[BUG-28] Effects queued during effect execution are silently dropped

[shroy]

Severity: P2 · Batch: B3 (effect system)

Location

src/core/hookContext.js:147-154

Root cause

forEach doesn't visit items appended mid-iteration and effectQueue.length = 0 then wipes them — a useEffect registered from inside another effect never runs.

Proposed fix

Drain with while (effectQueue.length).

Verification recipe

Spec: hook whose useEffect body registers another useEffect(spy); run in context; assert spy ran. Expected on unmodified code: never runs → confirmed.

[shroy]

Phase 1 verification: CONFIRMED

Method: runtime repro + static analysis

Root cause (confirmed)

src/core/hookContext.js:147-154 — executeEffectQueue drains the queue with effectQueue.forEach(...). Array.prototype.forEach snapshots length at the start of iteration, so any effect appended mid-iteration (i.e. a useEffect registered from inside another effect's body, which pushes to context.effectQueue at line 310) is never visited. Line 154's effectQueue.length = 0 then wipes the un-visited entry. Net effect: a useEffect registered from inside another effect never runs.

The forEach/append semantics are not implementation-defined; verified directly:

$ node -e "const a=[1],s=[];a.forEach((x,i)=>{s.push(x);if(i===0)a.push(2)});console.log('seen',JSON.stringify(s),'final',JSON.stringify(a))"
seen [1] final [1,2]

2 was appended during iteration and never visited — exactly the mechanism the finding describes.

Repro spec

Placed at src/tests/__verify__/issue-39.spec.js, asserts the correct/documented behavior (nested effect must run):

import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest'
import { useEffect, withHookContext, runCleanupFunctions } from '../../core/hookContext.js'
import { initConfig } from '../../core/config.js'

describe('issue #39 — effects queued during effect execution', () => {
  let container

  beforeEach(() => {
    initConfig()
    container = document.createElement('div')
    document.body.appendChild(container)
  })

  afterEach(() => {
    runCleanupFunctions(container)
    document.body.removeChild(container)
    vi.restoreAllMocks()
  })

  it('runs a useEffect that is registered from inside another effect body', () => {
    const innerSpy = vi.fn()

    // expect() must not live inside a framework-invoked effect body (swallowed);
    // capture into an outer spy and assert outside the framework-managed call.
    withHookContext(container, () => {
      useEffect(() => {
        // Registering another effect from within an effect's setup body.
        useEffect(innerSpy, [])
      }, [])
    })

    // The nested effect's body should have executed exactly once.
    expect(innerSpy).toHaveBeenCalledTimes(1)
  })
})

Test output (unmodified code)

 ❯ src/tests/__verify__/issue-39.spec.js > issue #39 … > runs a useEffect that is registered from inside another effect body
     → expected "spy" to be called 1 times, but got 0 times

 AssertionError: expected "spy" to be called 1 times, but got 0 times
 ❯ src/tests/__verify__/issue-39.spec.js:45:22
      44|     // The nested effect's body should have executed exactly once.
      45|     expect(innerSpy).toHaveBeenCalledTimes(1)
        |                      ^

 Test Files  1 failed (1)
      Tests  1 failed (1)

Interpretation

The nested useEffect's body ran 0 times instead of the expected 1 — the effect queued during effect execution is silently dropped, exactly as reported. The finding is CONFIRMED. The proposed fix (drain with while (effectQueue.length) and shift/index forward rather than forEach + length = 0) would resolve it; care is needed so effectOrder / initializedEffects bookkeeping stays consistent for the newly-drained effects.