Found by the new #288 post-deploy @smoke suite (PR forthcoming) — it did exactly its job: caught a real, deployed-but-broken defect that every other test is green through.
Symptom
The live production sign-in page throws a JavaScript parse error:
pageerror: Invalid or unexpected token
What I ruled out (repro notes)
- The page still renders (the "Continue with GitHub/Google" buttons are visible), apex serves
200, and OAuth client_ids are valid — so it's a partial client break, not a white screen.
- Loading each of the 13
_next/static/*.js chunks standalone via <script src> does NOT reproduce the error. So the offending script is not a standalone static chunk — it is most likely an inline hydration script in the page HTML (self.__next_f/flight data) or a dynamically import()-ed chunk whose content carries the bad escape. Next step for the fixer: capture the page's inline <script> bodies + the dynamic-import chunks and strict-parse each ("use strict" / module mode) to find the \0<digit> (or raw NUL) token, then trace it to the source construct / dependency and re-apply/adjust the SWC override.
Impact
A deployed JS parse error on the sign-in page — the #287-class "the product a customer touches is broken while every test is green." Priority to (a) confirm what functionality is actually dead, (b) fix the octal-escape source, (c) add a build-time detector that parses chunks the way the browser does (strict/module), since the current sloppy-mode check-chunks-parse.mjs misses it.
Relationship to the @smoke suite
The @smoke suite (#288 item 2) ships with its "sign-in loads with zero console errors" check skipped, annotated with this issue number. Un-skip that check when this is fixed — a green smoke run then re-confirms the fix on the live origin. The suite's other checks (providers enabled, OAuth client_id validity, canonical redirects) gate immediately.
Found by the new #288 post-deploy @smoke suite (PR forthcoming) — it did exactly its job: caught a real, deployed-but-broken defect that every other test is green through.
Symptom
The live production sign-in page throws a JavaScript parse error:
https://scripthammer.com/sign-in/.pageerrorhas an empty stack (the script failed to parse before any statement ran).node's (sloppy-mode) parser accepts the source, so the build'scheck-chunks-parse.mjsgate and the CI build both stay green — this is why it reaches prod undetected. This is the Adding cesium inlines a WASM module as a template literal → vendor chunk fails to parse → client JS dead app-wide #294 SWC-octal class (a\0<digit>octal escape that is aSyntaxErrorin strict/module/template context but not in a sloppy classic script) — TECHNICAL-DEBT [Gap-Audit] Stability: PaymentConsentModal Firefox focus timing + imperative/declarative hybrid #11, previously worked around via a@spz-loader/core@0.3.0pnpm override. This is a regression or a new instance of that class.What I ruled out (repro notes)
200, and OAuth client_ids are valid — so it's a partial client break, not a white screen._next/static/*.jschunks standalone via<script src>does NOT reproduce the error. So the offending script is not a standalone static chunk — it is most likely an inline hydration script in the page HTML (self.__next_f/flight data) or a dynamicallyimport()-ed chunk whose content carries the bad escape. Next step for the fixer: capture the page's inline<script>bodies + the dynamic-import chunks and strict-parse each ("use strict"/ module mode) to find the\0<digit>(or raw NUL) token, then trace it to the source construct / dependency and re-apply/adjust the SWC override.Impact
A deployed JS parse error on the sign-in page — the #287-class "the product a customer touches is broken while every test is green." Priority to (a) confirm what functionality is actually dead, (b) fix the octal-escape source, (c) add a build-time detector that parses chunks the way the browser does (strict/module), since the current sloppy-mode
check-chunks-parse.mjsmisses it.Relationship to the @smoke suite
The @smoke suite (#288 item 2) ships with its "sign-in loads with zero console errors" check skipped, annotated with this issue number. Un-skip that check when this is fixed — a green smoke run then re-confirms the fix on the live origin. The suite's other checks (providers enabled, OAuth client_id validity, canonical redirects) gate immediately.