Closure-compiler: Extending computed class name identifiers throws:

[mohsen1]

Input:

var a = {
  foo: class {}
}

class bar extends a['foo']{

}

CC throws:

JSC_DYNAMIC_EXTENDS_TYPE: The class in an extends clause must be a qualified name. at line 6 character 18
class bar extends a['foo']{
                  ^

Is this by design? The JS code is valid.

[mohsen1]

I think CC should not treat bracket property accessor with string value as "dynamic". In my example class bar extends a.foo works but class bar extends a['foo'] doesn't while both are the same.

TypeScript compiler can follow the code:

[MatrixFrog]

Can you elaborate a bit on what your use case is? i.e. why you don't want
to write 'a.foo' instead?

[mohsen1]

Webpack outputs code like this:

class APIClientError extends __WEBPACK_IMPORTED_MODULE_3_helpers__["e" /* CustomError */] {

And I am using CC to minify Webpack output.

[alexeagle]

Duplicate of https://github.com/google/closure-compiler/issues/2106 ?

[ChadKillingsworth]

Not a duplicate - it's different. Fixing this would be a pretty simple change to the transpilation. Should be non-controversial as well.

[tomccabe]

Just ran into this as well trying to replace the ever-bloating Babel with CC. Sadly I'm getting otherwise great results, but this error pretty much prevents us from adopting CC.

[TheLarkInn]

Webpack outputs code like this:

class APIClientError extends __WEBPACK_IMPORTED_MODULE_3_helpers__["e" /* CustomError */] {

@mohsen1 webpack would only replace that if you are extending a namespace / import / etc. that you have imported. Can you please share the original source?

[mohsen1]

I don't exactly remember the original code anymore but this is close enough:

// Errors.ts
export class CustomError {
// ....
}

// SomeFile.ts
import { CustomError } from './Errors.ts';

class APIClientError extends CustomError {
// ...
}

[TheLarkInn]

Thanks, @mohsen1 https://github.com/webpack/webpack/blob/master/lib/dependencies/HarmonyImportSpecifierDependency.js#L87-L99 if you wanted to hack.

[sokra]

As workaround:

// SomeFile.ts
import { CustomError } from './Errors.ts';

const CustomErrorTemp = CustomError;

class APIClientError extends CustomErrorTemp {
// ...
}

[vanhoavn]

The problem is from cases like this from typescript in ES6 mode:

class APIClientError extends CustomError<API> {
// ...
}

we cant do:

const CustomErrorTemp = CustomError<API>;

[mohsen1]

You should be able to do that with type:

type CustomErrorTemp = CustomError<API>;

class APIClientError extends CustomErrorTemp {
// ...
}

[vanhoavn]

Hmm error:

error TS2693: 'CustomErrorTemp' only refers to a type, but is being used as a value here.

[mohsen1]

oh right... sorry for misleading comment

[Treeki]

I ran into this issue and decided to take a stab at fixing it (despite never having touched the Closure Compiler codebase before...). As far as I can tell, the most elegant solution is to allow CC to consider terms like a.b['c'] as a qualified name, similar (but not equivalent!) to a.b.c.

I don't want to submit a PR for this as my fix isn't perfect -- it fails some tests -- but perhaps somebody more familiar with the codebase can build upon it and complete it: https://github.com/Treeki/closure-compiler/tree/allow-constant-getelem-in-qualified-names

[nichtverstehen]

I deal with this limitation by manually extracting the base into a variable. I wonder if applying such a transform mechanically could be a correct resolution for this issue.

I.e. for the code in the first comment, the rewriting could be:

var a = {
  foo: class {}
}

var barBase0 = a['foo'];
class bar extends barBase0 {
  // ...
}

Will this be typed as expected?

[nichtverstehen]

Hmm, I see I unassigned MatrixFrog here, but I didn't intend to; I was just posting a comment. I don't think I even have a permission to assign or unassign anyone here. Please disregard and re-assign the issue back if you are an owner.

[paulfisher53]

I'm also experiencing this issue with webpack + closure compiler. Is there any ETA on a fix?

[ChadKillingsworth]

See #2997

[ChadKillingsworth]

Closed by #2997