[Andrew Branch (andrewbranch)]

Fixes #64053

This makes TypeScript content mapping a better citizen in a system where another tool is going to emit content mapped files to disk. The current system basically assumes a bundler or running the source in place.

A content mapper's package.json can now specify typescript.contentMapper.outputExtensions, a mapping from a content mapped file's original extension to the output extension that another tool will emit:

{
  "name": "ember-content-mapper",
  "typescript": {
    "contentMapper": {
      "exec": ["node", "server.js"],
      "outputExtensions": {
        ".gts": ".js"
      }
    }
  }
}

The same configuration is supported in tsconfig.json as an override (the entire mapping is overridden; keys are not merged):

{
  "contentMappers": [
    {
      "package": "ember-content-mapper",
      "extensions": [".gts"],
      "outputExtensions": {
        ".gts": ".js"
      }
    }
  ]
}

This causes:

• The declaration file output for Component.gts to be Component.d.ts instead of Component.d.gts.ts, because there is expected to be a Component.js (that someone else will create).

• ⚠️ The program that uses this content mapper must set rewriteRelativeImportExtensions: true. Why? If configuration indicates that Page.gts is going to be emitted as Page.js, then a plain TypeScript file in the same program that imports it:

  // routes.ts
  import Page from "./Page.gts";

  must be emitted as

  // routes.js
  import Page from "./Page.js";

  which is exactly what rewriteRelativeImportExtensions does.

  The alternative would be to import from "./Page.js" or even "./Page" depending on module resolution settings, i.e., an output-compatible path. That request is Content mappers: let registered extensions take part in extensionless module lookup #64549, and while this PR allows us space to consider that in the future, I strongly prefer imports that will trigger a content mapper be immediately recognizable by file extension.

[Jake Bailey (jakebailey)]

I think the casing stuff might be out of date after the typed paths PR?

[Leonidas (leonidaz)]

Andrew Branch (@andrewbranch) Thanks for working on this. The output-extension support is a useful step forward.

I strongly prefer imports that will trigger a content mapper be immediately recognizable by file extension.

I understand the preference for making the source format visible, but I don’t think that should rule out extensionless resolution when a project explicitly configures it and its build tool supports it.

TypeScript already allows import Card from "./Card" to resolve to Card.tsx in the appropriate resolution modes. That import does not reveal that the source contains JSX or requires JSX processing. Why should components authored in a registered, mapper-supported format have to expose that implementation detail in every import? The mechanisms differ, but the developer-facing expectation seems equivalent. [TypeScript’s extension-substitution rules](https://www.typescriptlang.org/docs/handbook/modules/reference.html#file-extension-substitution)

More fundamentally, moduleResolution: "bundler" is intended to model the bundler’s resolution behavior. When the bundler is configured to resolve ./Card to Card.tsrx, rejecting that import in TypeScript creates a mismatch between the type checker and a working build. This is particularly difficult to justify for noEmit projects, where TypeScript is not responsible for producing runtime imports. [Extensionless-path documentation](https://www.typescriptlang.org/docs/handbook/modules/reference.html#extensionless-relative-paths)

The proposal in #64549 is deliberately limited to registered extensions and contexts that already support extensionless lookup. The resolver would use the registered extension list; the mapper would still process the file after resolution selects it. Node ESM’s explicit-extension requirements would remain intact.

There are practical benefits beyond shorter imports. Extensionless paths let a component change source format without requiring changes throughout its import graph. They also preserve existing conventions in projects whose language tools already support this through Volar, avoiding import rewrites solely to migrate to TypeScript 7.

The precedence and declaration-emission questions deserve explicit answers. However, they seem like reasons to define the feature’s constraints rather than require source extensions universally, especially when declaration emit is irrelevant to a noEmit application.

I really hope you would consider an opt-in setting on a contentMappers entry. That would let projects request behavior consistent with their build tooling while keeping explicit extensions available as a project convention.

[Jake Bailey (jakebailey)]

When would we ever want that flag off if it's config driven anyway?

[Andrew Branch (andrewbranch)]

We don't want the flag enabled in the type printback node builders for diagnostics / LSP (types should display as import("./Card.astro"), not import("./Card.js"))

[Jake Bailey (jakebailey)]

Curious if this needs to not be internal for those printing types back for other reasons, but maybe not

[Jake Bailey (jakebailey)]

Code LGTM if the design is what we all decided on.

[Daniel Rosenwasser (DanielRosenwasser)]

We actually discussed and said that null was an acceptable output value.

[Andrew Branch (andrewbranch)]

This is in the package.json, not the tsconfig. f78aff7 adds null handling for tsconfig.