[Bug]: Array.sort in hoist-jest might reversing all statements

[hanjunspirit]

Version

27.1.5

Steps to reproduce

for the code:

class Base {
  sum(a: number, b: number) {
    return a + b;
  }
}
class MMath extends Base {}
export { MMath };

Expected behavior

Actual behavior

It will be compiled into:

export { MMath };
class MMath extends Base {
}
class Base {
    sum(a, b) {
        return a + b;
    }
}

This leads to Base class not being found, causing an error.

Debug log

Additional context

This piece of code has different results in different versions of Node.js:
[1,2,3].sort(() => 1)

For example,
in Node.js 10.24.0, it returns [3, 2, 1],
But in Node.js 12.18.3, it returns [1, 2, 3]

This depends on the implementation of the sort algorithm, and the result is unstable.

So for the sort in hoist-jest(#3498), if all isJestGlobalImport and isHoistableStatement return false, it might reversing all statements.

Environment

Node: 10.24.0
Jest: 27.1.5

[tschoartschi]

@hanjunspirit I think I have a similar problem. How did you debug it? How did you get the compiled code? I think in my case also the order of the statements is somehow mixed up. Although I import the class definition on top of the file I get the following error:

ReferenceError: Cannot access 'Base' before initialization

I'm a little bit lost because I don't know how to debug and therefore I can not create a repo with reproduction and without a repo I can not create a bug report.

[tschoartschi]

If someone else has the same problem, we found a workaround that works for us (probably that's not possible in every project).

We were able to reference the problematic file in setupFiles (https://jestjs.io/docs/configuration#setupfiles-array) and now everything works as expected. For us, this is a good enough workaround because it unblocks us 🎉 and we can continue to work.

Nevertheless, a long-term solution would be better than a workaround. We also expect that the hoisting "feature" of jest breaks our code. It's super hard to create a repo that reproduces the problem. We think that our problem is very similar to what @hanjunspirit encountered.