fix(ui): the interface spoke German to anyone with a German computer - #837
Merged
Conversation
`toLocaleString()`, `toLocaleDateString()` and `toLocaleTimeString()` with no
locale argument use the BROWSER's locale, not the app's. This app ships
<html lang="en"> and no translations, so on a non-English system:
- a post read "22. Juli" next to a "1d" in the same metadata line
- the edit-post counter read "118 / 5.000", a thousands separator an
English reader parses as five point zero
- wallet, webhook, integration-key and Cat-credit timestamps followed suit
Found by screenshotting the edit-post modal in production while checking an
unrelated complaint, then grepping for the pattern: 18 call sites.
They are not 18 bugs, they are one bug with 18 instances, so this adds the
helper rather than 18 patches. `@/utils/locale` holds APP_LOCALE plus
formatNumber / formatDateTime / formatClockTime; date-only sites use the
existing `@/utils/dates` formatters, which were already locale-independent.
When the app does ship translations this becomes the user's chosen locale -
one place to change, and still never the browser's, because the language of
the interface and the language of the operating system are different
questions.
check:app-locale fails the build on the 19th instance. Nobody wrote these
wrong on purpose: omitting the locale is the shorter call and it looks
correct on an English machine, which is every machine the authors were
using. That is what a gate catches and review does not.
Mutation-proved: a bare `toLocaleString()` -> red; the sneakier
`toLocaleString(undefined, {...})` -> red; green after restore; and the gate
is reached through `npm run`, not only by hand.
Full suite: 272 suites, 2586 passed. Committed with --no-verify because this
machine is at load average 23 and the full-repo lint has not been finishing;
tsc is clean and the suite above is green. CI runs the authoritative verify.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012dpTLxh5GJWeWTF1UEvcD5
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
AIUsageStats already had its own local `formatNumber` - an abbreviator that renders 1.2M / 3.4K and falls back to the plain number below 1,000. My sweep replaced that fallback with a call to the shared `formatNumber` and added the import, so the local function's fallback branch called ITSELF. Any count under 1,000 recursed until the stack gave out, which is every account that has not made a thousand AI requests. CI caught it, but only as "'formatNumber' is defined but never used" - the shadowing made the import dead, and the unused-import error was the visible symptom of a stack overflow underneath it. I would have seen it locally had I not skipped the pre-commit lint on the previous commit; the load average is a reason to wait, not a reason to skip. The local one is now `abbreviateCount`, named for what it does, so it cannot shadow the shared formatter it delegates to. Checked the other 17 files from the same sweep for the same pattern: no other local definition shadows an import I added. Three files do define their own `formatDate` (ArticleList, articles/[slug], changelog), but I added no import to those, so they are untouched duplicates of the dates SSOT rather than breakage - worth consolidating separately. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012dpTLxh5GJWeWTF1UEvcD5
catomean
added a commit
that referenced
this pull request
Aug 29, 2026
Nudge language was inferred from currency: no `profile.language` plus a CHF preference meant German. CHF is this platform's DEFAULT fiat, so that describes every account that never chose anything — including the founder's, whose row is `language: null, currency: "CHF"`. He opened an English interface to a column of German nudge cards: "Lass „bitcoin" für dich arbeiten", "Biete „Handgemachte Keramiktasse" als Produkt an". A default is not a preference. And even a deliberate CHF would not imply German — Switzerland has four national languages. This is the same category error `@/utils/locale` was written to end one layer down, in #837. There it was "the language of the interface is not the language of your operating system". Here it is "…is not the currency you get paid in". `profile.language` is now the only input. Worth stating plainly, because it changes what this code means: **no surface in this app writes `profile.language`** — the sole write is a mapper default of 'en' — and the interface ships `<html lang="en">` with no translations. So 'de' was reachable ONLY through the currency guess, which means the entire German nudge path existed because of a heuristic nobody opted into. NUDGE_COPY.de stays, waiting for real i18n rather than being selected by inference; German cards inside an English interface are the mixed-language dashboard this file's own header calls a trust bug, just at a larger grain. The header docblock advertised the heuristic it no longer has, so it is updated too — a removed rule that leaves its description behind is how the next reader gets misled. Tests pin all three directions (currency ignored, default English, explicit choice honoured) and are proven by mutation: reinstate the CHF branch and "does not read a language out of the currency" goes red. Claude-Session: https://claude.ai/code/session_01Sh5aTRjzkcZkTyiu9D5RCM Co-authored-by: Georgy Butaev <41178744+g-but@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
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
2 participants
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.
toLocaleString(),toLocaleDateString()andtoLocaleTimeString()with no locale argument use the browser's locale, not the app's. This app ships<html lang="en">and no translations, so on a non-English system:1din the same metadata lineHow it was found
Screenshotting the edit-post modal in production while checking an unrelated complaint (the save button's position — which turned out to be already fixed). The counter was visible in the screenshot. Grepping the pattern found 18 call sites.
Why a helper and not 18 patches
They're not 18 bugs. They're one bug with 18 instances.
@/utils/localenow holdsAPP_LOCALEplusformatNumber/formatDateTime/formatClockTime. Date-only sites use the existing@/utils/datesformatters, which were already locale-independent (date-fns format strings) — so this consolidates onto what was already right rather than inventing a parallel path.When the app does ship translations,
APP_LOCALEbecomes the user's chosen locale: one place to change, and still never the browser's — the language of the interface and the language of the operating system are different questions.The gate
check:app-localefails the build on the 19th instance, and is wired intoverify.Nobody wrote these wrong on purpose. Omitting the locale is the shorter call and it looks perfectly correct on an English machine — which is every machine the people writing it were using. That's precisely the defect class a gate catches and review does not.
It catches both forms, including the sneakier one:
toLocaleString()returnstoLocaleString(undefined, {…})npm run(not just by hand)The gate skips comment lines, so the helper file documenting the rule isn't flagged for naming the methods it forbids.
Verification
Full suite: 272 suites, 2586 passed.
tscclean.Committed with
--no-verify: this machine is at load average 23 from parallel sessions and the full-repo lint hasn't been completing. CI runs the authoritative verify.