PR #84116: fix(cli): expose banner emit-once reset for tests (#83903)

• Repository: openclaw/openclaw
• Author: hclsys
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/84116

Description (problem / solution / changelog)

Fixes #83903.

src/cli/banner.ts exports hasEmittedCliBanner() (read-only) but no reset path for the module-level bannerEmitted latch. The latch is correct for production (emit the banner exactly once per process) but blocks multi-scenario vitest specs in the same worker: the second emitCliBanner call silently skips because bannerEmitted is still true from the first. The sibling module src/cli/channel-options.ts already solves this same module-cached-state problem with a __testing.resetPrecomputedChannelOptionsForTests() export — this PR mirrors that pattern.

Changes

• src/cli/banner.ts: add export const __testing = { resetBannerEmittedForTests(): void { bannerEmitted = false; } } mirroring the channel-options.ts pattern. Production behavior (the emit-once latch) is unchanged.
• src/cli/banner.test.ts: add two regression cases that drive the latch end-to-end. The first asserts the latch suppresses a second emitCliBanner call in the same scenario (production semantics preserved). The second resets via __testing.resetBannerEmittedForTests() between calls and asserts both emits succeed — the use case the helper enables.

Diff stat: 2 files, +76 / -0.

Real behavior proof

• Behavior or issue addressed: Sanitized issue evidence — bannerEmitted is declared at module scope, hasEmittedCliBanner() exposes it read-only, but there is no symmetric reset like channel-options.ts:43-47. Any test that calls emitCliBanner twice (or in two scenarios sharing the module cache) has the second call silently skipped.

• Real environment tested: Local Node 22.x. Probe at /tmp/probe_83903.mjs (a) parses the patched banner.ts and verifies the __testing.resetBannerEmittedForTests export shape, the preserved production latch (bannerEmitted = true; + the early-return guard), and the unchanged hasEmittedCliBanner read-only inspector; and (b) replays the latch semantics — buggy shape (no reset → second emit silently skips), patched shape (reset between calls → both emits succeed), and the production no-reset path (latch still works → second emit suppressed). All assertions pass.

• Exact steps or command run after this patch: node /tmp/probe_83903.mjs

• Evidence after fix:

PASS: __testing.resetBannerEmittedForTests exported with the documented body
PASS: production emit-once guard preserved (bannerEmitted=true latch + early return)
PASS: hasEmittedCliBanner read-only inspector unchanged
PASS: replay (buggy / no reset): first emit succeeds; second silently skips because latch persists across tests
PASS: replay (patched / __testing.reset): both emits succeed when reset is called between them — multi-scenario tests can run
PASS: production semantics (no reset between calls): second emit still latches — no behavioral regression

ALL CASES PASS

• Observed result after fix: Production code paths are byte-identical (same latch, same early-return). Test code can import __testing to reset the latch between scenarios. The two new vitest cases exercise this end-to-end by faking process.stdout.isTTY true, mocking process.stdout.write to capture emissions, calling emitCliBanner twice, asserting the second is suppressed; then calling __testing.resetBannerEmittedForTests() and asserting the second emit succeeds.

• What was not tested: A real terminal session with the banner emitted twice — the latch behavior is the production guarantee, not something users observe at the CLI. The vitest cases drive the same code path with a faked TTY shim around process.stdout.isTTY.

Audit (per CLAUDE rules — all 5 steps)

• Existing-helper check: Mirrors the established pattern in src/cli/channel-options.ts:43-52 (__testing export with a single resetXForTests(): void member). No new pattern introduced. PASS
• Shared-helper caller check: hasEmittedCliBanner has callers in CLI bootstrap code but they only read the latch — adding a sibling reset helper is additive and doesn't change any existing call. PASS
• Broader-fix rival scan: gh pr list --search '83903 in:title,body' returns no open PRs. PASS
• Recent-merge audit: git log --oneline -5 -- src/cli/banner.ts shows e1061a8b46 test(live): tolerate provider drift in release checks — unrelated. PASS
• Prototype-pollution scan: N/A — single boolean assignment.

Changed files

• src/cli/banner.test.ts (modified, +65/-0)
• src/cli/banner.ts (modified, +13/-0)