We switched our project to Typescript about a year ago, because we had to handle a lot of different types of data, and it was nice to know which type you have in your hands. I wrote a bunch of code to parse anonymous JSON objects and turn them into objects of the proper types. We added interfaces for all sorts of things.
End result: checks don't work, because all our data is run-time. Most of our parameters are still of type `any`. We need to explicitly specify types everywhere that are either `any`, or they're not actually checked. I discovered recently that someone had accidentally created a type that was conceptually the duplicate of an existing type. One of those should never ever get used, but the code doesn't complain, because it's only encountering this in runtime.
So I'm inclined to agree: Typescript doesn't really add what it's supposed to add. It adds a lot of work, without providing the security we expect from it.
I think you used the wrong tool or misunderstood how to utilize typescript. If you want guarantees of data integrity from external sources, try something like Json Schema.
Or even without that, your data should be entering the system as type 'unknown'. There's then a validator function that takes in unknown and outputs the intended type.
From there onwards, integrity is guaranteed by typescript, but only if you have the discipline to use it consistently. Otherwise, yeah, don't bother.
The second one is what I did: I've got a function that I can throw any of my anonymous data objects into, and it turns it into a typed object that knows its type and has the relevant class methods. And this worked fine with Typescript as long as nobody accidentally introduced a wrong class. Or actually it kept working fine after that, because the system simply ignored the types during runtime, so nobody noticed there was one bad class in the mix.
Anyway, we're going back to simple JSON objects. Typescript is clearly not helping in our case, and the whole process with parsing all incoming objects is fairly cumbersome without adding much value.
“io-ts introduces the concept of a runtime type (an instance of the Type class) which represents a runtime validator for a typescript (static) type. [2]”
I work with Haskell, a language that some people call "pinnacle of type systems". Guess what would happen if I use `unsafeCoerce`? Exactly the same that would happen if I use `<any>` in Typescript. All the type guarantees go out of the window.
In your code, if someone throws a bad class into the mix, it will blow up both in Typescript and in Java/C# and in any other programming language. But, somehow, blowing up in Typescript is way worse that blowing up in Java/C#. Go and tell me that when I'm not allowed to put `nulls` anywhere neither in Java nor in C#, because that's something I can do in Typescript today.
Being able to ban null values is absolutely fantastic. That's definitely a big improvement over Java, and seeing it in Typescript makes me wonder why Java allows it. In fact, seeing Option in Scala made me wonder why anyone ever thought null was a good idea.
Even if it's only a compile time check, and can therefore still be circumvented at runtime, this check for null values has still helped me out a lot. At least within a class it ensures my assumptions are consistent, which is not nothing.
>this worked fine with Typescript as long as nobody accidentally introduced a wrong class
The problem with receiving a wrong class is not intrinsic to any particular type system:
A *a = reinterpret_cast<A*> &a;
// have a nice gdb day
C++ doesn’t control types at runtime too, it is just bytes (runtime data) at precompiled offsets. Either you cook types wrong, or that data is designed as a set of untagged unions that get in the way of any typing, forcing to any-cast everywhere. <any> is for unknown pojos, not for already-typed values. And even if Foo is passed to makeBar(arg:any):Bar, it should treat it as pojo and still emit Bar as a result, not just ‘arg’ as it is. All typed languages do exactly the same for incoming binary data, the only difference is that you have pojos instead of byte-ptrs.
>Typescript doesn't really add what it's supposed to add.
Underspecified/force-casted types couldn’t help with these promises. It is like programming in C with a heavy preference to void pointers and va_lists.
I don’t see how having at least some types is more risky than not having them at all. Type system is an instrument that prevents you from e.g. using/setting x.id instead of x.y.id silently and succesfully (for some time). The entire argument is stretched on the particular situation of a malformed input.
I think it is well-known fact among programmers that a prototyping phase of 10kloc+ project requires types. Not for a formal proof, but to help programmers to understand their code better and faster at monday morning through a type error feedback. There is no reason to not use ts the same way there is no reason to not have a ruler and yesterday’s sketches in your toolbox.
Everyone may have their opinion, but this modern “media presence in software development” nonsense goes beyond all limits.
Watching the talk "Typing the Untyped: Soundness in Gradual Type Systems" by Ben Weissmann (https://www.youtube.com/watch?v=uJHD2xyv7xo) gave me an appreciation for the challenges in designing in a gradual type system like TypeScript.
When I first started using TypeScript, I was surprised to find that it wouldn't compile code I absolutely knew was correct, or that it would allow "risky" code in some cases as the author says. The talk formalized the issues I faced using TypeScript by teaching me about soundness, completeness, and the trade-offs that are often taken between the two.
Generally, TypeScript has made my life a lot easier than the occasional trouble it's brought me.
We have a few TypeScript projects where I work at, one of them in particular had to ingest data from a large number of our internal APIs and do some crunching with it.
Validation of this data was essential, the crunching was complicated, and we wanted to be able to make assumptions about the data through the type system to make the code simpler.
We found JSONSchema validation was a perfect fit for this purpose (via ajv). In all the overhead of maintaining JSONSchema files was much lower than we thought it would be, and it even exposed some bugs in our internal APIs that we patched up at the same time.
I do wish there was a nice way to generate that sort of validation code directly from a TypeScript interface though.
I think the point about arbitrary data entering your system once everything is compiled is quite valid: it is common to call into typescript-compiled code from arbitrary JavaScript. Typed Racket had an interesting solution to this, where it would generate run-time contract checks at all of the typed-to-untyped boundaries, and so it could be sure that a compiled typed function was indeed receiving objects of the correct form.
It is not a platform language. Thus just like any guest language brings its own tooling, idiomatic library wrappers around platform libraries, an additional code generation layer to debug, having to write FFI (or type declarations to platform libs).
Having said this, I do use TypeScript in the context of Angular or BabylonJS, but that is because those frameworks are the "platform".
Yes, sort of. I don't bother with it since I just develop demos.
Anyways, it doesn't really matter that much given source line annotations. I wish TypeScript went with some more aggressive sugar so we could have nice things like operator overloading.