, and there is compile time overhead. In some cases, you have to work your way around the type system. But at least with Scala, I am confident my code will work as intended as long as it compiles successfully. I spend less time debugging in runtime and writing tests. I think static typing might save time in the long run but not short run.
Would love to know how your journey was. At first I was super hesitant, but now I am actually floored at the power Typescript has. The ability to tell the compiler to simply "trust me" has enabled me to build many abstractions I'd battle for hours to do in Scala.
Personally, I think structural typing systems are the path forward for functional languages, and that means accepting that Haskell needs to go to move on to a retirement home. I was pleasantly surprised by how well Scala did subtyping with ZIO's `Has` + variance annotations. Shame TypeScript is bivariant with function parameters.
I'm 100% TypeScript now and miss Scala. I miss using monads. But since Typescript doesn't have for comprehensions, using any of the monad libs for TS (like fp-ts) is messier than plain old TS. Also, Typescript doesn't have pattern matching, thus ADTs are more painful in TS.
But TS is a nice language. Instead of monads, they have stuff built into the language to make things easier and cleaner like null coalescing (compared to Option) and async/await (compared to using IO or Future). I can live with that. I've grown to like TS.
I'm curious what abstractions you had easier time with in TS compared to Scala.
I do miss the for comprehensions, but have built an abstraction that does the same thing for my purposes. It isn't quite as good, due to the whole monad thing, and me fitting it to the average Javascript programmer who won't want to learn functional programming all at once. So at its core it is pretty much async/await with sugar.
I quite enjoy exploiting structural typing for unit testing purposes. Gone is the need to learn a mock framework and its DSL; just write plain ol' JS for the parts you are interested in, and tell the compiler to chill. ;) Long term, of course, you should probably just have doubles.
"Pass in three parameters with a callback in the fourth position. The second param is configuration so pass in a string, but if you need to really customize stuff actually pass in an object instead and by the way if you don't have a third parameter then just chuck the callback there, why not!?"
Typescript is great in that somehow it is technically possible to write types for existing code written like this, but writing new code like that is cumbersome enough to prevent it.
The problem is, you'd have to have the documentation open to make the slightest tweak to anything. No help from the type system.
"This is an array of strings, but it could be an object. If it's an object it will have this key and based on the name of the key it's this type. But it could also have or not have this other key and..."
ARghhhh! You can tell what kinda language the back-end was written in with that crap because while TypeScript can actually type most of that stuff it's a huge mess in something like C#, Golang, Java, etc.
It's the same BS as Python and JS, the lack of typing but fundamental nature of systems that require data to be typed means half my code is checking that everything is properly typed!
Dynamic types have their uses, but in general they're more trouble than just writing typed code to begin with. And when your type system is more a suggestion than anything it provides little benefits.
This is true, and confirmed by the TypeScript team
https://stackoverflow.blog/2020/06/15/talking-typescript-wit...
> Do you remember why the team came up with TypeScript, why you wanted to release something like this?
> A: When I joined the team, there were a lot of people at Microsoft who wanted to develop JavaScript at what we call “application scale.” Teams like TFS and Office wanted to build large JavaScript applications. A lot of those people had familiarity with statically-typed languages— C++, C#, Java, that kind of thing. They wanted to have that static typing available both for conceptual scalability and for the tooling.
> The language experts at Microsoft looked at the situation and said, well, we could try to write a new language like people had done before. There are projects like Script#, which takes C sharp and turns it into JavaScript, or languages like CoffeeScript that said, what if JavaScript had a different syntax? But what they decided instead was to just take JavaScript and add static types on top of that.
> At the time, JavaScript was experiencing a renaissance of growth, where they were adding features to the language again for the first time in a pretty long time. People wanted to use those cool new features, like arrow functions and classes, and they didn’t want to wait for all browsers to adopt them. They wanted to be able to use those features right away. I think those were the three things that TypeScript offered upfront: static typing for error finding purposes, leveraging that static type information for tooling, and providing the JavaScript features of tomorrow, today.
Seriously?
> They wanted to have that static typing available both for conceptual scalability and for the tooling
Sounds like they understood the loose nature pretty darn well.
I use Typescript everywhere I need Javascript these days, it's a breath of fresh air, and my code feels maintainable, tractable, and much easier to refactor.
Ultimately, for any serious work, especially enterprise level work, your team members may include people of different calibers.
In such a case, where you have to work with code written by other members of your team, it becomes a serious problem to develop quality code.
For lone developers, it should not be a problem. But in enterprise settings, I think that is the reason why Node is not popular.
They like TS a lot. And it’s obvious why. Having said that, I know no “old” programmer who used to prefer a different paradigm to express a wish to switch to TS.
Rich Hickey put it well in this quote “ Functional programs are desirable because they are simpler and easier to reason about, due to being mathematical i.e. they are free of time and place.”. I agree.
I mentioned that paradigms of functional programming languages like Clojure translate well to JavaScript. They do not to typescript, because it is a statically typed language.
If you are interested in why are you dynamically typed language has benefits for functional programming, please see this talk: https://youtu.be/oytL881p-nQ
Typing is optional both for individual variables via the `any` type, and compiler-wide, by configuring it to assume untyped variables are `any` by default.
I disagree, though. Instead of going with a lengthy list of why I think TS is great and disproving your arguments, which I assume everyone else will do, I'll admit that types can be hard, and so doing them right, or even doing them as best as one can, will always take some time. For a small project or a small team, that time might be hard to justify. But I think this is resolved with a good understanding TypeScript.
Even so, TypeScript might not make sense for many projects or small teams, but I find it hard to defend that TypeScript won't be a net win for any medium team or medium-size project.
This was my believe before, but I've seen multiple teams go from JS to Flow or TypeScript and the result has always been, when there's expert guidance, extremely positive. Productivity goes up, more bugs are fixed because there's more confidence in the safety of the changes, more people feel confident to work on any part of the codebase (vs just the people that originally built it or have worked more on it), and the defect rate dropped dramatically.
Accidentally mixing incompatible types is mostly a problem for junior developers. I'm much more likely to accidentally put on my pants, socks and shoes in the reverse order in my daily life than mix up incompatible types in my code.
The solution is not to add stickers all over your clothes to remind you which way is the front or what order you need to put them on; you don't need more tools, you just need more focus.
Compile time errors are much quicker and cheaper to detect and fix than runtime errors.
I don't agree with you. I don't think that loose typing is based on opinion, and I do not think it is because people who start out in statically typed languages are unable to "grasp" loose typing. I think we are seeing a people abandon loose typing because the industry has observed that loose typing is harmful to even small, a few thousand lines, projects.
Static typing catches bugs. It makes your code easier to maintain. It makes your code safer to refactor. It checks your code to ensure correctness. It provides documentation about variables. It reduces the amount of code you have to read in order to understand a specific piece of code. These are empirically good things that you do not get in loosely typed languages.
An untyped value does not carry any information or explicit assertions about the variable. Take a URL for example: A "string" URL does not carry any information about weather or not the url string is a valid URL.
In an untyped environment, only by understanding ALL upstream code are we able to make assumptions about downstream code. We know that the "string URL" is a sequence of characters, and nothing more.
However, if we have a type, we can have a "URL" type. Since we have a type, we do not need to concern ourselves with all upstream code, only the code associated with the "URL" type. So, instead of potentially thousands of places to seek information about our "URL" variable, we have only one: The "URL" type. We look at the implementation of that type, and we know weather or not the URL has been validated.
Types dramatically reduce the amount of code you have to read in order to understand a given piece of your application. This makes your application more secure, more readable, and easier to maintain.
Oh, wow, given this unnecessarily combative start, am I about to finally read conclusive proof that will stop that over sixty years old flamewar?
> I think we are seeing a people abandon loose typing because the industry has observed that loose typing is harmful to even small, a few thousand lines, projects.
People are not abandoning or adopting anything, both static and dynamic typing have existed for as long as computers have been a thing. Hell, LISP is older than pretty much every statically typed language still widely used, methinks.
> Static typing catches bugs. It makes your code easier to maintain. It makes your code safer to refactor. It checks your code to ensure correctness. It provides documentation about variables. It reduces the amount of code you have to read in order to understand a specific piece of code. These are empirically good things that you do not get in loosely typed languages.
Where's this empiric evidence of all these advantages you mention? I'm a bit shocked that I managed to miss it... And that so did half the industry. Such conclusive study would have ended that and the industry would have coalesced into this unquestionably superior type system you mention.
> An untyped value does not carry any information or explicit assertions about the variable. Take a URL for example: A "string" URL does not carry any information about weather or not the url string is a valid URL.
Can't do that with static languages either. Call it a type (or class), call it a parsing function, ultimately the result is the same. You need something to ensure that it is an url. Those exist in dynamically typed languages and are not obviated in static ones.
> Types dramatically reduce the amount of code you have to read in order to understand a given piece of your application.
Conversely, types substantially increase the amount of overall code one has to write.
These has never been a matter of opinions. The flamewar still rages after decades because what's the factually best type system hasn't been found. Hell, we might be likely closer to prove that that question has no answer.
In short, ultimately everything you said was just, like, your opinion, man.
In a loosely typed language you cannot have compiler-level enforcement that a variable has been produced by a factory function that ensures that the passed in variable has been parsed and determined valid.
Additionally in a loosely typed language you do not even get the OPTION of labeling a value as "validated/unvalidated", you only have one option: Pass a string, and expect the new developer to read upstream code to understand where the variable came from.
> In short, ultimately everything you said was just, like, your opinion, man.
Is it my opinion though? Or did I pull it straight out of Building Secure and Reliably Systems, a book written by Google engineers to express industry best practice for developing software at a scale?
Hundreds of books have been written on the matter, and the results are still inconclusive.
> Is it my opinion though? Or did I pull it straight out of Building Secure and Reliably Systems, a book written by Google engineers to express industry best practice for developing software at a scale?
Yes and yes, you happen to share their opinion.
Now, should we pretend that appeals to authority turn an opinion into fact now? If so, to whose authority should we appeal? Theirs? The different google engineers who work in more dynamic languages? Hell, designers of dynamic languages who have worked at Google? The people who designed and work in LISP or any of its derivates? Whose authority weighs more?
Or maybe we should acknowledge that appeals to authority hold no real decisive value in deciding something like this, and instead contribute to studies trying to gauge how effective different type systems actually are.
Except that even those turn into flamewars: https://www.hillelwayne.com/post/this-is-how-science-happens...
> Conclusion This paper described an experiment comparing static and dy- namic type systems for programming tasks in an undocu- mented API. We gave 27 subjects five programming tasks and found that the type systems had a significant impact on the development time: for three of five tasks me measured a positive impact of the static type system, for two tasks we measured a positive impact of the dynamic type system. Based on the previous discussion, our overall conclusions for the use of static/dynamic type systems in undocumented APIs are the following: 1. There is no simple answer to the static vs. dynamic typing question: The question of whether or not static types are helpful for developers cannot be generally an- swered without taking the programming tasks into ac- count. In fact, this is completely consistent with the results of previous experiments, such as Prechelt and Tichy’s [22], or our own experiments[10, 28]. 2. The type system has an influence on the development time: The choice of the static or dynamic type system had an influence on the development time for all program- ming tasks in the experiment. Again, this is consistent with previous experiments (such as [11, 22, 28]). 3. Dynamic type systems potentially reduce develop- ment times for easier tasks: Although we are currently not aware of how to determine exactly what “easy” and “hard” means, it seems that if a dynamic type systems has a positive impact, it is for easier tasks (which is consistent with the experiments described in [10, 28]). Although there was one counter example in the experi- ment (task 1), we think that the result for this task is a consequence of the chosen subjects’ low familiarity with the dynamic language, Groovy (despite the presence of a warmup task). 4. Static type systems reduce development times if (a) the type annotations explicitely document design de- cisions, or (b) the number of classes to identify in the programming tasks is relatively high.
https://www.researchgate.net/publication/262317340_An_empiri...
Given the challenges to getting a good result identified in the study how do you think it could have been done better today?
And, of course there are exceptions in both directions for all of findings 2, 3 and 4; and those exceptions would also contribute to 1.
It's for findings like those that I wish people stopped treating type systems like religions. And paradigms, for that matter.
It just happens to be that I have never been exposed to the circumstances that would make me re-evalute TypeScript merits.
People who use types regularly feel the same way about typed languages. Types don't test as many things of unit tests of course, but feedback you get on what they do test is immediate - literally as you type in a IDE. The speed of that feedback is important. It's a well know in quality circles that errors discovered later cost far more to repair. In the case of types in makes you don't write a hundred lines of code under the expectation a type provides method foo(), only do find out later the foo() provided takes an argument argument you don't happen to having lying around.
While payback for types over unit tests is smaller, the effort required to use types is at least commensurately smaller, and the amount of typing required is way smaller, particularly now we have type inference. Also in my experience unit test's need to reach every line of your code forces some pretty tight constraints on how you structure it. Who hasn't had to refactor some block of code so you could test some obscure corner case catering to an unexpected error? Types don't require that - the compiler always sees everything. None of this matters much if you are going to write unit tests anyway of course, but no one does that for small projects. This is another advantage of the type system - it's always there, even for 5 five liner.
Finally, types come with an added bonus. They give the compiler more information to work with, which means the compiler can make programs run faster. Rust talks about cost free abstractions, which the extreme case: the compiler knows so much about what you are doing it generates no code at all. In rusts case it rarely has to check for if a pointer references null before accessing, and greatly reduces the number of array bounds checks, has no garbage collector, and the pattern of a function checking the number and type or argument in order to implement overloads just disappears - along with the runtime overhead of doing the checking.
> TypeScript isn’t proving itself helpful to organize Deno code. On the contrary, the Deno team is experiencing the opposite effect. One of the issues mentioned is that they ended up with duplicate independent Body classes in two locations
This feels like process immaturity or unfamiliarity. Thousands of other projects manage to do just fine.
These folks are free to do what they want with their project, but this is not a good look, especially to those that are skeptical of the javascript ecosystem.
There are also huge classes of bugs that exist in these projects that a statically typed language completely eliminates!
If they're running into trouble from having to have types, it's almost certainly design and architecture issue and does not make me think that deno is going to be a solid project.
Strong typing is from OOPS, itself an optional programming methodology. There are many JS programs that simply access other (JSON) objects directly without going through an interface. In these instances, typing is counterproductive.
One of the beautiful things about JS (and Lisp) is that variables and functions are interchangeable and, the case of JS, both can be accessed directly without worrying about an interface.
At the end of the day those type exist in JavaScript. If you access ‘first_name’ instead of ‘firstname’ it will not work. The difference is that typescript would let you know before running the code instead of getting an obscure runtime error.
I'm honestly wondering: What are people doing to end up in a position where TypeScript compile times are a problem?
If they dropped use of classes, or differentiated the names of the two things in collision the problem would appear to be solved.
To turn that around and blame TypeScript as the point of failure is an extremely bad omen suggesting their code style opinions are more important than product delivery. I could live with that nonsense if this were a framework or some minor dependency, which this application is not.
Edit for the curious, here's the monster type that caused such pathological build times...
type ImmutablePrimitive = undefined | null | boolean | string | number | Function;
export type Immutable<T> =
T extends ImmutablePrimitive ? T :
T extends [infer U] ? readonly [Immutable<U>] :
T extends [infer U, infer V] ? readonly [Immutable<U>, Immutable<V>] :
T extends [infer U, infer V, infer X] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>] :
T extends [infer U, infer V, infer X, infer Y] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>, Immutable<Y>] :
T extends [infer U, infer V, infer X, infer Y, infer Z] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>, Immutable<Y>, Immutable<Z>] :
T extends readonly [infer U] ? readonly [Immutable<U>] :
T extends readonly [infer U, infer V] ? readonly [Immutable<U>, Immutable<V>] :
T extends readonly [infer U, infer V, infer X] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>] :
T extends readonly [infer U, infer V, infer X, infer Y] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>, Immutable<Y>] :
T extends readonly [infer U, infer V, infer X, infer Y, infer Z] ? readonly [Immutable<U>, Immutable<V>, Immutable<X>, Immutable<Y>, Immutable<Z>] :
T extends Array<infer U> ? ImmutableArray<U> :
T extends ReadonlyArray<infer U> ? ImmutableArray<U> :
T extends Map<infer K, infer V> ? ImmutableMap<K, V> :
T extends ReadonlyMap<infer K, infer V> ? ImmutableMap<K, V> :
T extends Set<infer M> ? ImmutableSet<M> :
T extends ReadonlySet<infer M> ? ImmutableSet<M> :
ImmutableObject<T>;
type ImmutableArray<T> = ReadonlyArray<Immutable<T>>;
type ImmutableMap<K, V> = ReadonlyMap<Immutable<K>, Immutable<V>>;
type ImmutableSet<T> = ReadonlySet<Immutable<T>>;
type ImmutableObject<T> = { readonly [K in keyof T]: Immutable<T[K]> };So, instead of waiting a minute for the typescript compiler, you wait for the engineer to write tests.
Firstly, they'd be written in the same language that you are coding in, not a weird, half baked type language, that adds visual noise to your code.
Secondly, they'd have much more power to define meaningful and useful behaviour rather than being restricted to talk about correct behaviour via types.
Thirdly, they'd run when you wanted the tests to run, and you could tier different tests to run at different times, so they aren't all slowing you down while you're doing fast iteration.
I used to like types, but then I realised that what matters is how quickly you see the bug. Seeing it in your code editor is brilliant, but if it's slower in time than hitting control S and seeing the actual app in the other pane auto reload, and fail or not, then it's worse.
These days I look on the idea that you know the exact types of all data your program will interact with at the time you write your code as the same sort of mistake that we made when we thought we understood the deep inheritance hierarchies of the real world.
I write a lot of Purescript and a lot of Clojure. Purescript, being basically a Haskell variant, is about closing down every last little part of the system into types. You _do_ pick up a lot of visual noise for this (like `liftEffect` ugh...). Whereas Clojure, is the polar opposite. It's about keeping the system open, using large chunks of data, and having functions take/change what they need and pass along everything else none-the-wiser to what's present.
Bouncing back and forth between these two worlds, I think I've begun to lean towards the opinion that I really only care about strict types at the boundaries of my program and certain very specific checkpoints along the way (generally, something like a module/namespace boundary).
Clojure has Spec, but it misses the mark in my opinion because it doesn't solve refactoring. It's not (currently) instrumented enough to help the me as a human make changes to the code without also just "bumping into the guard rails" to try and figure out what I broke along the way.
A middle ground orthogonal type system sounds really appealing to me. "Type coverage" is an idea I've been kicking around. Something like Spec in Clojure, but more symbiotic with the host code such that it can tell you things _about_ your code and help in refactoring, symbol resolution, etc..
That's a lot of words, but tl;dr I think I agree with you : )
It's quite obvious that there are downsides to types that most developers are missing - because most developers seem to feel that they get massive benefit from types, but empirical studies seem to show that if there really are benefits, they are not massive.
I see the downsides in a few ways - they open up another avenue for 'architecture astronauting', they are not in fact accurate representations of the real world in many situations - you're dealing with files, or messages coming in from the network which are not typed, and even when they are (e.g. databases), they can change under your code while you're running. That's not to mention the points I made above around usually being an entirely different, hobbled language that gets sprinkled through your source code. Types encourage people to build ridiculous code generation pipelines during build time, and builds getting larger and flakier are about the worst thing in the world for fast iteration. I find the Smalltalk approach of coding in a live image really interesting, and I worry that obsession with types are closing off that kind of future. There's also the fact that your language, and particularly your type system constrains what code you write - like a programming version of Sapir Whorf, and most type systems are bad at expressing very high levels of abstraction. Scala had to invent a documentation tool that hid the real types of things because the type signature of higher order functions like map and reduce were scaring people.
I was interested in your opinion of Spec, because I'm aware of it and find it interesting, but I don't do a lot of Clojure so I haven't really used it practically. I liked that it aimed to address more than just verification - schemas, object generation for property testing, these things are all closely related to the general concept.
I think your ideas about a middle ground type system seem very interesting.
I saw recently that there's a version of Nim 'DrNim' that incorporates the Z3 theorem prover which allows you to assert and prove much more interesting things than typical type systems. It sounds fascinating.
Which, if we're doing TDD, involves no waiting at all, since it was already written before the implementation. Ideally anyway.
Type safety is easy in a dynamic language. Stop mutating variables and stop mixing types. Suddenly that whole class of errors evaporates. No need to test things that are impossible.
And for those times you want some magic in your life, nothing is stopping you, no need for generics, just proceed with caution and test accordingly.
Remove the browser practical monopoly, and JS popularity would vanish.
Also, languages are always a function of their ecosystems. You could argue, remove Apple products and Swift/ObjC would vanish.
Easy to reason about, great performance for most cases, simple to debug and write code for, a plethora of libraries, easy to ship, no need to have different teams for backend/frontend, and minimal tooling needed.
GP comment was presented somewhat snarkily, but I'm not sure they're all that wrong.
But I have yet to find a language that has the same set of features except JS. If you want to offer up an alternative, please do so.
There is not a single feature in JS you can't find elsewhere, often better implemented.
If anything, the vast majority of scripting languages (Python, Ruby, PHP...) have more features than JS.
JS is so lacking in features that half of its ecosystem is dedicated to compensate for that (typescript, babel, webpack, undersacore...).
Yet one feature does not a language make. A language is the whole of it's behavior, and JS is a perfectly fine solution with a wide mix of good features.
Is it the fastest? No. The most performant? Not close. Does it have the best ability to write language parsers? No!
But it's not supposed to. It's a general purpose language that's well suited for a wide swath of cases.
Most languages have transpilers and utility libraries - and those are useful tools but also not meant to be part of a language. Even python has dozens of compilation tools depending on how you intend to distribute your code.
It's the mark of a poor intelligence that buys on pros only without considering cons. We must weigh not only our features but how well the average engineer can use it, how quickly we can produce features and how safe our code is, and how easily we can actually hire and train engineers.
JS is one of the most popular languages in the world. My 10 year old nephew can write it and do a decent job of it.
JS is a fine language. It doesn't really bring anything new/different to the table (other than prototype inheritance, but that doesn't amount to much IMO), so the real differentiating factor is that it has a privileged position because of browsers.
Very dishonest take btw.
Here's a forum of people who could tell you why they use JS instead of your $favelang, yet instead of trying to figure it out, you just assume everyone is lazy or a junior dev except for you.
You should be a little more suspicious of convenient little truths like that where you're the cool elite engineer protagonist and everyone else is a bumbling idiot.
I use JS on the server. Any questions for me?
I also didn't say everyone is lazy. I said a lot of developers.
Most people here do not belong to that use-case. As they are mostly here by interest and are willing to learn.
But most people, after work, don't do anything with their knowledge.
And are more willing to learn node as it's the same on the backend + frontend.
Or keep using what they already know, with no interest for other languages.
So for the rare cases where async does matter (and they are a niche, given even big names like facebook code behind async proxy and do long polling), there are still little reason to choose JS.
I can't think of a single mission where, if I were not being forced to use it, I would chose it over something else. And I say that while coding a LOT of JS all year long. Even training people regularly in the language.
I double down on my initial statement, if tomorrow JS is removed from the browser, it would die in a few years.
It's a badly designed language that has been accumulated fixes over the years to make it acceptable while never removing what's bad. It was so initially bad that half the ecosystem of JS exists to not code in JS (typescript, coffeescript, babel, jsx, webpack...).
And it has received that much of attention because we had no choice. Google spent millions to make a technical marvel of a JIT to run it at decent speed.
Any tech that would have had half the resources poured into it would today be the incarnation of skynet.
It's a language surviving on artificial life support than can do so because it has the best blackmail game of all time.
Java?
The JVM is one of the most amazing piece of tech ever created. It's so good it supports not only what has been most the popular language in the world for years, but many newly popular languages like Kotlin or Closure. The perf have been optimized so much they are getting about 70% of C/C++, but with high level paradigms. It can pretty much do everything.
In fact, it runs on everything. It used to run on old school nokia phones. It runs on some credit cards! Despite being slowed down by patent trolling, Java found its way to more than half the modern mobile phones in the world throught Dalvik reimplementation.
It's everywhere.
Today, it is still ranking #2 on TIOBE and PYPL, despite having to leave with 2 decades of legacy design decisions and no monopoly.
You don't think JS is obviously unique for its ubiquitous Promise, async/await, + async-everything abstraction. You tend to only get that in other languages (Rust, Python, Ruby) by limiting yourself to a fraction of the ecosystem.
Btw, what's lazy is making convenient uncharitable assumptions about others than just... asking.
This forum is full of people who could tell you why they still use JS after they analyze pros/cons of other languages.
You are here because you are interested and want to learn. You probably know more than 1 language.
But that is not the general developer. Most developers are lazy and at the end of the day, want to go back to their wife and kids ( for example).
The real problem is that businesses don't want to train devs in the right tools for the jobs at hand, and are instead looking for the shortcut that will let them 'ship it'.
I agree, never claimed otherwise.
> You don't think JS is obviously unique for its ubiquitous Promise, async/await, + async-everything abstraction. You tend to only get that in other languages (Rust, Python, Ruby) by limiting yourself to a fraction of the ecosystem.
A fraction like nodejs or whatever subset of NPM you decide to use?
Core JS is what you get in a browser. That includes async/await, but it's hardly ubiquitous in usage in the core.
The main distinction JS has over Perl, Python, Ruby etc is that it's in the browser, so you can assume almost everyone has it and it's accessible in some manner if they access a web property you are responsible for.
I'd like to know what language you use as a baseline for this?
I made my first "real" Javascript code back in 2005 (a vector map system that worked in realtime) and I've been working mostly with frontend for the last 3 years so it is not lack of familiarity. In between that however I have programmed a lot of other languages, worked in a number of different teams, written new code, maintained old code and generally gotten some perspective on life.
That perspective has sent Javascript down my list of languages that are easy to reason about.
The lack of OO means no typecasting, inheritance models, interfaces, etc.
Arrays are fully dynamic, not type restricted and operate strictly by reference.
No memory mapping, manual garbage collection or even GC adjustments.
Suddenly the list of things someone has to understand to be a competent JS engineer is way smaller than a language like C or Java.
There are other languages that have a similar level of being easy to understand such as Python and Ruby, and they're good languages too but have a different set of tradeoffs.
Here's my attempt to create something similar from my perspective:
- Being able to know to at any point what a variable can contain is very liberating. (Ok, you can define a list of Object that behaves just like a list in JS: "fully dynamic, not type restricted", but thankfully for people like me that is not common in Java anymore.)
- Knowing that the compiler has my back allows me to work faster.
- My preferred languages (Java, C#, TypeScript) together with version control allows me to refactor fearlessly.
+ An engineer had written a script with a Dictionary of type <Transform, Material>. If you've touched Unity, you immediately caught the flaw: Transform's aren't unique enough keys and it's really easy to end up trying to add Duplicate keys, even though you really mean different objects.
+ So, since we don't care about uniqueness, shift from a Dictionary to a List. Except we have a problem, because we need to keep that Key->Value relation so we know what transforms get what materials. Well, we have a solution to that, it's yet-another-type called a KeyValuePair.
+ Well we're now nesting a generic KeyValuePair that needs type information inside of a list that is holding <KeyValuePair> objects, which also changes how we iterate through that list and add items to it.
+ In a language like JS, everything I just said doesn't matter at all because instead of using a Dictionary (Map in JS), we would of just said "Use a Set instead". Done.
If they started using PHP, without monopoly, they would not have switched to JS.
But plenty of people have switched from PHP to python or ruby.
> Also, languages are always a function of their ecosystems. You could argue, remove Apple products and Swift/ObjC would vanish.
Python, C and Java are not.
Right now, you have options to avoid writing js at all. There might be some glue code in js but you can write vast majority of your code in other languages and target js. People still choose to write in js.
Purescript, elm, clojurescript, kotlin, bucklescript, and the list goes on. There is a to-js transpiler for every popular language. Why aren't they seeing more usage?
So i don't subscribe to idea that js is not popular because of browser monopoly.
Because a layer of indirection is a heavy price to pay.
If I could code in Python with zero cost in the browser, I would. Hell, I would code in Lua, Lisp or Ruby if that was the alternative.
Lisp programmers use check-type[1] a lot, they proclaim[2] types for performance optimizations and they often deftype[3] if not only for code clarity. And then we have CLOS and its dispatching on types/classes...
[1] http://www.lispworks.com/documentation/HyperSpec/Body/m_chec... [2] http://www.lispworks.com/documentation/HyperSpec/Body/f_proc... [3] http://www.lispworks.com/documentation/HyperSpec/Body/m_deft...
I object, C is that.
There might be hundreds of API calls in X but the glue code is all going to be of the same general form: translate the X parameters to corresponding Y parameters, call the matching Y function, and then translate results back.
Within all these glue functions there will be a lot of common actions. X calls with integer parameters for instance might pass them as ASCII strings and Y might expect integers as 64-bit big-endian binary.
In this situation there is a good chance you can make a table that lists each X call, the types of its arguments and results, the Y call to map it to, and some flags to indicate that some things need special handling and what those are.
You can then have a program, probably a fairly short and straightforward program, that reads that table and writes out the glue code for you, sticking in comments marking places where you need to edit it to deal with those places marked as needing special handling.
As a maintainer of a popular 3k LOC typescript library, even at 1k LOC you start running into frequent type-related bugs with JavaScript
It did take quite a bit of discipline, but with a skilled team it really wasn't all that hard.
> It did take quite a bit of discipline, but with a skilled team it really wasn't all that hard.
Out of curiosity, what was the turnover within the team? My personal experience is that static typing helps a lot when you're new to a codebase. Probably less so when everybody can recite the architecture in their sleep.
That's what a typechecker gives you–an automated enforcement mechanism for ruling out certain errors during refactoring.
I had actually been watching Deno in case they solved any of that.
I recently watched this talk by Johnathan Blow[0] where he talks about quality of life in programming languages and spends a fair amount of time talking about compile times and their effects.
I think we collectively need to re-align on being performance focused. Tools are simultaneously amazing but crummy at the same time.
Note also: * type checks can run in parallel, so they won't block your other build block * the tsc has been making gradual improvements to performance and techniques to improve compilation and type checking * there are speedy alternatives to the TypeScript compiler (tsc) (eg: https://sucrase.io) for transpilation
I've witnessed teams that have switched from static typing to dynamic typing before, due to complaints about compile times and a lack of ability. It doesn't take long before their testkit balloons and they spend 10x the time on testing, dwarfing any savings they made by eliminating compilation. Or worse, the code turns into something magical but untouchable, lest something break...eliminating the ability that they imagined they would get.
Knowing they are aware they could easily fix this issue themselves but are instead dumping TS is cause for hesitation to invest in the project.
Things like when this system was built all the code assumed social security could not be null. We now need to make the system able to handle a null social security system. What code do I change?
Do I need to null check the results of this value?
What does this function return and how can I be sure I'm handling all use cases?
On the contrary, I think it shows great maturity in the decision making process. Many highly experienced developers have stayed away from TypeScript for reasons like those mentioned in the article (mostly to do with the build environment, source mapping, versioning complexity, etc...) These are very significant problems and not worth the small benefits which static typing brings.
pshh
There are places where no-typescript would be appropriate, but they’re vanishingly small. The idea that the benefits static typing brings are small is just laughable.
The reason this is such a hard choice is because static typing brings great benefits, so choosing to forgo that is not easy.
This post adds little to what's already said there.
If people are serious about types, then why isn't more front-end moving over to Elm, Reason, PureScript, OCaml etc? Hell, GWT is still out here even. TypeScript has convinced people they're getting the benefits of (strong?) types when they're just populating the auto-complete in VS Code.
Even then, why not just use any one of Elm, PureScript, Reason, ScalaJS, etc? Why implement yet another FP runtime? If something like this is the surest way to make optimal use of TypeScript then I feel like we're just right back where we started.
For most devs, I'd wager they're still ending up writing all the same tests they were before TypeScript.
tsc even offers to just consume jsdoc annotations for crying out loud. If its just some kind of type convention sandbox, what did it really bring to the table?
PureScript I really like but it's a massive jump from JS/TS to PS/HS.
All of these languages are harder to abandon than a library in the same language you'll keep using if you decide against the paradigm down the road. Do I like this compromise? Not really, but it makes sense.
Also, advertising use of a functional language might open you up to candidates who wouldn't otherwise apply.
It doesn't come with the fp-approach that io-ts takes, and has, in my opinion, significantly better error reporting. Just plugging this here because I believe that zod is grossly underused :)
Note that I'm not saying that Elm would be better than TS or any other language or anything like that, just commenting about the runtime checks. Obviously languages have their strengths and weaknesses (Elm has plenty of both) and TS has had good reasons to go with their design goals.
I have been using GWT, too hard to use comparing to typescript with pretty much same benefits.
I wish them luck!
One note of concern though: in their design doc they say incremental compilation is the number one problem and list multi-minute compile times. That’s huge! I wouldn’t use TS either of it took minutes to compile my projects. However, I think there’s room for improvement in their tooling if they took some time to explore what other large TS projects do instead of just giving up on it entirely —- VSCode for instance has incremental compilation in hundreds of ms or less.
Someone is trying to rewrite tsc in rust for speed up here - https://github.com/swc-project/swc
Cool project for runtime checking with ts - https://github.com/gcanti/io-ts
Edit: Remove unmaintained runtime type checking library.
const ofId /*: (Sql, number) => Promise<User> */ =
(sql, id) =>
...
Lack of libdefs is not a huge problem in my case. I have strong preference for shallow or no dependencies in projects and for things I use 3rd party code there are types or were converted from ts or simply added. Better support would help here but it was not a deal breaker.
Code editor support is not as good as ts but it does the job.
Another interesting effect is that arbitrarily deep npm linking is so much easier, the code is just js, doesn't need transpilation step, can be edited as is; I'm free to use any directory structure, ie. I often use root directory to drop files, they have natural require/import paths, no configuration etc.
For me, it's a joy to code like this.
I've found this very useful for gradually migrating a node.js project to a more maintainable state. Types also mean autocomplete works!
Additionally since it's typescript we can download the `@types/` packages for our dependencies and get type checking for those as well.
types can be imported either explicitly via `require()` or via the `@type {import("foo").Bar}`
There are definitely some limitations and caveats, like `Object` being aliased to `any` (I think this is being changed in a new version), but it's still way better than untyped JS, and having typescript integration is nice since our devs are already using typescript in other places.
Can even using things like https://github.com/typescript-eslint/typescript-eslint
Flow is both more advanced and faster than Typescript, but the community isn't as large, so Typescript continues to gain marketshare
I don't konow why they aren't separating type checking and dev builds from each other, but there is just absolutely no way simple TS to modern JS (no compiling to old ES5) conversion would take that long for a 10k LOC project, it must be the type checking. I'm using TS through Babel and it has been a breeze for much larger projects than that. Babel just strips away the TS types and we do type checks via editor integration and a separate CI step. Can't recommend this setup enough.
1. It seems that they write a ".d.ts" file manually in addition to using TypeScript for the code. This is dumb, since TypeScript generates the declarations automatically. However, for them "it was too much overhead and complexity when we attempted it before", which caused them to give up, a very dubious course of action.
2. They claim that changes take minutes to recompile, but TypeScript can compile incrementally, so this shouldn't happen assuming they are organizing their code properly. Also, you can just translate without type checking, which is no worse than using JavaScript instead.
3. They claim that "having two Body classes is obviously wrong" (?!). Of course having two classes with the same name in different namespaces/packages is perfectly fine in properly designed languages (including both JavaScript and TypeScript). Their "Header" shadowing problem also might be due to a lack of understanding of namespaces.
4. They seem to conflate JavaScript vs TypeScript with single file vs multiple files. Of course you can have a single TypeScript file or you can have multiple JavaScript files and bundle them with any JS bundler or just concatenate them.
Given that their codebase seems to be complex according to their compile time claims and that they don't seem particularly skilled given their claims, using a type-deficient language will most likely result in software full of bugs.
Interested to know how you're actually supposed to do this. I don't do much web development, but the few times I've had to dive into our typescript code at work, I end up just modifying the compiled javascript to debug things because the compile loop is painfully slow.
https://www.typescriptlang.org/docs/handbook/release-notes/t...
I'm with you on the rest of the points, though. I don't quite understand it, either. But then I'm not working in that codebase every day. Easy to criticize people from the outside!
Until those languages make themselves feel just like js or completely remove excessive js interlop, they won't be as mainstream.
Sure, if you want to make maximal use of the strong typing of F#, you'll have to annotate JS data as it comes in, but that is necessary complexity, not incidental. And if you're happy with having your F# code "dynamically typed", there's zero ceremony around importing JS stuff.
You do have that class of bug. That's like a cpp programmer saying they don't need smart-pointers because they never introduce memory leak bugs. You're a human.
... choice of programming languages is irrelevant here.
The Haskell community strongly disagrees...
Though, Haskell figures out the types for you, instead of you typing them in, maybe thats a big difference. In Haskell, do you find you have to think about the types anyway, or can you code it like you would clojure?
This strikes me as a similar sentiment to “Requirements are for managers, not engineers”
Hard disagree. Types are for people. They're for people to tell compilers what to tell people. Which is incredibly useful. They are first and foremost a communication and code-navigation/wayfinding tool—for people.
As for all this extra time some folks keep complaining about, we must be using TypeScript entirely differently, somehow. I don't get it at all. Its overhead is less than the time it saves me in typo-spotting alone, let alone all the other benefits. Takes 5%, gives back 20-30%. It's not even close. Maybe it's a problem for really, really slow typists? I'm not even some kind of editor wizard and it hardly slows me down. It can't be thinking up the type definitions since you need to do that anyway—right? I hope? And at that point you may as well write them down.
Their arguments remind me of someone saying they want to use Assembly Language instead of C++, because assemblers are faster than compilers, or to not have to worry about memory heap issues, etc.
All you're doing by switching from TS to JS, is trading one set of minor problems (which are solvable) to a more vast set of even larger problems that are NOT solvable.
https://github.com/Clay-Ferguson/quantizr
I have 5 years experience in TS and 20 in JS. Trust me, if you don't yet understand the value of type-safety, it's only because you haven't lived long enough yet.
If I misspell something, or provide the wrong argument type to a method, or have args in wrong order, etc, it's caught instantly.
With JS you can have a typo that causes user malfunctions for YEARS before being found by a developer and fixed. Any developer who's done many years of both TS and JS would never in a million years consider going back to JS. Absolutely out of the question.
Except, apparently, for the topic of this article.
> If I misspell something, or provide the wrong argument type to a method, or have args in wrong order, etc, it's caught instantly.
Yes, I've heard the pitch before. I've even given it more than a few times.
Of course, if both of your arguments happen to be 64-bit floats, then the type system won't help you much if you get them in the wrong order.
However, when you want to change that function to take a closure instead of a number, the type system can be invisible or it can fight you. Some of us love the structure that a type system can provide. Some of us, like the one who started this website we're on now, also appreciate the power of languages that get out of your way and let you make your own brilliant mistakes. And some of us can appreciate the beauty of different language designs and approaches, and try to rise above dogma. Which, if you've been in this line of work for 20+ years, should be you.
Well, Deno devs did try to use TypeScript, and just ran into some challenges, and decided to throw the baby out with the bath water. I've seen this a lot in my 30yrs exp.: some developer/team runs into problems, gets frustrated, gives up, and decides the best decision is to remove some important piece rather than solve the actual problems.
The reason Deno is giving up TypeScript is representative of failure, and not an indication them wanting JS instead of TS. What they want is TS.
The rest of your reply is just stating some minor imperfections that are a billion miles from invalidating TS. And yeah, there's a place for JavaScript still, for small-scale tinkering, or perhaps even prototypes with only a few hundred lines of code. Any large scale project in JavaScript (not TS) is simply unfathomable to any modern developer worth his salt.
And you dismiss my reply (including the part where I point out that the website we are on was first developed in a Lisp) as "some minor imperfections". It's like you're satirizing the tunnel vision of the fundamentalist.
And, true to form, you "know" what the deno team wants better than they themselves do! I hope you can come back to this thread in a year or two with some perspective.
And about me knowing Deno did want to, and did attempt to use TypeScript: the way I know that is because I read the article, lol. Apparently you did not.
The facts are that they started with TypeScript and moved away (for a particular internal component). The interpretation of that as wanting to use it and failing is yours. You could just as well see it as having made a mistake and then realizing it. But all of these interpretations, by those of us who didn't have to make the choice and weren't there when it was made, aren't very helpful.
[0]: https://docs.google.com/document/d/1_WvwHl7BXUPmoiSeD8G83JmS...
(1) Use a faster bundler like esbuild, here vite's similar considerations [1]
(2) Use Flow [2] and just strip it away before distribution like this Babel module [3]
---
[1] https://github.com/vitejs/vite#typescript
[3] https://babeljs.io/docs/en/babel-plugin-transform-flow-strip...
But for anyone enjoying to specify their types in detail, have fun!
are you working with a team? if so was it easy to get buy-in?
If the request ends up with the query param as an array, there isn't a runtime to validate the cast and now the variable is typed as `string` but is actually `string[]`.
I think a safer approach is to use a validation library or refine the type with an if stmt and returning an error when it isn't a string.
I use the following lint to prevent all casts except `as const`: https://github.com/typescript-eslint/typescript-eslint/blob/...
Finally people are speaking out against the madness of TypeScript.
Don't get me wrong. It is a technical achievement to put a static type system on top of JavaScript. And for libraries, it is useful. It let's the users of the library get hints and direction from an editor like Visual Studio.
But as a language. It is just not good. No way anyone would end up with a design like that had it been built from scratch without going through JavaScript first.
Remember when Java did generics and got flak because of type erasure? Compare with TypeScript ...
(I'm just referring to this nice talk about the regretful mistakes done in Node: https://youtu.be/M3BM9TB-8yA)
As a counter example, VSCode is easily bigger than the Deno codebase. Again, that doesn't mean it's good for your project.
Well, I didn't have a clue what Deno was before, but now it's barely letters in a screen. This post says Deno way too much. Deno.
This is not a language flaw. It is a tooling problem.
TS is the best language I've used. It supports frontend and backend and strict typing and it's scalable so I can refactor and know that my code still works.
The one thing its not good at is ML where Python is better but that's mostly because Python has better library support (not language design).
Here's the deal. When dealing with a large amount of code 80% of your dev time is spent maintaining code - NOT writing new code.
You can save a lot of time now writing new code and not using strict typing. OR you can spend a little bit of time now, maintaining your types, and save a MASSIVE amount of time later.
> - TypeScript compile time when changing files takes several minutes, making continuous compiling an excruciatingly slow process
I have a HUGE typescript repo and using a Macbook Pro from 2015...
With --incremental and --watch my build takes like 5 seconds. About 1 minute on the first build then about 5 seconds after that.
If your build is taking a long time - you're doing something wrong.
> The internal code and runtime TypeScript declarations must be manually kept in sync since the TypeScript Compiler isn’t helpful to generate the d.ts files
What? How? That doesn't make any sense. You're doing something wrong here. If all the code is in typescript, you have no additional work to do. If it's in JS you just have to write your own .d.ts files manually.
Honestly it seems like this team just needs to sit down and understand how their tools actually work.
The one issue where Typescript DOES kind of suck is when dealing with webpack, typemaps, and types.
Many older projects just don't publish types and if you rely on them, and they refuse to publish types, you're going to be in a bit of pain.
You can write your own types though and we've been doing this internally and for some repos just forking them so it's a bit easier to work with them. They usually lag on publishing their types.
If you want to avoid all this pain you can just use webpack for your build system and Typescript for your code. We use lerna to build a multi-module system with --hoist to avoid duplicate dependencies.
https://github.com/burtonator/polar-bookshelf
... is the project we're working on if you want to check it out. Our internal build system isn't fully published yet as we're in the process of reworking it but we're open source so you can take a look if you want.
But I suppose they are aware of this flag.
A sign of the times I guess that a build system requires 8gb of ram to run. Could be we're doing something wrong, but from what I understand we're using a pretty vanilla setup.
Woah, woah, hold on. This case is showing that in the very specific, atypical case the Deno team are using it in, TypeScript is not the right fit.
I’d rather read something that actually interviews the Deno folks about what they are doing, the underlying document is a pretty free ranging discussion and this writeup is making a few assumptions.
N=1. That is all.
(Though I also think TS is a verbose mess made so that enterprise programmers/managers can feel safe, but even if it was great I still think this would have been a good choice)
That being said I run ts-node all the time with transpile-only and that is decently fast. Even on production where the minimal cost is incurred once when loading the module.
As for their problems mentioned on the document. That Header class has a .d.ts conflict on an interface. That could be solved with having namespaces in .d.ts or having I prefix for class interfaces. We do this all the time in our codebase.
I’m not entirely sure why they have two copies of things. The doc doesn’t give much detail.
If you want things to be fast like nodejs, then the other option is to use //@ts-check comments on JS files, or with compiler allowJS and checkJS settings. Webpack checks all their JS.
TSC now also has an option to automatically generate .d.ts from JS files with jsdoc comments. With declarationOnly compiler flag. Many existing JS projects already do that.
So TLDR is. You don’t have to ditch Typescript totally. Typescript can purely just be a checker that you run at CI time and it will work quite well with existing JS code. Although if you want super strict safety the .ts syntax is less verbose and nicer to write. Even then transpile only mode is much faster than running full tsc.
Remember: TS is a superset of JS. All JS is valid typescript. Sometimes we gotta structure the JS properly and that’s where the issue is rather than the typesystem.
Getting closer to the metal seems good!
Would Deno become the new Node?
Do you see a vibrant ecosystem building around Deno?
What I am interested to know primarily is, is the architecture / stack / goals and vision that resulted in Deno viable in the long term?
Personally, it makes me cringe to see hard-coded HTTP URLs in source code files for loading dependencies. They present it like we're stupid for using package managers all this time. What used to be a simple command line option (changing a package repo) is now a thousand file edit. Excellent.
> TypeScript compile time when changing files takes several minutes, making continuous compiling an excruciatingly slow process
Umm, you can compile single files, and serve them individually. It's no different from bundling (e.g. with webpack) Javascript.
> The Typescript structure that they’re using in the source files that create the actual Deno executable and the user-facing APIs is creating runtime performance problems
Then use a better structure?
? TypeScript isn’t proving itself helpful to organize Deno code. On the contrary, the Deno team is experiencing the opposite effect. One of the issues mentioned is that they ended up with duplicate independent Body classes in two locations https://github.com/denoland/deno/issues/4748
Then consolidate the two classes...? How is the the language's fault?
> The internal code and runtime TypeScript declarations must be manually kept in sync since the TypeScript Compiler isn’t helpful to generate the d.ts files
Well, it can be. Fix your build rules.
> They’re maintaining two TS compiler hosts: one for the internal Deno code and another other for external user code even though both have a similar goal
Again, not the language's fault.
---
We've been using Typescript for a couple of years now, and it's been great. By trying to do things the right way, it actually forces us to organize our code better. I think the Deno project would benefit from refactoring their code to solve their issues, instead of blaming the tools.
I think that's the whole point. It's not the right tool for their specific goal.
In their design doc[0], they agree that Rust glue code should not be written in TS: compiled code "full of weird namespaces."
From the design doc discussion[0]:
ry: manually managing the d.ts files has been great - it's very much part of the public interface - we need to have 100% control over it we do not want random compiler-generated __namespace0123 in there This is the same situation - but for the bundle It's not good that we do not have visibility nor control over what code is in there. for example, the TS compiler worker, has 15k lines of code in its bundle - EXCLUDING typescript.js TS compiler worker is about 500 line thing that interfaces between ops and TSC this is caused by our attempt to "self host" the internal code self hosting is cute - but if it's effecting performance and the ability to make improvements - it's an unnecessary feature https://gist.github.com/ry/a6d4b1466158d82750a6447342fd3af4 ^--- here is what we ship for the compiler worker we call this code every time deno runs
lucacasonato: Ok, wow. I hadn't seen this. That changes things...
ry: does the compiler worker really need to generate a UUID? https://gist.github.com/ry/a6d4b1466158d82750a6447342fd3af4#... (no) yes, this is fixable within the current system .. but the point is that we're abstracting away the important parts V8 does not run typescript. It runs JS. We need to have a good handle on the JS we ship If we don't try to self-host, we don't need a sourcemap for the internal code. There's 1mb off the executable right there. The point is the internal typescript comes at a cost - and that cost is too high.
lucacasonato: So the actual issue is the bundling and opaqueness that that brings with it. It makes sense to stick to pure JS in that case. Its annoying that type checking is gone, but youre right. This is not the kind of code that should be in the runtime.
ry: not like the code isn't full of "!" anyway. TS is great, but it gives a false sense of security.
[0]: https://docs.google.com/document/d/1_WvwHl7BXUPmoiSeD8G83JmS...
[edit: not trying to take a stance in support of the GPS post or either side of this discussion, to be clear]
Taking up Typescript then ignoring best practices when trying to use it is unfortunate. This kind of thing would be a teaching moment for a junior engineer.
It's like saying you don't want to use a dishwasher because if you stack bowls on top of each other, it doesn't clean them well. Well, a dishwasher has its faults - a lot of dishes are not dishwasher-safe, etc. - but your reason kind of misses the point.
"TypeScript compile time when changing files takes several minutes, making continuous compiling an excruciatingly slow process"
and
"The internal code and runtime TypeScript declarations must be manually kept in sync since the TypeScript Compiler isn’t helpful to generate the d.ts files"
If you're compiling Typescript, the .d.ts files are output by the compiler. The only time you need to manually keep them in sync is when you're writing .js and manually writing the .d.ts.
Aside from the compiling takes time issue, the rest sounds like trolling.
If the code was organized in a way which is inefficient and confusing, I fail to see why checking types at runtime instead of compile types improves the structure. It’s a completely separate problem.
I will refuse deno. Too much ego in JavaScript-land. “Everybody wants to be like Mike”. Errr Linus... Lowrey
They must be doing something horribly wrong, because tsc takes literally 1 second for my 3KLOC codebase[1] in incremental mode.
EDIT: I didn't read the article thoroughly. It looks like they're still supporting Typescript user code.
> Most people don't have the context to understand this narrow technical document - it is only applicable to a very particular, very technical situation in the internals of Deno. This is not at all a reflection on the usefulness of TypeScript in general. It's not a discussion about any publicly visible interface in Deno. Deno, of course, will support TypeScript forever. A website or server written in TypeScript is a very very different type of program than Deno - maybe much more so than novice programmers can appreciate - little of Deno is written in TypeScript. The target audience is the 5 to 10 people who work on this particular internal system. Please don't draw any broader conclusions.
(Almost) all npm packages that are written in typescript expose .js and then .d.ts files, so you can't "just import the .ts". This is because of huge variances in tsconfig.json files.
Why add complexity to a new project which starts from scratch? JavaScript is good enough and there's no need to make it confusing by adding another layer.
To some people, JS itself is relatively confusing, and TypeScript reduces that confusion.
EDIT: And... someone's clarification is saying this is internal framework code, not userland support for TS. Seems like a non-issue, really. They don't want to use it, but want to provide support for others to use it in their projects.
This doesn't change anything, it's for their internals. Why anyone would ditch static typing on a large collaborative project is beyond me though.