fix nested acts

[snowystinger]

What:

Closes #1231
And removes the need for testing-library/user-event#1283

Why:

testing-library/user-event#1283

In React, nested synchronous act calls can prematurely teardown the act queue creating console warnings.

How:

By tracking if we're already inside an act, we can avoid wrapping it in a new one. At the same time, we can keep the wrapper around all events, which is an expectation using testing library's configure

Checklist:

• Documentation added to the N/A
  docs site
• Tests
• TypeScript definitions updated N/A
• Ready to be merged
  

Originally I was going to make this change over in user event, we have two PRs to do it, however, it's really a React and Preact concern. Changing it in User Event would affect everyone. Angular testing library already wraps all events for a different reason and so removing an event from that callback wrapper could be unexpected or breaking. I would guess this change is also needed in the Preact testing library repo as well.

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit b115de7:

Sandbox	Source
react-testing-library-examples	Configuration

[eps1lon]

Can you write a test showing the underlying issue instead of testing implementation details (nesting depth)?

[snowystinger]

sorry about that, I was stumped on it for a little bit
I got there today though, just needed to sleep on it. done.

[filipems]

There is now a React-only public-API reproduction of the underlying failure this PR is working around: react/react#37374

The minimized case removes Testing Library, user-event, Radix, and focus/blur behavior entirely. A re-entrant act() entered while an outer flushActQueue is running sees actScopeDepth === 0, takes the outermost-exit path, cannot flush because isFlushing is already true, and nulls actQueue mid-flush. The outer act() then returns with React updates still pending on the real scheduler.

That issue includes a standalone node repro.mjs, expected vs actual behavior, causal controls, and the queue trace. It may also provide the behavioral test case requested in this PR without relying on act-depth instrumentation.