fix(effects): make within/tryTo/hopeThat/retryTo work under tsx/cjs (#5632)

[mirao]

Fixes #5632

Problem

within() and tryTo() (and hopeThat() / retryTo()) silently do nothing in 4.x when a TypeScript test imports effects through a CommonJS loader, e.g.:

import { within } from "codeceptjs";
import { tryTo } from "codeceptjs/effects";

run under require: ['tsx/cjs'] (the setup the official Quickstart generates).

Root cause

CodeceptJS 4.x is native ESM, and it loads test files through Mocha's synchronous require() (lib/mocha/factory.js), i.e. the CommonJS realm. So when a test imports codeceptjs/effects, tsx/cjs turns it into a require() and Node materializes a second, disconnected CJS copy of effects.js together with its own copies of the recorder and container singletons.

That recorder is never started (running = false), so every recorder.add() short-circuits to Promise.resolve():

• tryTo() / hopeThat() return undefined without ever running their callback (no error is reported even if the inner steps would fail),
• within() skips all of its inner steps, and its empty CJS container never calls _withinBegin, so the context is never applied,
• retryTo() never resolves and hangs.

Because test files are always loaded via require(), this is the default for essentially every TypeScript project — including the bare Quickstart.

Fix

Make the singletons realm-agnostic using the same globalThis bridge the framework already uses for global.codeceptjs (lib/globals.js):

• lib/recorder.js — start() registers the running instance on globalThis.__codeceptjs_recorder (only the instance the runner actually starts registers itself).
• lib/container.js — create() registers the live container on globalThis.__codeceptjs_container.
• lib/effects.js — resolves the recorder/container through _getRecorder() / _getContainer(), which prefer the globalThis instances and fall back to the local module imports. Under pure ESM both point at the same object, so there is no behavior change on the ESM path.

Tests

Adds a runner regression test that drives a real tsx/cjs project and asserts all four effects execute:

• test/data/effects-tsx-cjs/ — fixture (config with require: ["tsx/cjs"], a helper, and a test that imports effects through the CJS loader).
• test/runner/effects_tsx_test.js — asserts within, tryTo, hopeThat and retryTo all run.

It fails without the fix and passes with it. The exec call has a timeout guard so that a future regression which re-hangs retryTo() fails cleanly instead of hanging the suite.

The existing pure-ESM coverage (test/unit/effects_test.js, test/runner/within_test.js) continues to pass — that path previously had no coverage for the CJS-loaded case, which is why the regression shipped.

🤖 Generated with Claude Code

[mirao]

Heads-up: #5636 (fix for #5635) addresses the same dual-realm split as this PR, but at the source rather than in effects.js.

#5636 makes the recorder and container singletons themselves realm-agnostic (via a shared globalThis instance in the new lib/realm.js), so the CJS-loaded copy of effects.js resolves the live recorder/container directly through its normal import — no _getRecorder()/_getContainer() indirection needed. The same source-level fix also covers config, event, store, output, and transitively within/session/pause/inject/etc.

I verified empirically on the #5636 branch (which does not include this PR — effects.js is unpatched there) that all four effects run from a CJS-loaded test:

EFFECTS_TRYTO result=false
EFFECTS_CLICK withinActive=true
EFFECTS_PASS ran
EFFECTS_HOPETHAT result=true
EFFECTS_FLAKY try=2
EFFECTS_RETRY done   ← retryTo resolves, does not hang

So this PR's _getRecorder()/_getContainer() changes become redundant once #5636 lands (both sides of globalThis.__codeceptjs_recorder || recorder resolve to the same shared object).

To avoid losing the valuable regression coverage from this PR, I've ported its effects test into #5636 (test/data/internal-api-tsx-cjs/effects_test.ts + the corresponding runner assertions), so within/tryTo/hopeThat/retryTo under tsx/cjs stay guarded.

Suggest we close this PR in favor of #5636 — but happy to keep it if you'd prefer the narrower change first. Either way the effects coverage is preserved.

[mirao]

Suggest we close this PR in favor of #5636 — but happy to keep it if you'd prefer the narrower change first. Either way the effects coverage is preserved.

Closing as it was fixed in #5636 by more general way.