Typedoc confuses any error types.

[dragomirtitian]

Search terms

type alias, rename

Expected Behavior

Typedoc should not confuse any error types.

Actual Behavior

Tyedoc ends up reporting types as being a rename of another type even though the types are wholly unrelated

Steps to reproduce the bug

//@ts-ignore
export type { T1 } from 'bad-module' // no types, no definitions

//@ts-ignore
export type { T2 } from 'bad-module-2' // no types, no definitions

//@ts-ignore
import { T3 } from 'bad-module-3' // no types, no definitions

export default T3;

This produces:

T2

Renames and re-exports T1

default

Renames and re-exports T1

Even though T2 and default are not in any way related to T1. For default it's even more confusing as T3 is not even mentioned, it gets resolved to T1

Environment

• Typedoc version: 0.22.10
• TypeScript version: 4.5
• Node.js version: 15
• OS: Windows 10

[Gerrit0]

I'm really tempted to say that there's nothing to say that those aren't the same symbol. It wouldn't be difficult to always treat the error symbol as a new symbol, but then we might run into the opposite problem if anything references it (typeof T1 resulting in a "not exported" warning).

[Gerrit0]

Having thought about this some more, we might be able to fix this by changing how symbol references are stored. Right now, TypeDoc relies on reference equality to determine if symbols are the same. If we change that to be based on paths + names (paths + names + line number to handle declaration merging?)...

This is something I've wanted to try for a while, since doing it should make it easier to fix #1606.

To do this, we'd need an update to the validateExports function, and it would probably also end up breaking typedoc-plugin-missing-exports... which is annoyingly popular.

[Gerrit0]

I did some more investigation on this, and... I don't think this is something I want to "fix". It is technically possible, but doing this means that a core assumption about how aliased symbols work would have to be changed, and would require a lot of rework with consequences I'm fairly certain would be detrimental in other cases (could probably hack the change in with 50loc, but I think it would break several internal plugins in very easy to miss ways).

An any is an any is an any...