ChatGPT解决这个技术问题 Extra ChatGPT

How do I decide whether @types/* goes into `dependencies` or `devDependencies`?

I use TypeScript 2 in my project. I'd like to use some js library, but also typings for that library. I can install types with simple npm install @types/some-library. I'm not sure if I should --save or --save-dev them. It seems to me that even DefinetelyTyped GitHub readme kind of mentions both versions, but never explains them. I would think that @types should be in devDependencies, as types are needed for development and aren't used in runtime, but I saw many times @types in just dependencies. I'm confused.

How should I decide whether @types/* goes into dependencies or devDependencies? Are there actually some more or less official instructions?

Are you generating a bundle or is this a package that will be used by others? As I see it you only need to make the distinction between dependencies and devDependencies in the latter case.
I make some game in js/ts from scratch. I bundle everything with webpack. There's no backend at all atm, but it's possible that I'll wrap it all in Electron to make it standalone some day. I don't think anyone will ever use it as a dependency in their own app, but I guess it could be possible (think of mini games in GTA games; and my game is open source). Still, I want to learn and follow best practices and it's the main reason I make that game. I hope I clarified my use-case well enough. :)
Yes, it makes sense, just wanted to make sure that my original answer was relevant to your use case. I still think that the distinction between devDependencies and dependencies is irrelevant when building a bundle, it's something that create-react-app enforces as well but ultimately it's up to you to choose

A
Anton Rudeshko

Let's say you're developing a package "A" that have @types/some-module package in devDependencies. For some reason you're exporting the type from @types/some-module:

import { SomeType } from 'some-module';

export default class APackageClass {
  constructor(private config: SomeType) {
    // …       
  }
}

Right now TypeScript consumers of package "A" are unable to guess what SomeType is, since devDependencies of package "A" are not installed.

In that particular case you need to place @types/* package with regular dependencies. For other cases devDependencies are good enough.


So you imply that, if I only use the type in implementation, it’s type definition can be devDependencies?
Yes @FranklinYu . As soon as the type appears in declaration file, you need to place it on dependencies. Otherwise devDependencies is fine
But a package works for both TS and JS. JS developers doesn't need those types to compile their code. Adding the type definition to dependencies will make dependency tree bloated.
@TylerLong Correct. It's not perfect but that's reality. Optionally You can also use "optionalDependencies" but I believe at scale it might be very annoying.
I suppose that in case I'm developing an end-user application and not a package I can move all the @types/.. into the devDependencies, correct?
V
Valentin

If you're just generating a bundle there may be no need to make the distinction between dependencies and devDependencies. This feature of npm is generally useful when publishing a package that can be used by others and you don't want to spam them with redundant dependencies.

There may be other use cases where splitting dependencies can be helpful but unless you have an express need for this, then my advice is to just pick either one and place everything there. It's not difficult to split them afterwards if the need should arise.

A well-known example of this practice IRL is create-react-app, by default the un-ejected boilerplate it creates places everything in dependencies, see this thread and this answer


If you're not publishing the package, that's correct, but if you are, it has nothing to do with development vs. runtime and everything with what's needed to build this package vs. what's needed to use this package.
@Yogu That's why I made the distinction in the first place so yes, I completely agree with you
I disagree with this advice. devDependencies are not installed when you do npm install --production (or npm ci --production) and thus not available when running production code. This is a very meaningful difference for a service, not just a library.
@BradWilson You have a point, there are many npm workflows under the sun, if your use case requires you to make the distinction then by all means, do it. Feel free to provide your own answer to this dilemma.
I've updated my answer to mention the existence of other use cases where the distinction may be meaningful and gave real world examples. Thanks for the feedback!
C
Carsten Führmann

In the particular case of deploying a Node.js application to production, one wants to install only the dependencies needed to run the application.

npm install --production or

npm ci --production or

yarn --production

In that case, the types should be in the devDependencies, to keep them from bloating the installation.

(To avoid misunderstandings, the --production option must not be used on machines where the application is built, otherwise the TypeScript compiler will complain.)

Remarks: I'm aware this was mentioned in a comment by Brad Wilson to another answer. This point seems worthy to be an answer, though.


I am developing an app using 3rd party libs, put all of the @types/* packages in devDependencies, but then production build fails as typescript is looking for these types in the production build.
@omerts: When you wrote "looking for these types in the production build", did you mean "looking for these types in the package.json during compilation"? I'm asking because a production build is what's run on the target machine, and it would be strange if the TypeScript compiler were invoked at that stage. If on the other hand you were talking about compilation on a developer machine, the devDependencies should still exist at that stage, because --production should only be used when installing for production - not on the developer machine.
When running yarn/npm install with the production flag on, and then yarn build, either in a CI or docker env, the typescript compiler will fail due to missing types.
The issue seems to be that you use the --production option already on machines where the application is built. Instead, the option should be used only on machines where the application is run, not built. I clarified my main answer in that regard.
@omerts While it is sometimes possible to copy the node modules from the build machine the hosts (i.e. the machines that run the application), this seems like a questionable practice, because: Even if the Node.js version number on the hosts is identical to that of the build machine, the target machines might be a different OS. For example, you may develop on Windows, but run things on Linux. And there are cases where e.g. the output of npm install on Windows is not correct on Linux.
Z
Zim

Other answers made great sense, but I'm gonna add that a peerDep's type declaration package should also be placed in dependencies instead of peerDependencies.

Assume that b is a plugin of a. And c uses a and b.

Why shouldn't @types/a be placed in b's peerDependencies?

If b's package.json is like:

{
  "peerDependencies": {
    "a": "1.5.x"
    "@types/a": "1.4.x"
  }
}

c may use only interfaces defined in @types/a@1.2.x but c is forced to install @types/a@1.4.x.

Furthermore, c may be a regular javascript package rather than typescript package, but c is also forced to install @types/a@1.4.x.

So, the correct package.json of b should be like:

{
  "peerDependencies": {
    "a": "1.5.2"
  },
  "dependencies": {
    "@types/a": "1.4.x"
  }
}