The main problem with JavaScript and TypeScript is that there is such a little entrybarrier to it, that way too many people use it without understanding it. The amount of times that we had major issues in production because someone doesn’t understand TypeScript is not countable anymore and our project went live only 4 months ago.
For example, when you use nest.js and want to use a boolean value as a query parameter.
You see this code. You don’t see anything wrong with it. The architect looks at it in code review and doesn’t see anything wrong with it. But then you do a GET https://something.com/valueOfMyBoolean?myBoolean=false and you get “myBoolean is true” and if you do typeOf(myBoolean) you will see that, despite you declaring it twice, myBoolean is not a boolean but a string. But when running the unit-tests, myBoolean is a boolean.
This is more a condemnation of nest.js than ts. It seems great in theory. I like the architecture and the ability to share models and interfaces between front and backend, but it’s objectively makes everything more complicated. It adds layers of abstraction that should not be necessary and it’s such a niche/unpopular framework for backend systems that you generally have to jump through hoops to do anything moderately complex. Not only do new devs have to learn typescript to use it, they have to learn the nest architecture to know how to do things “the right way” and you still end up in situations like this which looks perfectly valid but isn’t. Typescript was never meant to be used for backend, and trying to make it do so and then complaining about it is like jogging while carrying a gun, shooting yourself in the foot, and blaming the gun.
I’ve never used TS, and I’m not exactly sure what nest.js even does, but building a TypeScript project on top of a JavaScript library not designed for it seems like asking for trouble. Is that standard practice?
If I really needed to use a JS library in TS, I’d have to build some sort of adapter between the two that crashes whenever the JS library (that doesn’t know anything about your types) breaks the typing rules. Anything else will inevitably lead to the above “fun” kind of bugs.
I don’t think that this would work, there are no types anymore during runtime because everything is translated into plain js on build. TypeScript only exists during development
It would be dependent on what I was doing exactly, but a lot of times I’d end up reimplementing the type system in the runtime code, basically. Maybe I’d test the data every time it returns from the library, which would be slow, but if performance is an issue and there’s still no TS options I should really just be making my own libraries.
I really don’t want to deal with the above kinds of bugs, yo. Bugs that are actually visible in the code are hard enough. Maybe I’d have to do it but I’d kick and scream the whole way there.
Honestly, programming is great for teaching you that you are the stupid one. This is still a feature.
The main problem with JavaScript and TypeScript is that there is such a little entrybarrier to it, that way too many people use it without understanding it. The amount of times that we had major issues in production because someone doesn’t understand TypeScript is not countable anymore and our project went live only 4 months ago.
For example, when you use nest.js and want to use a boolean value as a query parameter.
As an example:
@Get('valueOfMyBoolean') @ApiQuery( { name: 'myBoolean', type: boolean, } ) myBooleanFunction( @Query('myBoolean') myBoolean: boolean ){ if(myBoolean){ return 'myBoolean is true'; } return 'myBoolean is false'; }You see this code. You don’t see anything wrong with it. The architect looks at it in code review and doesn’t see anything wrong with it. But then you do a
GET https://something.com/valueOfMyBoolean?myBoolean=falseand you get “myBoolean is true” and if you do typeOf(myBoolean) you will see that, despite you declaring it twice, myBoolean is not a boolean but a string. But when running the unit-tests, myBoolean is a boolean.This is more a condemnation of nest.js than ts. It seems great in theory. I like the architecture and the ability to share models and interfaces between front and backend, but it’s objectively makes everything more complicated. It adds layers of abstraction that should not be necessary and it’s such a niche/unpopular framework for backend systems that you generally have to jump through hoops to do anything moderately complex. Not only do new devs have to learn typescript to use it, they have to learn the nest architecture to know how to do things “the right way” and you still end up in situations like this which looks perfectly valid but isn’t. Typescript was never meant to be used for backend, and trying to make it do so and then complaining about it is like jogging while carrying a gun, shooting yourself in the foot, and blaming the gun.
On the other hand, this can be considered a downside because it locks you into using JS/TS on the front and backends.
Alternatively, if you define your models and interface with an Open API spec, you can write the front and backends in whatever language you want.
I’ve never used TS, and I’m not exactly sure what nest.js even does, but building a TypeScript project on top of a JavaScript library not designed for it seems like asking for trouble. Is that standard practice?
Yes. As of this writing there are 7,738 type definitions in a central repo maintained by users for plain JavaScript packages.
https://github.com/DefinitelyTyped/DefinitelyTyped
Many package owners write type definitions included with their package that is written in JavaScript also.
Web dev continues to be cursed, I guess.
If I really needed to use a JS library in TS, I’d have to build some sort of adapter between the two that crashes whenever the JS library (that doesn’t know anything about your types) breaks the typing rules. Anything else will inevitably lead to the above “fun” kind of bugs.
I don’t think that this would work, there are no types anymore during runtime because everything is translated into plain js on build. TypeScript only exists during development
It would be dependent on what I was doing exactly, but a lot of times I’d end up reimplementing the type system in the runtime code, basically. Maybe I’d test the data every time it returns from the library, which would be slow, but if performance is an issue and there’s still no TS options I should really just be making my own libraries.
I really don’t want to deal with the above kinds of bugs, yo. Bugs that are actually visible in the code are hard enough. Maybe I’d have to do it but I’d kick and scream the whole way there.
I instantly noticed this line. Shitcode is so fun.
Typically when creating API interfaces you’d be better off marking the inputs as unknown, and then using something like Zod to validate the types