The last days of my type checker for JavaScript(jsmonk.github.io) |
The last days of my type checker for JavaScript(jsmonk.github.io) |
But I don't think I agree with the authors conclusions that there's less interest in type safety or strict type checking. I suspect the author finds Typescript too leanient (there's plenty of ways to trigger Typescript's unsoundness) for their liking, but I do think there is an overall trend to use Typescript to design and build more reliable software.
Do not forget you learned a lot along the path. That's something no one can take away from you.
> unfortunately, I see that nowadays, TypeScript won. The idea of strict type-checking for JavaScript is less popular than in 2019.
> [Typescript's types] give people a fantastic tool to cover existed code with types but doesn’t force to re-design it to make it safer.
Many other languages haven't passed the PoC stage. Several are ready, and doable (C#, Go?) but hardly as easy and common as with rust.
At least that was the state when I looked six months ago. It's a fast moving area.
So, yes, that time is now. But the only practical option is rust. Which may be enough, idk.
My only reference is a memory of this talk by Richard Feldman but I might be wrong or misremembering. https://www.youtube.com/watch?v=Tml94je2edk
I’d be shocked to find this to be true. Back then I’d always wonder, “are there typings for this library?” Now I don’t have to.
I'm unconvinced this would be a good thing.
I was also happy to see the shout-out to Ben's Ezno. I'd recommend people start sponsoring that kind of work, it has the potential to make our JS life much saner.
I honestly don’t get the problem TypeScript is trying to solve. JavaScript has lots of wacky problems but its main advantage to me was the loosy goosy approach to types. I was a smalltalk programmer back in the 1990s and it felt pretty comfortable if very primitive in the way of tools.
Typescript just seems unnecessary. Not sure why nobody built something like reentrent exceptions or cool stuff like that, instead they built a stupid jail that down compiles to unreadable sludge.
Now I find out there were competing jails. What a strange ecosystem.
There is a lot of text written and voice recorded over what type safety is good for. It's reasonable to disagree, but to "honestly don't get" what the point of it is seems kinda strange. Like you haven't really looked into it.
It's quite annoying IMO. Some of the things it can be infer are reasonably impressive but then that gets balanced out with having to write out a lot of crap that something a bit more HM-like would be able to nail immediately (rather than just "Any!? Fuck you").
That and the refusal to add other value-adds (despite being a compiler in all but name) hobbles the union types and so on.
Smalltalk is type safe & not statically checked.
page 263 "Programming Languages: Application and Interpretation"
[pdf] https://cs.brown.edu/~sk/Publications/Books/ProgLangs/2007-0...
It's generally better to do proper micro-services.
Type safety is the inferior way of doing it where you get hit with a double whammy of long compile times and extreme code verbosity.
> instead they built a stupid jail that down compiles to unreadable sludge
Does it? The Typescript compiler mostly just erases types, leaving the code otherwise the same. Here's a non-trival example https://tsplay.dev/mA078N, where the output is basically the same as the input, just with type syntax erased.
IIRC the only 'language feature' that Typescript actually compiles is enums, which is a bit of a big blob. There's compiler flags to 'fix' that, or you can just not use enums.
Smalltalk (and languages in the same academic orbit, like LISP derivatives) seems to be a tool that fits a few people's hands very well, but there are no large smalltalk projects for two reasons:
- that doesn't actually suit most people, so it's harder to find developers
- the language maps well to one (1) human brain that keeps a model of what's happening, but that breaks down when larger teams of developers are involved.
I guess that keeps the "small" in smalltalk.
Seems like we need do no more than assert: Yes there are!
~
But it's more interesting to look at how a real Smalltalk project worked:
"A very large Smalltalk application was developed at Cargill to support the operation of grain elevators and the associated commodity trading activities. The Smalltalk client application has 385 windows and over 5,000 classes. About 2,000 classes in this application interacted with an early (circa 1993) data access framework. The framework dynamically performed a mapping of object attributes to data table columns.
Analysis showed that although dynamic look up consumed 40% of the client execution time, it was unnecessary.
A new data layer interface was developed that required the business class to provide the object attribute to column mapping in an explicitly coded method. Testing showed that this interface was orders of magnitude faster. The issue was how to change the 2,100 business class users of the data layer.
A large application under development cannot freeze code while a transformation of an interface is constructed and tested. We had to construct and test the transformations in a parallel branch of the code repository from the main development stream. When the transformation was fully tested, then it was applied to the main code stream in a single operation.
Less than 35 bugs were found in the 17,100 changes. All of the bugs were quickly resolved in a three-week period.
If the changes were done manually we estimate that it would have taken 8,500 hours, compared with 235 hours to develop the transformation rules.
The task was completed in 3% of the expected time by using Rewrite Rules. This is an improvement by a factor of 36."
from “Transformation of an application data layer” Will Loew-Blosser OOPSLA 2002
But mostly because it was expensive. Objective-C and Ruby got you halfway there for free, and there is nothing developers love more than something that is free.
Someone properly trained with something like Smalltalk will on average output 3x the amount of software as someone trained on something like Java.
There are big advantages to doing it.
I've worked with both Javascript and Typescript and the ease of patching up Typescript projects you're not familiar with is unmatched compared to Javascript. At least when my code compiles, I'm sure the plumbing is correct
It runs in the browser. That’s always been its only advantage. Even now that it’s much better than it used to be: still true.
> Not sure why nobody built something like reentrent exceptions or cool stuff like that, instead they built a stupid jail that down compiles to unreadable sludge.
TIL machine-readable and -verifiable versions of the documentation you should be writing anyway (but probably aren’t) is a jail.
Also one of the reasons I didn’t worry about it going the way of other compile-to-JS languages is that it’s output is nearly identical to what I would have written anyway, so it’s got a clean and easy “escape hatch”. But maybe I write unreadable sludge.
(_c = this.log) === null || _c === void 0 ? void 0 : _c.withIndent(' ', () => {
var _a;
// (pointless ?. silence eslint)
(_a = this.log) === null || _a === void 0 ? void 0 : _a.pn(`Buffer address: 0x${utils.hex8(bufferAddress)} `);
});
this.log?.withIndent(' ', () => {
// (pointless ?. silence eslint)
this.log?.pn(`Buffer address: 0x${utils.hex8(bufferAddress)} `);
});
At my workplace a team of test engineers wrote an entire little test framework around our application. It‘s using webdriver.io under the hood but defines all test classes in type script. Everytime I need to inspect a failing test I open it up in my Emacs and jump around the test method definitions quickly and easily and enjoy inline doc strings popping up wherever I set my cursor on. So the task of fixing a test becomes quite easy despite not having written the framework. Back when we used JS or Perl this task used to be quite obscure.
On the other hand when I see react applications written in typescript I become quite skeptical of the supposed architectural benefits of static typing in front end logic. While React applications were already difficult to grasp, static typing constraint make refactorings even more headachy and I receive Java‘s spring framework vibes constantly.
+1: I'm much slower working in environments where I have to do this manually. This is why I've been infiltrating Python type annotations in our codebase.
If you've only done small projects or projects where you are the primary author then you may not run into the issues. You essentially end up doing what Typescript does but in your head without writing anything down.
The problems it solves are:
* Bugs! It catches lots of bugs.
* Navigating code efficiently. You can just click stuff to go to its definition or uses.
* Refactoring code reliably. You can now rename variables without making mistakes.
* Reading code. It can be extremely difficult to understand code without static types because you don't know what anything is or where anything goes unless you run it (which can be very difficult).
Not Typescript, but as an example I was trying to understand some code in Gitlab like `auto_cancel_from_rules.merge(auto_cancel_from_config)`. "what does `merge` do?" I wondered.
Literally could not find it. You can't search for `merge`, way too many results. I don't know what the types of the variables are so I can't look it up. I spent about 15 minutes looking around and eventually gave up.
With static types you can literally just click it and it will take you to the definition.
Dynamic typing doesn't have to get that messy, but it takes discipline that is very hard to enforce at scale. Static typing can sort of approximate that level of mess but it takes a lot more work, and perhaps even more importantly, it's a lot safer to fix. But every significantly-sized dynamic scripting code base I've encountered has certainly had at least one of those two issues deeply, deeply ground into it.
Unlikely. Maybe if you are using a language with a complete type system (let's face it, you are not), but otherwise you are necessarily going to have to encode all the type information into your tests anyway, so you haven't gained anything on this front. A partial type system cannot replace these tests.
Your other points are the real wins, though.
To me it's one of the top requirements for a modern language.
Edit: Also you can follow typedefs of a library and make sense of it without relying on its documentation quality or deciphering the code.
If you're building something simple, and the type/shape of function parameters is either very obvious or very well-guarded, TypeScript's returns diminish.
However:
As a reviewer of complex code, I want to see annotations so that I can better understand what I'm reviewing.
As an author of complex code, I want to clearly communicate the intent of functions I write, so that they are not misused by other developers.
As someone who supports the use of my software, I want to reduce errors by guaranteeing the correct types/shapes of function arguments.
...and I'm just getting started. These are all very compelling needs in my experience.
I think there might be a couple of other tools around in your setup (that might influence your feelings). TypeScript mostly just strips out type annotations, so the output looks pretty close to your input.
Wouldn't it be more accurate to say that WASM hasn't passed the PoC stage? There are quite a few languages that have great WASM support, to the extent WASM allows, but have to bring all kids of workarounds to deal with the shortcomings of WASM as it sits. An especially challenging problem as it relates to browser use with those workarounds is that they bloat the artifact size. Few are willing to subject users to multi-megabyte downloads.
Work is being done on WASM to address those shortcomings, but hands are pretty much tied until that arrives. Until then, the "lower level" and heavily constrained languages will be left to stand alone.
Network communications are the most tedious and annoying part of programming for me because of the validation of the request body and checking that it has all the fields needed.
Microservices only increase the amount of documentation needed. I would prefer to lessen the amount of documentation by letting the compiler/code do the work instead.
It's like British tanks in WW2, your system is composed out of standard parts. Not necessary the best, but standard parts and that's what tends to be important.
The industry spends billions of dollars on developers writing Java. If they could spend 3 times less by simply training developers to use Smalltalk instead, they'd be doing it.
They're not, because that's not the problem.
No they wouldn't. While not bad long-term thinking, by the time developers are up to speed with Smalltalk and realizing three times the output, the competition using the ecosystem its developers already knew will have launched something and won over the customers. It doesn't matter how much software you can crank out after you've already lost.
[edit] oh no wait that’s in newer JS versions, you’re just “compiling” to a version that doesn’t support it. Update your target to a version that supports the feature you used and it should be fine.
(The compile target in my case is ES2017, as configured in August 2018. I can probably update that.)
Dart 2.0 implemented a new sound type system
Dart's type system, like the type systems in Java and C#, is sound.
My question was different though and I was asking (quoting comments further up the chain) whether "google's original plan with Dart" really was to allow for a programming language "which can be properly type-safe from the beginning".
It seems that claim about the original intent is false. I checked Wikipedia [0] which states "Dart 2.0 was released in August 2018 with language changes including a type system" and the reference they link seems to corroborate that, before 2.0, Dart didn't have a sound type system.
Thanks for prompting me to dig deeper (I probably wouldn't have done it without seeing your comment).
[0] Quote found under the history section. https://en.m.wikipedia.org/wiki/Dart_(programming_language)
It's literally impossible to write tests to find all bugs. Most tests don't come remotely close to that, so there's always going to be bugs that static type checks find that tests don't.
There's a great paper that shows that even after tests, Typescript would catch 15% of Javascript bugs: https://earlbarr.com/publications/typestudy.pdf
Also the fact static typing eliminates entire classes of bugs means you need to write far fewer tests. I've seen Python tests that literally fed different types into functions and verified that they accepted them. What a misguided waste of effort!
It is always telling when someone starts getting worked up about something, especially when that something wasn't even said or implied.
> Also the fact static typing eliminates entire classes of bugs means you need to write far fewer tests.
You don't need to write any more tests in the absence of static types. The purpose of testing is not to act as a replacement for static types and if you find yourself writing tests just for the sake of testing types, you know you're doing something horribly wrong. But, you necessarily have to encode type information into the tests in order of them to execute. After all, if that wasn't the case, you wouldn't even be able to catch bugs with a type system. As such, you gain that indirectly.
> I've seen Python tests that literally fed different types into functions and verified that they accepted them. What a misguided waste of effort!
Sure, I've seen developers do all kinds of stupid things too. In fact, give them Typescript and they will just litter the code with `any` everywhere – something I've witnessed far too often. There is no technical solution to bad developers. Was there supposed to be some meaningful takeaway here?
To achieve the same level of quality, yes you absolutely do. This is actually a fairly fundamental fact about static typing. Static typing is a weak form of formal verification, or equivalently formal verification is just really really strong static typing. Clearly you don't need to write as many tests for formally verified code. The same is true (to a lesser extent) for "ordinary" static typing.
> Sure, I've seen developers do all kinds of stupid things too.
It's not stupid if you have a static typing system available! The stupidity was relying on tests to verify types, instead of ... you know, the thing whose whole purpose is to verify types.
And as I said in the other comment, tests won't catch all of the mistakes that static typing will catch anyway.
If you have a complete type system. But as we are specifically talking about partial type systems, you have to already cover all your bases for where the type system is lacking, and by virtue of that you are going to overlap with what the type system also covers.
> It's not stupid if you have a static typing system available!
There is no reason for it, static type system or not.
Nope. You don't need a Turing complete type system.
Anything to avoid the actual topic at hand, I guess?
The partial type systems found in languages people actually use, though, only constrain the most glaring of mistakes that you are almost never going to make. Not impossible that you could make one, but highly unlikely. Certainly not on the order of "lots". More like once in a blue moon.
Refactoring is a different story. Having an editor that lights up like a Christmas tree when code changes breaks existing relationships is a huge boon, and with good refactoring tools you almost never have to break the relationships in the first place. This is where (partial) static typing truly shines.
Not true. People make these mistakes all the time.
I was literally reviewing code today where someone write Python code to write a field that didn't exist. They didn't realise because they hadn't written a test. They asked me to check if it fixed my issue, which wasted 5 minutes of my time checking, when an IDE with static types would have told them their mistake instantly.
You can pull the "well I don't make mistakes" classic, but it doesn't matter. People do make these mistakes.
I linked an empirical study showing that 15% of bugs are precisely these mistakes. If you don't think it's worth eliminating 15% of bugs then frankly I just think you have crazy priorities and I can't really fix that.
If someone hates future developers so much that they wouldn't write a test, why would they suddenly take care to write proper types? I've worked with those kind of people before. The reality is that they won't define any specific types either, relying on `any` (or whatever the language's equivalent is) to minimize the social footprint.
> They asked me to check if it fixed my issue
I get you work with an asshole who hates other developers, but why would they need to ask if it solves your issue when they can just run the tests you wrote to assert the issue you discovered? Well, okay, being an asshole they maybe just wanted to see what meaningless busywork they could force you to do, but then why didn't you at least run your tests rather than spending minutes hunting up and down for something that your test would have found instantly? Something doesn't quite add up here.
> when an IDE with static types would have told them their mistake instantly.
And most likely even an IDE with type inference, a la the original article. Did you even apply a type checker when faced with this problem? It would be quite ironic if you didn't.
> when an IDE with static types would have told them their mistake instantly.
As would the test, at least for any developers who don't have an intense hatred for other developers.