Repository navigation
Tags
Tags
Refresh corrected definitions to tzdata 2026e and add Winnipeg/Inuvik (… …#2) * Refresh the corrected definitions to tzdata 2026e and add Winnipeg/Inuvik Northwest Territories (America/Inuvik, tzdata 2026d) and Manitoba (America/Winnipeg, tzdata 2026e) adopted permanent daylight time from 2026-11-01, so 0.3.62's bundled data falls back where they no longer do. America/Rainy_River and Canada/Central are linked_timezone entries pointing at America/Winnipeg, so they follow without their own files. Regenerated all 25 definitions from tzinfo-data 1.2026.5 so every shipped file comes from one snapshot. Of the 23 already corrected, 22 change only in the header's recorded gem version; Asia/Tehran's 1979-05 transition moves one day earlier (296598600 -> 296512200), which is upstream's correction and ships with the refresh. Also commits the generator the file headers have always named but which was never in the repo. It reproduces the 23 definitions generated for DEV-9301 from tzinfo-data 1.2026.3 byte for byte, which is how this refresh was validated. Verified against the JRNI contract: 157 assertions over 25 zones and 13 aliases, including that every affected zone's final period stays open-ended with std_offset 0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Bump version to 0.3.62.2 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * [self-review-1] Address own review findings on the 2026e refresh Regenerate America/Bogota from the 1.2026.5 snapshot so the corrected set is uniformly tzdata 2026e. Its 1992 transition moved a day earlier upstream between 1.2026.3 and 1.2026.5, and comparing the two releases' full definition trees shows it was the only in-window zone the refresh missed: Europe/Dublin changes only in 1925, and CST6CDT/EST5EDT/MST7MDT/PST8PDT differ from this fork only below 1970 and above the 2070 ceiling. Correct the generator header. The double-quote normalisation is this fork's convention for the corrected definitions, not a 0.3.x dialect difference -- upstream uses single quotes in both. Document OUT_ROOT, and name the manifest that the GEM_VERSION clause defers to. Record in jrni/diff-report.md what the 1.2026.5 refresh supersedes, leaving the 1.2026.3 stamp as the honest record of when the sweep ran. Three rows are re-annotated rather than recomputed, and the note names the sections that still hold the 1.2026.3 set. Known gap: the downstream manifest has no America/Bogota entry and nothing enumerates this directory, so no spec exercises the new definition yet. The manifest entry and the CORRECTED_ZONES breadth pin follow with BookingBug/bookingbug#13057. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * [self-review-2] Record the held identifiers and guard OUT_ROOT The supersession note recorded what the 1.2026.5 comparison corrected but not what it examined and deliberately held, leaving those five decisions only in the PR description. Europe/Dublin differs solely in 1925; the four POSIX identifiers flipped back to standalone timezone definitions in 1.2026.5, matching the kind this fork already ships, and differ from it only outside the 1970-2070 window. The note names rows 375, 389, 521 and 529 rather than the shared "linked in the vendored release" wording, which is still accurate for CET, EST, HST and MST. The generator guarded its inputs but not its output, so a mistyped OUT_ROOT materialised a parallel tree and exited 0, leaving the operator to believe the refresh had landed. The guard reads "is not a directory" rather than "does not exist" so it is also true when the path exists as a file. It sits after the usage abort because File.directory?(nil) raises and a nil out_root is only reachable with fewer than four arguments, which that abort already rejects. All 26 corrected definitions still regenerate byte-identical from tzinfo-data 1.2026.5, and the suite is unchanged at 271 tests / 2393 assertions. Two findings from the same review are deferred to DEV-10037: stamping provenance from argv rather than the source snapshot's version.rb, and the absence of any check on the reproduction claim. Neither has a runtime effect, and the data itself is guarded by the consumer's specs. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Add the 23-zone derivation report from BookingBug PR #12847 Documents how the emitted set was derived (sweep of all identifiers, resolved-offset comparison 1970-2070) and why each excluded zone was excluded (Morocco regression, abstract-alias zones, pre-epoch-only differences). Provenance documentation for the data commit below. Claude-Session: https://claude.ai/code/session_014FkyUqPaFbvHZmprq4jnMn
PreviousNext