fix(1685): refine export assignment diagnostics to match strada

[a-tarasyuk]

Fixes #1685

---

This patch addresses the issue of incomplete export assignment validation by aligning the check with Strada logic.

const declaration = getDeclarationOfAliasSymbol(exportEqualsSymbol) || exportEqualsSymbol.valueDeclaration;
if (declaration && !isTopLevelInExternalModuleAugmentation(declaration) && !isInJSFile(declaration)) {
    error(declaration, Diagnostics.An_export_assignment_cannot_be_used_in_a_module_with_other_exported_elements);
}

[jakebailey]

This one's a regression.

[a-tarasyuk]

@jakebailey This may not be directly related to the change; it seems the issue is that the jsdoc callback is emitted as exported in the .d.ts files.

typescript-go/internal/parser/reparser.go

Line 73 in d2c442d

// !!! Don't mark typedefs as exported if they are not in a module

export = MyClass;
....
export type DoneCB = (failures: number) ;

while strada handles this case as

export = MyClass;
....
type DoneCB = (failures: number) => any;

I think the way to address this regression is to elide the export modifier. WDYT?

[jakebailey]

That sounds plausible. Thankfully I believe ExternalModuleIndicator is set before reparsing.

[a-tarasyuk]

It looks like ExternalModuleIndicator is set only after parsing top-level statements

typescript-go/internal/parser/parser.go

Line 350 in c0aa663

p.finishSourceFile(result, isDeclarationFile)

typescript-go/internal/parser/parser.go

Line 379 in c0aa663

ast.SetExternalModuleIndicator(result, p.opts.ExternalModuleIndicatorOptions)

[weswigham]

Yeah, the declaration emitter or reparser definitely needs to be adjusted to handle this. Probably the reparser, in this case - it probably shouldn't add an export modifier if there's an export assignment in the file. Truth be told, though, this is one of those cases where we can't accurately represent the input as a declaration file, and no matter what we do we're probably going to change behavior from strada.

[jakebailey]

@sandersn @weswigham Is this the thing we were discussing about relaxing this for even TS code and not erroring at all? I can't remember where we left that.

[weswigham]

We're at the "just do it like strada" stage because trying to improve it just moved where the jank in our code is - better the jank we know.

[weswigham]

Definitely an improvement. I'm sure we'll have to keep adjusting the JS declaration emit for export assignments, though, since they're still only partially supported.

[weswigham]

Hm. This is because we don't actually support constructor functions yet, right @sandersn ?

[jakebailey]

This one is regressing; maybe a mad interplay with the new code?

[jakebailey]

This one too? (I hate JSDoc typedef rules)

[jakebailey]

This file's also regressing in a similar way, it seems...