microsoft / TypeScript

TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
https://www.typescriptlang.org
Apache License 2.0
101.31k stars 12.53k forks source link

disable certain global types for specific files, or specify type roots for specific files. #37053

Open trusktr opened 4 years ago

trusktr commented 4 years ago

This continues some ideas from https://github.com/microsoft/TypeScript/issues/17042, which was closed by @microsoft (for inactivity, I think?).

Search Terms

disable certain global types for specific files, specific type roots for certain files

Suggestion

Ability to specify which @types (or in general, which specified typeRoots) apply to which files in a project.

Use Cases

My project's src folder contains both .ts and .test.ts files, and I'd like an easy way for globals like describe to be defined only for the .test.ts files.

Examples

Not sure, but maybe some sort of new options in tsconfig.json for specifying which types items, lib items, or typeRoots items apply to which files.

This would be a beneficial feature because various editors (VS Code aside) look for a tsconfig.json file at the root of a project, and at the moment the only way to make types work in all files is to just provides all types for all files, which is undesirable because describe is not a function that is available in source files (as an example).

Example configuration: maybe instead of a typeRoots array, it can be an object like:

"typeRoots": {
  "./src/**/*.test.ts": ["../path/to/test/types"], // test-only types
  "./src/**/*.ts": ["../path/to/@types"] // other types other files
}

Or something. That's just an example to get the idea rolling.

Checklist

My suggestion meets these guidelines:

trusktr commented 4 years ago

The main concept is that we often need to install @types packages as devDependencies to use global types in things like tool configurations, scripts, tests, etc, and then those global types leak into source files (although they actually aren't available at runtime in those files).

If someone tries use a describe() test function (for example) in a source file instead of a test file, then at runtime this will be an error.

The idea here is that some option would allow us to specify which files certain types or typeRoots apply to, and our configuration would allow intellisense in editors like VS Code to show autocomplete options for APIs like describe() (for example) only in test files.

trusktr commented 4 years ago

The the new composite feature (described in the Project References doc) solve this request? Does it allow globals to be scoped per project?

The doc says

Previously, this structure (f.e. having both test/ and src/ folders that can import from each other) was rather awkward to work with if you used a single tsconfig file:

  • It was possible for the implementation files to import the test files

This seems to imply that isolation of globals may be possible (f.e. so that things like describe() and it() in test files don't leak into source files) .

EDIT: Yes, composite (Project References) does allow segregating global types per each reference. However, migrating from regular projects to project references can be cumbersome and lead to issues requiring refactoring, which is not ideal. Having an option that works with or without (tangential to, agnostic of) project references would be great.

trusktr commented 4 years ago

This issue keeps popping up the more TS projects I work on.

What we need is a more robust way of managing which global types are included or excluded. A couple cases are illustrated next.

Case 1: ergonomics:

Setting types to an empty array and manually listing all global types only to exclude, for example, a single global lib, is not ideal.

For this use case, we simply need a way to exclude certain global types.

Case 2: libraries with global builds and modules:

The three package on NPM can be imported as ES Modules, and the type defs work that way. However, that package also includes global types for people that may be using a <script> tag to load the lib as a global variable. When this package is added to a project, then even if we wish to use ES Modules for all types, the global types are still available (although the global variable itself is not), so this error can happen:

import {Camera} from 'three'

// all good, no type or runtime error
const camera = new Camera

// no type error, but runtime error "can't read property Scene of undefined"
const scene = new THREE.Scene 

For this use case, we simply need a way to exclude certain global types.

Case 3: multiple libs in a project depend on the same global namespace, but expect different types

Sometimes a project needs to use both React (for old parts of the code base) and some new library that also relies on JSX (for new parts of the code base). What ends up happening is that both React and the new lib (f.e. Solid.js, Preact, and a bunch of others) need to define JSX types. Libs that depend on JSX types don't all define the JSX types in a way that plays nicely with each other (React doesn't because it assumes to be the king, but Preact does) so what can happen is conflicts in JSX type declarations.

For this case, it would be great to be able to control which global types are available for which sets of files (I imagine it similar to a paths mapping, but mapping sets of files to type roots or similar). Then the project author can put React files in one place, and other-library files (f.e. Preact files or Solid.js files) in another place, and simply configure which files particular global types are available to.

An alternate existing solution for this case is to split the project into multiple TypeScript Project References, but this can be cumbersome and lead to other issues.

Having an option to control which globals are available to which files (which is a more fine-grained control than just "which globals to exclude" with the set of files to which we control global visibility being the whole project, as shown in the previous cases) would be useful here, without needing to restructure a project and introducing new issues. Furthermore, this option could still be helpful in Project References (it is tangential, and can be an option that works in any tsconfig.json, regardless if that tsconfig.json is for a project reference).

Case 4: source files vs test code

Projects often include both .ts and .test.ts files co-located with each other (f.e. src/foo.ts is next to src/foo.test.ts). We may want certain global types available only in the test code (f.e. describe, it, expect, etc) but not in the source code.

For this case the solution is the same one suggested for Case 3, for example exposing or excluding some globals to or from one set of files.


Here's a set of issues with people effectively asking for more control over global types:


@RyanCavanaugh Hoping this comment can satisfy the Awaiting More Feedback tag. (I'll move this comment to the OP)

aleclarson commented 4 years ago

Can we get an official response on this?

It's very annoying when I'm working on a browser-related project, and @types/node leaks in because a package in devDependencies depends on it.

I tried the "solution project" approach described in #37239 without success.

Related: #22331

aleclarson commented 4 years ago

This seems to be the current proposal: https://gist.github.com/RyanCavanaugh/702ebd1ca2fc060e58e634b4e30c1c1c

Each file's global scope is determined individually

Really not a fan of that idea. Excluding type packages with file globbing sounds much nicer.

Fallenstedt commented 4 years ago

I experienced a global clash issue today, and reproduced it here https://github.com/Fallenstedt/ts-globals-clash

Say you had a monorepo with two packages a and b.

The problem is b's test files are only reading jasmine types...when they should have jest types. Having the ability to exclude types for specific directories would be amazing. We are at the mercy of teams using the global feature of typescript well. Having a way to correct their mistakes would be great!

RyanCavanaugh commented 4 years ago

Excluding type packages with file globbing sounds much nicer.

This is just going to create compile errors in lib files that need them (and they surely must need them, otherwise they wouldn't reference them in the first place). The problem here isn't that random definition files end up in your program, the problem is that transitive dependencies end up in your program even though they're immaterial to your particular use case.

trusktr commented 3 years ago

Hello everyone and @RyanCavanaugh (and @aleclarson because you pointed out Ryan's proposal).

I've come up with a new compiler option idea that may solve some of the problems above (unless I missed that such a solution already exists):

https://github.com/microsoft/TypeScript/issues/42003

The hypothetical new declareOnImportOnly, when set to true, would prevent global and module augmentations from running unless you explicitly import the file that the augmentation code is inside of.

Rather than all declare global and declare module definitions eagerly making their way into your scope from any file in your project (including node_modules), they will all stay out, unless imported.

This seems like a very practical and standard role for import syntax to have (because you get what you import).

To make this work well, you'd want to import particular files into your project, rather than simply importing package index files that import everything you may not use (this good practice also helps reduce bundle size without any tooling).

trusktr commented 3 years ago

The reason for my previous comment is, apparently, global types are picked up even when tsconfig.json contains

    "types": [],
    "typeRoots": []

and you've imported anything from a package that has global types (even if you never import the files that declare global). It completely ignore types and typeRoots in this case.

The problem here isn't that random definition files end up in your program, the problem is that transitive dependencies end up in your program even though they're immaterial to your particular use case.

@RyanCavanaugh Another problem is that some parts of one project need certain globals, while other parts of a project need other globals.

For example, having two forms of JSX in a project where some files need to pick up React's global JSX, and in the other files we want some other JSX types without React's JSX types polluting global.

AlexGalays commented 3 years ago

Everyone has the same issues, which indicate something must be done.

A purely browser project shouldn't be forced with ambient node types which almost always end up being loaded (I don't think we can fix it at this level as it would require every single lib and their dependencies to be a good actor) and pollute the global namespace.

aleclarson commented 3 years ago

@RyanCavanaugh I have a suggestion for your proposal.

In addition to setting strictEnvironment: true, projects should have the option to include a specific @types package in any file matching a defined glob. My suggestion is to let the types option be a "glob map" like below:

// tsconfig.json
{
    "compilerOptions": {
        "strictEnvironment": true,
        "types": {
            "**/__tests__/**": ["jest"],
            "src/node/**": ["node"],
            "src/client/**": ["lib:dom"], // Maybe allow lib definitions too?
        }
    }
}

edit: On second thought, once strictEnvironment lands, this will be achievable by creating one tsconfig.json per glob.

// tsconfig.test.json
{
    "include": ["**/__tests__/**"],
    "compilerOptions": {
        "strictEnvironment": true,
        "types": ["jest"]
    }
}

// src/node/tsconfig.json
{
    "compilerOptions": {
        "strictEnvironment": true,
        "types": ["node"]
    }
}

// src/client/tsconfig.json
{
    "compilerOptions": {
        "strictEnvironment": true,
        "lib": ["dom"]
    }
}

But would I need two tsconfig.json for tests (one with [ jest, node ] and the other with [ jest, dom ]), or would tsconfig.test.json merge with the other two when they match the same file?

andreialecu commented 3 years ago

One other use case for this would be when a misbehaving package adds a global type that it shouldn't.

For example, @apollo/client adds the following:

declare global {
    interface Observable<T> {
        ['@@observable'](): Observable<T>;
    }
}

(https://cdn.jsdelivr.net/npm/@apollo/client@3.3.12/utilities/observables/Observable.d.ts)

This conflicts with rxjs's own Observable which creates editor issues in VSCode. A way to override this at the project level would be great.

See: https://github.com/apollographql/apollo-client/issues/7839

While this can be resolved by manually adding the import. It breaks editor auto import functionality as the editor thinks Observable is already a known type. This is a major pain as rxjs/Observable is used in nearly everywhere (angular app).

LaysDragon commented 3 years ago

yeah....I really need this. I try to develop a plugin under typescript for Blockbench, which is a combined environment under electron. .Its need lib.dom.d.ts type but its override some lib.dom.d.ts global type in plugin context. And its drive me crazy, since it just simple won't work caus of duplicated declared problem and seems not possible to override the global object under typescript d.ts which can be done under javascript. I guess I have to made a custom lib.dom.d.ts type as workaround :/

sabberworm commented 3 years ago

This is one of the few things about TypeScript that have irked me enough times over the years that I was going to open my own issue. But then I found this, which is sorta close enough, so I’m chiming in here:

IMO, global type declarations, despite the name, shouldn’t be made available globally at all. The “global” refers to them defining something in the global context. That doesn’t mean, however, that every file in a project should have access to it implicitly.

Especially in modules (files with imports or exports), where everything is side-effect-free and files are essentially meant to be isolated, having global types be implicitly available seems flat-out wrong.

Suggestion 1

I suggest adding another opt-in “strict” option, to remain backwards compatible. Having it enabled would mean global declarations would only be available in the file that imported it explicitly and importing a type declaration with global types wouldn’t make it implicitly available in all other project files as well. This is probably similar to https://gist.github.com/RyanCavanaugh/702ebd1ca2fc060e58e634b4e30c1c1c, but for types instead of libs.

This way, tests could be compiled along with source files in the same tsconfig without the source files suddenly having test, expect and whatnot defined. Each test file would instead have to explicitly use import 'test-framework' or /// <reference types="test-framework" />.

Suggestion 2

I think what lies at the core of this issue is that many module type declarations contain global declarations and exports, for legacy/UMD reasons. So one other partial solution would be ignore the global declarations when importing something from a module (as opposed to importing the module itself).

In this version, import 'jquery' and /// <reference types="jquery" /> in any file (or "types": ["jquery"] in tsconfig) would still make the $ global variable available in all files but import $ from 'jquery' would not.

This wouldn’t help much with the unit test example since many unit test frameworks’ typings only expose global declarations and have no exports declared, as the globals are injected by the test framework.

Open questions

FrederickEngelhardt commented 3 years ago

Any updates on this feature request? I've run into this limitation many times.

Adding a global way to control typeRoots would really make things much easier than having to manage configuration via inheritance and overrides.

So far I've been creating child tsconfig.json files that scope types better. Also been generating config dynamically for builds especially if you need type completion for testing purposes but are not shipping those types in your build.

samwightt commented 2 years ago

Any updates on this? Dealing with an issue where @types/express-jwt modifies some types in the express namespace (which is incredibly hard to track down because TypeScript never tells you that this is the case in its errors). I cannot remove the types because one of my packages depends on them. This means my only choice currently is to list out about 20-30 items in the "types" array, which just isn't ergonomic. And this is on a smaller project.

It's very annoying that something like this exists for 'inside the project' files but not for types. The current experience just isn't a good developer experience and is a massive annoyance on larger projects.

luxalpa commented 2 years ago

I don't have any suggestions, but another use case:

I want to use Storybook which internally adds @types/react with a vue.js project. Unfortunately the React types are overriding / merging the vue shims, therefore Vue's JSX syntax can not be used anymore :( Not sure what the solution to that problem is?

publicvirtualvoid commented 2 years ago

Monorepo using NX here - We build browser, NodeJS, and workers from a set of common libraries. The issues this aims to solve are easy to make and very time-consuming to resolve.

ajubin commented 2 years ago

I've encounter a similar issue, when developing a web application with nextjs.

I have a function called reportError which is also a function declared in the global namespace here but this doesn't always exists.

I forgot to import my function but I got no warning from the ts compiler.

I've fixed this by renaming my function, but I'd love to remove reportError from the globals to prevent this issue to happen again

zatloeri commented 2 years ago

I am running into this while developing a app with require js environment.

I am using jest for testing, but as soon as I include jest/utils node global variables get pulled in and override global requirejs variables like require and module thus making the code intended for requirejs env not compile correctly.

For anybody looking for workaround ideas: I have worked around this by splitting the project into two ts projects (app and jest tests) adding a import './globalTypeOverride' (where I have specified the global types I need for the code to compile) in my jest part of the project and using declare to redefine the global variables ad-hoc in jest (for example when I actually need to use Node require).

lgarron commented 2 years ago

The reason for my previous comment is, apparently, global types are picked up even when tsconfig.json contains

    "types": [],
    "typeRoots": []

and you've imported anything from a package that has global types (even if you never import the files that declare global). It completely ignore types and typeRoots in this case.

The problem here isn't that random definition files end up in your program, the problem is that transitive dependencies end up in your program even though they're immaterial to your particular use case.

@RyanCavanaugh Another problem is that some parts of one project need certain globals, while other parts of a project need other globals.

For example, having two forms of JSX in a project where some files need to pick up React's global JSX, and in the other files we want some other JSX types without React's JSX types polluting global.

Having spent several hours trying to get mocha and chai types to work "without magic", this rings very much true.

One particular place I've recently found this frustrating is module augmentation for testing frameworks. chai provides:

const { Assertion: untypedAssertion } = await import("@esm-bundle/chai"); // Partial ESM compat workaround
const Assertion: typeof import("chai").Assertion = untypedAssertion;

Assertion.addMethod("identicalAlg", function (expected: string): Promise<void> {
  // ...
});

The following doesn't work:

declare module "chai" {
    // .. only augment Chai.Assertion
}

So I have to do this:

const { Assertion: untypedAssertion } = await import("@esm-bundle/chai"); // Partial ESM compat workaround
const Assertion: typeof import("chai").Assertion = untypedAssertion;

Assertion.addMethod("identicalAlg", function (expected: string): Promise<void> {
  // ...
});

declare global {
  export namespace Chai {
    interface Assertion {
      identicalAlg(expected: string): Promise<void>;
    }
  }
}

Now these types are available from every file in the project. But:

1) I only want this to be available in test files, 2) only when the types are valid, which is only the case when Assertion.addMethod has happened in an import.

The current situation forces me to choose between several non-obvious behaviours that make it harder to make the project easy to maintain and contribute to.

(Of course, there are also issues like others have described in this thread, such as @types/node being available everywhere instead of only the files where node: modules are directly imported, or similarly for @types/web-bluetooth.)

kesupile commented 1 year ago

Any updates on this issue?

It's something that has come up quite a few times recently. Personally, I think @aleclarson's suggestion would be the most easy to understand while giving the most fine grained control.

RyanCavanaugh commented 1 year ago

The linked suggestion doesn't really solve anything IMO. If different parts of the program can "see" different global files, then they have different global scopes, and they have different interpretations of the same type.

Let's say you had something like this

// lib.object_rotation
interface Object {
  rotate(): void;
}

where file1 can "see" lib.object_rotation and file2 can't. What happens if you try to take an Object from file2 and give it to file1? Isn't it a type error, since it doesn't have a rotate method? What about assignability caching -- if file1 sees that Object is assignable to Rotatable, then file2 will have the same interpretation. So the entire assignability cache is now useless, or has an additional cache key of some kind.

These problems can be "solved" by the same mechanisms that we use for project references (IOW, separate compilations), but it's not clear what this extremely confusing per-file-global system would have over the existing solution of project references.

muuvmuuv commented 1 year ago

Eslit does a great job here with env and so on. Where you have a global and override scope. Whereas overrides can provide the same options as the toplevel scope but overrides object properties (or extends them) and arrays.

See https://github.com/angular-eslint/angular-eslint "Notes on eslint" for an example what I mean.

kesupile commented 1 year ago

The linked suggestion doesn't really solve anything IMO. If different parts of the program can "see" different global files, then they have different global scopes, and they have different interpretations of the same type.

Let's say you had something like this

// lib.object_rotation
interface Object {
  rotate(): void;
}

where file1 can "see" lib.object_rotation and file2 can't. What happens if you try to take an Object from file2 and give it to file1? Isn't it a type error, since it doesn't have a rotate method? What about assignability caching -- if file1 sees that Object is assignable to Rotatable, then file2 will have the same interpretation. So the entire assignability cache is now useless, or has an additional cache key of some kind.

These problems can be "solved" by the same mechanisms that we use for project references (IOW, separate compilations), but it's not clear what this extremely confusing per-file-global system would have over the existing solution of project references.

True, but I don't think anyone asking for these changes would actually use the system in that way. If file1 and file2 share data, then usually they're either in the same environment or the interface between them is explicitly defined. If the interface isn't explicitly defined, it wouldn't be surprising if the compiler raised an error.

clshortfuse commented 1 year ago

Reproducible:

npm install @types/mocha
npm install typescript

mkdir test
echo context > test/test.js
npx tsc test/test.js
# should pass ^^^

mkdir src
echo context > src/src.js
npx tsc src/src.js
# should fail ^^^

We want /src to not include @types/mocha but everything else instead. I haven't found a way to make this happen.

Perhaps compilerOptions.types can work with a glob/regex pattern:

/tsconfig.json:

{
  "compilerOptions": {
    "types": ["!(mocha)", "*"]
  }
}

/test/tsconfig.json:

{
  "compilerOptions": {
    "types": ["*"]
  }
}
MicahZoltu commented 1 year ago

Am I correct that if you want a web project that has tests which run in node (meaning @types/node present), the only way to have require not be accessible to your web project is to have two separate tsconfig.json files and have types: [...] in the web tsconfig include every @types/* except @types/node? There is no way currently to have a tsconfig that excludes a single @types folder?

jsejcksn commented 1 year ago

@MicahZoltu Assuming you want to include all other @types packages, yes — because it's an allow list.

From https://www.typescriptlang.org/tsconfig#types:

If types is specified, only packages listed will be included in the global scope.

As a workaround in the absence of negated pattern matching: you could write a simple script to generate a complete list by reading dependencies from package.json (or the dir structure from node_modules) — and remove the ones that you want to exclude. Perhaps such a script could be converted into a plugin for easier use.

wongchichong commented 1 year ago

for now, in install those types into another folder

pnpm add -D types_react@npm:@types/react

and refer it as needed in specific files

/// <reference types="types_react" />

gernotpokorny commented 1 year ago

Can somebody please summarize in short what the problem is in regards to why there's not much progress and this issue is still not solved after three years, given how serious it is? It's obvious that a glob pattern is needed (as OP already mentioned), because the problem is on a file basis.

jsejcksn commented 1 year ago

Can somebody please summarize in short what the problem is in regards to why there's not much progress and this issue is still not solved after three years, given how serious it is? It's obvious that a glob pattern is needed (as OP already mentioned), because the problem is on a file basis.

@gernotpokorny Other tasks have been identified to be more important.

TypeScript is Free Open Source Software, so anyone (including you and me) can fix the issues — and even contribute our fixes upstream (using the Pull Request feature), so that everyone can benefit.

gernotpokorny commented 1 year ago

@jsejcksn What is more important then having clean types in TypeScript? ^^

Shouldn't this be at the top of the list after critical bugs?

ajubin commented 1 year ago

If you use eslint, you can use this rule https://eslint.org/docs/latest/rules/no-restricted-globals to disable some globals

If you want to enable/disable on certain file, you can use either /* eslint-disable no-restricted-globals */ at the beggining of the file or use overrides proprerty in eslintignore to match a given list of pattern

clshortfuse commented 1 year ago

This may now be solved with extends in v5.0.0 though I haven't confirmed this personally yet:

https://devblogs.microsoft.com/typescript/announcing-typescript-5-0-rc/#supporting-multiple-configuration-files-in-extends

egasimus commented 1 year ago

@jsejcksn

TypeScript is Free Open Source Software, so anyone (including you and me) can fix the issues — and even contribute our fixes upstream (using the Pull Request feature), so that everyone can benefit.

Nice copout. What do you gain from it?

Consider the differential between the complexity of the TypeScript codebase vs. the resources of any individual developer blocked by any particular issue.

Consider that there is no guarantee that the proposals of contributors outside Microsoft will be included: in my experience, maintainers react with "idk what this guy is talking about" to issues that do not further the product agenda - regardless of years upon years of valid arguments provided by the community.

Do you honestly expect all PRs to be reviewed in good faith?

Of course, TS devs are a competent bunch, so they excel at making excuses for TS being ridden with antifeatures that disallow people from just solving their own problem. See https://github.com/microsoft/TypeScript/issues/18588 - or this absolutely brilliant and totally fair rationale for closing https://github.com/Microsoft/TypeScript/issues/14979:

typeRoots is rather an advanced feature. that and the limited number of these folder, suggest that this can be left to humans to figure out, and does not need to be done by the compiler.

I'd rather humans be left to figure out the correctness of types for themselves, too: by writing tests to verify their code actually works - rather than delegating about half of that task to a static analysis tool that also, by necessity, is a compiler, then disregarding the other half. However, my coworkers only know TypeScript. They do not seem to know JavaScript. Or testing. I like them, and I like where I work. So I'm stuck wrangling TS, too - for mine and their sake.

For example, let's say I want to include . and exclude ./dist in typeRoots. Am I dumb or is there no way to specify that? Glob is not supported, exclude is not supported... what then? Move all my code under ./src? Sure, "everybody does that", and I could totally do that, too - except what if I have a valid reason not to? If TS was committed to doing things compatibly and correctly, one wouldn't need to adapt their workflow to accommodate the blind spots of a bunch of Microsoft people that act deaf to the frustration of their fellow programmers.

The FOSS solution would be to fork TypeScript - and then the community would be forced to forever maintain an "alternative TS", that would slowly grow incompatible with upstream (just like TypeScript is now practically incompatible with JavaScript, talk of "supersets" notwithstanding). Good luck getting people to know about it and switch over!

Boo, Microsofties. Shame on you. Instead of accommodating developers' needs, you keep gaslighting them into exhausted compliance. I hope one day every TS dev and TS apologist learns the error of their ways. Right now, I'd say TS5 sucks about 10% less than TS4, so maybe in TS6 another batch of papercuts that wasted innumerable dev hours will be resolved :crossed_fingers:

Sorry, your comment struck a nerve there. I stand by my words, though: TypeScript is fake "Open Source Software".

P.S. I've got half a mind to email the downvoters and ask them when was the last time they got a PR merged into TypeScript. But instead, I'll just leave this here https://graydon2.dreamwidth.org/306832.html and leave you to experience your knee-jerk reactions - I've got some AST rewriters to implement...

acusti commented 9 months ago

Can somebody please summarize in short what the problem is in regards to why there's not much progress and this issue is still not solved after three years, given how serious it is? It's obvious that a glob pattern is needed (as OP already mentioned), because the problem is on a file basis.

@gernotpokorny Other tasks have been identified to be more important.

TypeScript is Free Open Source Software, so anyone (including you and me) can fix the issues — and even contribute our fixes upstream (using the Pull Request feature), so that everyone can benefit.

@jsejcksn is there a solution to this problem that core typescript maintainers have indicated support for that someone from outside that core group could implement and PR upstream? the only comments in this thread from a core member are https://github.com/microsoft/TypeScript/issues/37053#issuecomment-730725238 (in which @RyanCavanaugh indicates that a file glob pattern solution is inviable) and https://github.com/microsoft/TypeScript/issues/37053#issuecomment-1355747946 (in which @RyanCavanaugh reiterates that the “linked suggestion doesn't really solve anything IMO”).

over in #52433, which requests support for treating libs as a union (not an intersection, as they are right now) in order to support type checking code that runs in multiple environments, there’s https://github.com/microsoft/TypeScript/issues/52433#issuecomment-1405570230, which again indicates a posture that is generally closed towards any of the proposed solutions (“The problem we've had here in the past is that a huge amount of code can be seen to be running in one environment by inspection, but not in a way that's susceptible to static analysis.”), but it does end with this:

There's nothing today stopping someone from auto-genning variant DOM/webworker/etc libraries with every top-level declaration marked possibly-undefined. Real-world usage of such output would be good evidence that this kind of feature would be usable in practice.

that’s a suggestion for a user-land-based solution that would help with the isomorphic problem-space described in #52433, but would do nothing to help with the issue here as described by Ryan Cavanaugh as:

transitive dependencies end up in your program even though they're immaterial to your particular use case

all of this leaves me with two questions:

  1. what is the solution for this problem that someone could work on that has any indication of support from any core typescript team members?
  2. has anyone tried to auto-gen variant DOM/webworker/etc libraries with every top-level declaration marked possibly-undefined? if no one has, i’d be happy to give it a go and try to see if it solves any problems, though as a library consumer, not a library author, i’m not sure if there’s an existing auto-gen tool i can use to do that or if the assumption is that whoever creates those libs would need to also create the tools needed to parse and transform the existing libs in the way suggested. what i really want is to do something like import type * as DOMTypes from 'dom'; export type OptionalDOM = Partial<DOMTypes>;, so that i could continue to rely on the libs and not have to maintain a fork, but 🤷.
dominictobias commented 8 months ago

Driving me nuts. JSX is becoming a more and more popular choice on the server, but as soon as a React project imports anything from the server, like a strong typing contract for the API, all types in React break because the non-React JSX redefines JSX globally

HitkoDev commented 8 months ago

@acusti What I got from this whole debate is that the right way of dealing with this issue would be for libraries to provide a separate globals.d.ts file, and for the developers to include it as { "types": ["library/globals"] } only where they actually rely on particular global values.

However, doing it that way is discouraged by the fact that most users would view it as an inconvenience rather than a feature, as well as being a major change for existing libraries, many of which are no longer actively maintained in the first place. So other than doing it the wrong way, I don't think there's another practical solution to this issue.