Temporal oracle
temporal-oracle.ymlEvery Monday at 06:17 UTC
Next run
Last 8 · latest Failed, 16 November
⟨ NO SIGNAL — ⟩
Workaround check (Node)
Each workaround's probes on Node: is the polyfill defect still there?
Lists each workaround that every engine in its scope no longer needs in the workflow's rolling issue.
- Watching, now
- Fixed upstream
- Retired
Retires with its workarounds: 29 still needed, 0 removable, 1 guard, 4 only on some engines.
| Defect | What it covers | Still needed? |
|---|---|---|
| D1 | calendar fields → ISO, add, until and Duration relativeTo near the limits (temporalCompat) | ✓ Yes |
| D2 | buddhist read through ICU4C's Julian/Gregorian hybrid | ✓ Yes |
| D3 + D4 | Hebrew years <= 0 (polyfill leap-year %, ICU4C one day off) | ✓ Yes |
| D5 | Indian before ISO year 1 (stale V8 bug 10529 detector) | ✓ Yes |
| D6 | non-ISO until re-constrains the day while counting months | ✓ Yes |
| D7 | Hebrew until by years throws mixed-sign (tc39/proposal-temporal#3159) | ✓ Yes |
| D8 | japanese pre-proposal era codes | ✓ Yes |
| D10 | fields → ISO skips a 5–6-day month 13 in far years (tc39/proposal-temporal#3329; ethioaa, so coptic and ethiopic) | Kept on purpose |
| D11 | the calendar nudge window is never retried, so total, round and until with a calendar smallestUnit answer over the wrong bounds (relativeTo on the 29th-31st) | ✓ Yes |
| D12 | a zoned string's offset written with seconds is matched by minutes instead of exactly (zonedWallClock.ts defect 4) | ✓ Yes |
| D13 | two offset changes inside one 14-day transition search step are skipped, or stall the search for ever (zonedWallClock.ts defect 5) | ✓ Yes |
| D14 | ZonedDateTime#round to the day does not clamp an instant in the second piece of a date that starts twice (zonedWallClockOperations.ts roundZonedDateTime) | ✓ Yes |
| D15 | `Z` string read by wall clock under offset: "ignore" instead of as an exact instant (zonedWallClock.ts defect 6) | ✓ Yes |
| D16 | formatRange drops O and long zone-name widths (ICU-20548) | ✓ Yes |
| D17 | formatRange writes AM/PM for a flexible dayPeriod next to a timeZoneName (SpiderMonkey) | ✓ Yes, on Firefox |
| D18 | Islamic calendar dates before the Hijra read as positive years in a second era (WebKit) | ✓ Yes, on WebKit |
| D19 | ZonedDateTime#equals answers false for two identifiers of one zone where resolvedOptions keeps the identifier (JavaScriptCore) | ✓ Yes, on WebKit |
| D21 | Temporal.Now reads one millisecond early after a clock before 1970 (temporalCompat/clock.ts) | ✓ Yes |
| D22 | a zoned relativeTo or start in a repeated hour is measured from the wrong instant, so total, round and until in a date unit answer wrongly from the later pass, and wrongly or not at all from the earlier pass when the end falls in the later pass at a lower wall clock; where a TC39 Assert fails no value exists (tc39/proposal-temporal#3336), including after a skipped hour, and GMT returns its sentinel where the polyfill extrapolates | ✓ Yes |
| D23 | formatToParts and formatRangeToParts abort V8 on a week-of-year year field (nodejs/node#67076) | ✓ Yes |
| D24 | CLDR date patterns print the week-of-year year where the calendar year is meant (internal/weekYearGuard.ts) | ✓ Yes |
| D25 | the japanese calendar prints Japanese eras before 1873-01-01 (internal/yearPrinting.ts) | ✓ Yes |
| D26 | an islamic calendar prints a negative year before the Hijra (internal/yearPrinting.ts) | ✓ Yes |
| D27 | ja japanese leaves the era year unpadded under year 2-digit and in Latin digits under another numbering system (internal/yearPrinting.ts, internal/japaneseNumerals.ts) | ✓ Yes |
| D28 | chinese and dangi pad the related year under year 2-digit (internal/yearPrinting.ts) | ✓ Yes |
| D29 | a range drops, repeats or garbles a field of one date (internal/rangeFields.ts) | ✓ Yes |
| D30 | islamic and islamic-rgsa resolve as written (internal/formatterCache.ts) | ✓ Yes |
| D31 | a coptic date before AM 1 prints 1 - year and no era (internal/copticEra.ts) | ✓ Yes |
| D32 | the Islamic era changes in the middle of 1 Muharram of year 1 (internal/islamicEpochEra.ts) | ✓ Yes, on WebKit |
| D9 | non-ISO months added and counted one month at a time (heap OOM for in-range amounts; temporalCompat/largeMonthSpan.ts) | ✓ Yes |
| zoned.A | wall clock → exact time at the range limits (zonedWallClock.ts defect 1, zonedWallClockDifference.ts defect 1) | ✓ Yes |
| zoned.B | next transition missed near the maximum (zonedWallClock.ts defect 2) | ✓ Yes |
| zoned.E | transition search floored at 1847-01-01 (zonedWallClock.ts defect 3) | ✓ Yes |
| zoned.D | UTC fast path skips IsValidEpochNanoseconds (zonedWallClockDifference.ts defect 2) | ✓ Yes |
Native oracle
GMT's calendar results against Chromium's built-in Temporal. The test262 suite decides a disagreement.
Lists the mismatches no workaround accounts for in the same issue.
- Watching, now
- Fixed upstream
- Retired
Retires when every engine ships Temporal and the polyfill is removed from GMT (#316).
Native suite
The whole library test suite on built-in Temporal and on the polyfill, in Node, Chromium, Firefox and WebKit. Each browser also runs the workaround probes.
Lists, in the same issue, a section per engine for a difference no workaround accounts for, one for a host that failed and one for a stale tag.
- Watching, now
- Fixed upstream
- Retired
Retires when every engine ships Temporal and the polyfill is removed from GMT (#316).
Polyfill swap
The suite on the newest `temporal-polyfill`, beside `@js-temporal/polyfill`. The Node run also sweeps every zone's offsets and every calendar's years against Node.
Adds a section when every run passes: the swap is due.
- Watching, now
- Fixed upstream
- Retired
Retires when the fixes GMT waits on have shipped in `temporal-polyfill` and GMT has moved to it. Waiting on fullcalendar #127 fullcalendar #128 fullcalendar #131 fullcalendar #132.