fix(engine): Baileys post-review hardening — lazy-load (no Node floor), full inbound, lifecycle + store fixes#312
Merged
Conversation
…(Approach B)
Switch @whiskeysockets/baileys from static value-imports to a private cached
dynamic import() in both BaileysAdapter and BaileysMessageStoreService. The
package now loads only when ENGINE_TYPE=baileys is selected at runtime; the
default whatsapp-web.js path never touches it.
Changes:
- baileys.adapter.ts: remove static value-import; add private `lib` field +
`loadLib()` async loader; update connect() to call loadLib() and reference
all Baileys values (default/useMultiFileAuthState/fetchLatestBaileysVersion/
DisconnectReason/getContentType) via the loaded module.
- baileys-message-store.service.ts: remove static BufferJSON import; add
`baileysLib` + `loadLib()` loader; call loadLib() inside put()/getMessage().
- package.json: remove the engines.node >=20.19.0 floor (no longer required;
Baileys loads lazily so there is no boot-time ESM require() constraint).
- package.json jest.transform: override ts-jest tsconfig to CommonJS so
dynamic import() is downleveled for the CJS jest runner, making
jest.mock/@whiskeysockets/baileys resolve correctly in unit tests.
- CHANGELOG.md: replace the Node-floor warning with a lazy-load note.
Verified: dist files contain import('@whiskeysockets/baileys') not
require(), 627 unit tests pass, 6 e2e tests pass, lint 0 errors.
…pressions) Replace `private lib: any` / `private baileysLib: any` with `typeof BaileysLib` (namespace import type — fully erased at compile time, zero dist require). Loader return types updated to `Promise<typeof BaileysLib>`. Removes all `// eslint-disable-next-line @typescript-eslint/no-unsafe-*` suppressions that the `any` fields required. Wraps `saveCreds` in `() => void` to satisfy the ev.on void-return constraint now that the call site is typed.
…on + status filter
- handleMessagesUpsert fans out to async processInboundMessage per message
- protocolMessage REVOKE → onMessageRevoked (no onMessage, no spurious DB row)
- reactionMessage → onMessageReaction (no onMessage)
- mapMessage (now async): caption falls back to body; image/video/audio/document/sticker
downloads via downloadMediaMessage (guarded try/catch — failure logs + omits media,
never throws into event loop); locationMessage populates location coords + name/address;
contextInfo.quotedMessage populates quotedMessage {id, body}
- BaileysIncomingFields gains media/location/quotedMessage; buildIncomingMessageFromBaileys
passes them through to the neutral IncomingMessage
- session.service.ts onMessage: guard isStatusBroadcast at the top (mirrors onMessageCreate)
so incoming status@broadcast posts are never persisted or webhooked
- Tests: all 634 unit + 6 e2e pass; lint clean; build clean
Wrap the entire processInboundMessage body in try/catch so a throw from any path logs-and-drops instead of becoming an unhandled rejection. Use normalizeMessageContent (available in @whiskeysockets/baileys@6.7.23) to unwrap documentWithCaptionMessage/viewOnceMessage/ephemeralMessage before extracting subMessage, so documentWithCaptionMessage now yields a non-empty mimetype and fileName. Replace the (qm.conversation as unknown as string) double-cast in the quoted-message body extraction with a properly typed intermediate. Tests: add fakeStore.put not.toHaveBeenCalled assertions to REVOKE and reaction tests; add normalizeMessageContent identity mock; add fixture test proving documentWithCaptionMessage extracts non-empty mimetype.
…nnect, first-connect error surfacing
…K CASCADE on session delete
C1: set createdAt: new Date() explicitly in put() upsert so the stored value carries
millisecond precision matching the :createdAt bound param in enforceLimit(). Without
this, SQLite's datetime('now') stores second-precision (e.g. '…:11') while the JS
Date bound serializes as '…:11.000' — SQLite string-compares '…:11' < '…:11.000' = TRUE,
so every same-second row matches the eviction predicate, wiping the store to ~0.
Reply/forward/react/delete then fail with "message not found".
Add a real put()-driven eviction regression test (6 msgs, cap=3, same wall-clock second)
that must fail on old code (count=0) and pass with the fix (count=3, C4/C5/C6 kept).
I6: add @manytoone(() => Session, { onDelete: 'CASCADE' }) to BaileysStoredMessage so the
synchronize:true SQLite path emits the CASCADE FK (the migration path already had it).
Deleting a session now cascade-deletes its stored messages on both paths. Session entity
added to the unit-test DataSource; session rows seeded before put() calls.
SQLite FK enforcement in the unit test: TypeORM sqlite driver does not run PRAGMA
foreign_keys=ON by default, so FKs are declared but not enforced — existing put() tests
pass without a parent sessions row pre-existing. Session rows are seeded anyway for
correctness (and to be safe if FK enforcement is ever enabled).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Post-review hardening of the Baileys engine, resolving 12 findings from a deep source-traced review of the 5-slice stack (#299→#310). Stacked on #310 (base
feat/baileys-contacts-chats); the diff here is the hardening only.Why this PR exists
A pre-merge multi-agent review found 2 Critical + 10 Important defects in the engine — concentrated in the inbound path, the reconnect lifecycle, the SQLite message store, and the Node-floor decision. Rather than ship the stack with them, this PR fixes all 12 (each adversarially verified, then re-verified by an Opus whole-branch review).
What changed
Engine loading — Approach B (no global Node floor)
@whiskeysockets/baileysvia a cached dynamicimport()in the adapter + message-store service, so the ESM-only package loads only when the Baileys engine is actually used. The default whatsapp-web.js engine no longer boot-loads it.engines.node >= 20.19floor and the boot-time Node guard entirely — wwjs-only operators keep any Node version. (Build verified:distemits a dynamicimport(, zero eagerrequire("@whiskeysockets/baileys").)Inbound pipeline (full parity with wwjs)
downloadMediaMessage, withnormalizeMessageContentunwrappingdocumentWithCaption/view-once/ephemeral), caption → body, location, and quoted message now populated — previously these arrived blank.protocolMessageREVOKE) →onMessageRevoked; reactions (reactionMessage) →onMessageReaction— instead of being mis-persisted as spurioustype:'unknown'message.receivedrows.status@broadcastis dropped insession.service.onMessage(mirrors the existingonMessageCreateguard) — no longer pollutes the message list/webhooks.Connection lifecycle
connect()re-checksintentionalCloseafter its awaits, so adisconnect()/logout()/destroy()during connection setup can't resurrect a stopped session with a live socket.FAILED+onError, replacing the previous unbounded tight reconnect loop; the counter resets on a successful connection and the pending timer is cleared on stop.FAILED+onError(mirrors the wwjs adapter) instead of a stuckINITIALIZING.Persisted message store
datetime('now')vs the millisecond-precision bound param → string-compare matched every same-second row).put()now writes a millisecond-precisecreatedAtso eviction keeps exactly the cap. A regression test drives realput()(would return 0 under the bug, returns the cap with the fix).@ManyToOne(() => Session, { onDelete: 'CASCADE' })relation so the default SQLitesynchronizepath emits the FK CASCADE (the migration already had it for Postgres) — deleting a session no longer orphans its stored messages.Testing
Non-blocking follow-ups (deferred)
messages.upsertbatch, webhook/ws emission order is no longer strictly arrival-order (each message carries its own timestamp; consumers shouldn't rely on intra-batch order).dist/**/baileys*.jscontainsimport(and no eagerrequire("@whiskeysockets/baileys")would guard the lazy-load invariant against future regressions.