Skip to content

compatibility.md

Latest commit

 

History

History
115 lines (98 loc) · 8.76 KB

compatibility.md

File metadata and controls

115 lines (98 loc) · 8.76 KB
Raw

Compatibility

Status of each fixture zone against reference zic (tzcode 2026b). "Byte" = our output is identical to reference zic byte-for-byte (pinned in fixtures/expected/). "Semantic" = same meaning per zdump/decoded comparison. Verified by cargo test and zic-rs compare.

Zone Source Construct Semantic Byte
Etc/UTC fixtures/minimal/utc.zi fixed offset 0, literal abbr ✅ ✅
UTC fixtures/minimal/utc.zi Link (copy) ✅ ✅¹
Test/Fixed fixtures/minimal/fixed.zi fixed offset −5:00, literal abbr ✅ ✅
Test/Simple fixtures/minimal/dst.zi finite DST, Sun>=8/Sun>=1, wall AT, %s ✅ —²
Test/Euro fixtures/minimal/euro.zi finite DST, lastSun, UT AT, %s ✅ —²
Test/Sle fixtures/minimal/sle.zi finite DST, Sun<=25, standard AT, %s ✅ —²
Test/Eastern fixtures/minimal/eastern.zi recurring DST (max), POSIX footer EST5EDT,M3.2.0,M11.1.0 ✅³ —²
Test/FF fixtures/minimal/multi_ff.zi multi-era fixed→fixed offset change at UNTIL ✅ ✅
Test/MidDst fixtures/minimal/multi_mid.zi multi-era, era ends mid-DST; footer AST4 ✅ —²
Test/FR fixtures/minimal/multi_fr.zi multi-era fixed→recurring (final-era footer anchoring) ✅ ✅⁴
Test/Q fixtures/minimal/multi_q.zi multi-era, recurring FROM 2015 → DST from era start 2000 ✅ ✅⁴
Test/Mixed fixtures/minimal/mixed.zi single-era rule set mixing finite (April onset) + recurring (March onset) rows ✅⁵ —²
Test/FinalEffective fixtures/minimal/final_effective.zi multi-era final era: effective-in-era recurring-only (finite rows pre-era) ✅ ✅
Europe/London fixtures/iana-slices/europe_london_2026b.zi first real IANA slice (T4.0): 5 eras, LMT→GMT/BST/BDST, EU footer GMT0BST,M3.5.0/1,M10.5.0 ✅⁶ —²
America/New_York fixtures/iana-slices/america_new_york_2026b.zi second real IANA slice (T4.1): 6 eras, genuinely mixed-in-era final era (US finite 1967–2006 + recurring 2007–max), WWII war time; footer EST5EDT,M3.2.0,M11.1.0 ✅⁶ —²
Test/MinRule fixtures/minimal/minrule.zi obsolete FROM=minimum → coerced to 1900 (T3.2a) ✅ —²
Test/InlineLit fixtures/minimal/inline.zi inline-save era (literal HKWT, +8:30, isdst) (T3.2b) ✅ ✅⁷
Test/InlineZ fixtures/minimal/inline.zi inline-save era, %z → +0720 from total offset (T3.2b) ✅ ✅⁷
Test/InlineSolo fixtures/minimal/inline.zi single-era inline-save (+8:30 HKWT) (T3.2b) ✅ —²

¹ The link is materialised as a copy of Etc/UTC, so its bytes match the (byte-identical) target. With --link-mode symlink it becomes a relative symlink instead — a local representation choice, not a semantic difference.

² Rule-driven zones are behaviour-match (verified by the zdump oracle); we do not claim byte parity because zic's local-time-type/abbreviation ordering and explicit-transition horizon differ from ours while the meaning is identical.

³ Verified by the zdump behaviour oracle over 2019..2099 — the wide horizon exercises the recurring POSIX footer well beyond the explicit-transition window.

⁴ Test/FR/Test/Q happen to be byte-identical because the final recurring era footer anchoring rule emits exactly the same minimal output as reference zic (one anchor transition + the recurring footer). Byte parity is incidental here, not a contract; multi-era DST zones are validated semantically.

⁵ Accepted slim/fat structural difference. Reference zic emits a slim explicit set (finite history + one recurring anchor = 39 transitions); zic-rs emits a fat set (every year through RECUR_HI = 122). Both produce identical zdump behaviour over the finite window 1976..1997 and share the footer — see the slim/fat note in reference-zic-semantics.md. Behaviour is the contract; the explicit-transition representation is not.

⁶ T4.0 — first real IANA-zone slice. Verified by the zdump behaviour oracle over 1830..2045. The slice is the canonical (record-keys-only de-abbreviated) Europe/London from tzdb 2026b; reference zic compiles it byte-identically to the unmodified abbreviated tzdata.zi extract (faithfulness proof). zic-rs's explicit-transition count (159) differs from zic's by the same slim/fat orthogonality; behaviour matches across the full horizon.

⁷ Test/InlineLit/Test/InlineZ are ttinfo byte-identical to reference zic (the local-time-type table — utoff/isdst/abbreviation — matches exactly), and zdump-identical over their horizons. Byte parity of the whole file is still not a general contract.

Europe/London — PASS (T4.0)

  • Source: canonical de-abbreviated slice derived from tzdata.zi 2026b (record keys R/Z/L expanded; month/weekday/year prefixes left in zic-style form).
  • Faithfulness: reference zic output is byte-identical to the abbreviated extract.
  • Oracle: zdump behaviour match over 1830..2045.
  • TZif: v2; 5 local-time-types (LMT, GMT, BST dst, BDST double-dst, BST fixed); 159 transitions; footer GMT0BST,M3.5.0/1,M10.5.0.
  • Why it became admissible: the final EU era starts 1996 while Rule E's finite rows end in 1995, so the effective in-era rule set is recurring-only (the effective-in-era classification law).
  • Source form (T4.0.1): Europe/London is verified both from the canonical slice and from the abbreviated tzdata.zi-style extract — zic-rs now parses zishrink record keys (R/Z/L) directly, and produces byte-identical output for both forms. The canonical fixture is retained for readability/provenance; it is no longer required to get past the parser. zic-rs can also read Europe/London straight from the installed /usr/share/zoneinfo/tzdata.zi.

Whole-database frontier (support-report) — compile-clean ≠ behaviour-verified

The rows above are individually behaviour-verified (each passes the zdump oracle over a declared horizon). For the whole installed tzdata.zi (2026b), zic-rs support-report reports a coarser, honest compile frontier:

  • 338 / 341 canonical zones are compile-clean (zic-rs emits a valid TZif); 594 / 598 identifiers including links. Only 3 fail closed: Europe/Prague (negative inline SAVE), Asia/Gaza / Asia/Hebron (non-POSIX recurring day form).
  • Compile-clean is not behaviour-verified (still tracked separately, even now that they coincide). A comprehensive zdump oracle sweep over all 341 canonical zones (1900..2040, every zone compiled and diffed against reference zic) reports 341 match / 0 mismatch / 0 fail-closed (341 + 0 = 341 compile-clean) — after T5 #1–#5 + law 7 + law 10. Every canonical zone behaviour-matches reference zic over 1900..2040 — the canonical-zone behaviour frontier is closed. The match count rose 264 → 328 → 333 → 338 → 339 → 341 as the classes were fixed (an earlier 35-zone sample "28 / 6" caught only 6 of the original 74 — retired): T5 #1 standard-LETTER, #2 final-recurring anchor, #3 standard-clock absorption (−64), #4 prior-active rule-state seeding (−5), #5 clock-reference normalization (−5), law 7 inline negative SAVE (−1: Prague), law 10 non-POSIX ON + v3 footer (−2: Gaza/Hebron, e.g. Sat<=30 → M3.4.4/50). No canonical zone fails closed. (The remaining work is the operational shell (T9–T15) operational shell** — leap/right/v4, CLI/install, -r/-R/-b, warning parity — not a canonical-zone behaviour gap; not full-zic.) Full catalogue in reference-zic-semantics.md §10.

A zone is admitted to the table above (a behaviour-verified slice) only once it passes the oracle and carries a receipt; "compile-clean" alone never earns a row. See conformance-ladder.md for the admission gate and the status vocabulary.

Not yet on this table

Inline-save eras with a %s/slash FORMAT, 24:00/negative compiled times, recurring rules whose ON is a fixed numeric day (no weekday), leap seconds, and further IANA slices are not yet supported and fail closed today. (Negative inline SAVE and Sun<=N/Sat<=N recurring ON forms are now supported — laws 7 and 10.) They join this table as the remaining breadth (T3.2) and IANA-slice expansion (T4) land — each row added only once it passes the oracle. We do not claim compatibility for any zone without a corresponding passing comparison.