Repository navigation
fix(i18n): spell two bn numerals in Western digits so the numerals ratchet holds - #14526
chenmingwei23 wants to merge 1 commit into
Conversation
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS Cap left at 8 with count at 7: one new Bengali-digit string passes silently, then re-breaks an unrelated PR — confirm that headroom is deliberate. Suggestions
[DESIGN-REVIEWED] f0ec2b0 |
UX Review (Fable 5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing to check. [UX-REVIEWED] f0ec2b0 |
First Principles Review (Fable 5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Verify #14529 exists and tracks driving the cap to zero — the remaining 7-violation debt and the cap's blind spot rest on that filed issue. What this change shipsInventory (2 items) — 2 justifiedIntent: FIX — turn
Both items ARE the fix; nothing rides along. The test file and its cap of 8 are untouched, so no pin is deleted or rewritten. The defect is verifiable from the repo itself: [FIRST-PRINCIPLES-REVIEWED] f0ec2b0 |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5 Review — ✅ no blocking findingsReviewed Review detailsAdvisory only — nothing blocks. FINDING — website/src/i18n/style/bnStyle.test.ts:126 — converting two values drops the live Bengali-digit count from 9 to 7 while the ceiling stays [OPUS-REVIEWED] f0ec2b0 Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
85f2e54 to
bf2a1cf
Compare
…tchet holds The bn style guard caps pre-existing Bengali-Indic numerals at 8. Two strings took main to 9, so Frontend Tests (2) is red on origin/main and every pull request whose merge ref carries the tip inherits the shard, with Coverage Gate failing closed behind it. app.metrics_card_old_data and pages.membersPage.create_name_unusable now spell 90 and 500 in Western digits, matching en.json. Nothing but digit characters changed, and the count returns to 7 against the cap of 8. Driving the cap to zero is tracked in #14529: converting the remaining seven pulls one string into the same file's zero-tolerance register rule, which is scoped to changed values and which that string already violates, so it needs a translation change rather than a mechanical one. Fixes #14506
bf2a1cf to
f0ec2b0
Compare
|
Superseded by #14518, which merged at 21:31:06Z and made the identical change: Closing rather than rebasing: there is nothing left in this branch that main does The follow-up this work turned up is still open and still accurate: #14529 records |
Problem / Motivation
Goal:
Frontend Tests (2)passes onorigin/main, so no pull request inherits ared shard for a locale string it never touched.
website/src/i18n/style/bnStyle.test.tsencodes a rule fromstyle/bn.md: the Bengalicatalog spells numbers in Western digits. It is a ratchet capped at 8 pre-existing
violations:
Main is at 9. The shard is red on
origin/main, and every pull request whose merge refcarries main's tip inherits it.
Why it matters
The failure lands on pull requests that cannot have caused it and cannot fix it.
Confirmed red on #13640, #13448, #13727, #13936, #14095 and #13374 -- including ones
whose entire diff is Python.
Coverage Gatethen fails closed onfrontend-test=failure, so one string costs every open pull request two red lanes, andPR ReadinessreportsCIas a blocking item on top of that.Re-running cannot clear it. The assertion is a count against a cap, not anything
timing-dependent: a head-verified
gh run rerun --failedon #13640 reproduced it, andno re-run on any of those six pull requests has cleared the lane.
Not a goal
Raising the cap to 9, which would clear the red while leaving the guard one string from
the next break.
Driving the cap to
toBe(0), which is the right end state but is blocked on atranslation judgement -- see #14529. Converting the remaining seven strings pulls one of
them into the same file's zero-tolerance register rule, which is scoped to changed
values and which that string already violates. Measured on a real run: converting all
nine turned this shard red on the register rule. That work wants a Bengali speaker and
should not sit on the critical path of six blocked pull requests.
Nor does this address the sibling reds on the same main wave: #14491 (
Electron Shell Tests), #14503 (Backend Tests (3.12, 2)) and #14504 (security line budget). Those arestill red here, because this pull request's merge ref carries main's tip like everyone
else's.
What changed (symptom -> root cause -> change)
The violation count went 7 -> 9 between
2c3ebcf2c7and20ed6a1dc7, crossing the cap.The two strings that crossed it now spell their numbers in Western digits:
app.metrics_card_old_datapages.membersPage.create_name_unusableNothing but digit characters changed -- the two line pairs are byte-identical once digit
characters are removed from both sides.
create_name_unusablehas an Englishcounterpart already reading
500, so the two catalogs now agree.The count is back to 7 against a cap of 8, which is where it sat before this wave.
Tests
website/src/i18n/style/bnStyle.test.ts-- 3 passed, violations 7.Frontend Tests (2)passed on CI at this change's earlier head, on a merge ref builtagainst main's red tip. That is the lane six pull requests are stuck on, so the result
is the point of this change rather than a side effect.
Mutation-proved, and the result is worth stating precisely because it is not flattering
to the cap: reverting one of the two strings leaves the shard green, since 8
still satisfies
<= 8. Only reverting both fails, withAssertionError: 9 violation(s)-- the text CI reports. So this change is what moves the verdict, and thecap cannot see a single violation. That is the argument recorded in #14529.
The file was restored from a copy taken outside the tree and verified by sha256 after
each mutation.
Any other suggestions on the work
Bundle Size Gateis red here and on #14095, reportingassets/t-*.jsover its 955.0KB budget by 28 B. It is pre-existing on main; this change only removes bytes from that
chunk, 10 raw bytes, since a Bengali-Indic digit costs 3 UTF-8 bytes and a Western digit
costs 1. Not enough to clear 28 B, and not this change's to clear -- it needs its own
issue and owner.
app.metrics_card_old_datahas no counterpart at that path inen.json, which suggeststhe key is stale or named differently in the source catalog. Out of scope here, but
worth a look by whoever owns that card.
Pattern harvest
Rule candidate: a ratchet whose cap equals its current count is indistinguishable from a
broken gate on the next commit, and it misattributes the break -- the pull request that
adds violation N+1 passes, and the failure surfaces on an unrelated pull request later.
When a mutation of the fix leaves the guard green, read that as information about the
guard rather than about the fix: here it was the signal that the cap, not the strings,
was the thing to change. Drive such a cap to zero as soon as the backlog it was holding
is empty, and if something blocks that, file it rather than leaving the cap at its own
count.
Fixes #14506