Typescript 4.5 Beta infinite loop leading to JS heap out of memory

[mmvsk]

An issue appeared with TypeScript 4.5.0-dev.20211001 that made tsc go into an infinite loop leading to an out-of-memory error.

tsconfig.json:

{
  "compilerOptions": {
    "target": "es2020",
    "module": "commonjs",
    "moduleResolution": "node",
    "lib": ["es2020"]
  },
  "exclude": ["node_modules"],
  "include": ["run.ts"]
}

package.json:

{
  "type": "commonjs",
  "scripts": {
    "start": "tsc -p ."
  },
  "dependencies": {
    "zod": "^3.9.8"
  },
  "devDependencies": {
    "typescript": "^4.5.0-beta"
  }

when using tsc directly, the following import cause the issue:

import { z } from "zod";

when using ts-node instead of tsc, this is the minimum necessary code to cause the issue:

import { z } from "zod";

const schema: z.ZodSchema<string> = z.string();

Demo code in: https://github.com/mmvsk/bug-zod-typescript-4.5.0-oct-01

The same code works until TypeScript 4.5.0-dev.20210930, and the issue was probably introduced with the following TypeScript PR: microsoft/TypeScript#41821

[MikeRippon]

For reference, there's another issue here, which I think is probably tracking the same issue (?)

[mmvsk]

It's not the same, as Zod was working with TS 4.3, 4.4, and 4.5 until 4.5.0-dev.20210930 (included).

There was a pretty large refactoring in TypeScript PR microsoft/TypeScript#41821, and by quickly looking at the diff and the scope of that PR (refactoring recursion-related code) it may very well have introduced this kind of regression.

However, I don't know if Zod is using a faulty behavior of TS that was fixed in 4.5.0-dev.20211001, or if it's a real new TS issue.

[mmvsk]

I just tried on a blank setup (Mac M1, without node previously installed), same issue.

Better to figure this out before TS 4.5 is released.

[mmvsk]

Well, there is progress.

TypeScript was fixed (in microsoft/TypeScript#46326), so now we have a clear error message instead of a memory error.

I created a new issue #715.

[colinhacks]

Happy to see that fix land. I struggled to solve that issue for most of Saturday and made minimal progress. The DeepPartial thing is a much easier fix, I'll look into it when I have a chance.