[stevehu]

Correct the time changelog entry: missing seconds were already rejected. Describe the RFC 3339 fallback in README and document its space-separator and leap-second-table differences from Ethlo.

Share UTC minute normalization between time and date-time validation, make seconds mandatory in the formatter, and reuse the parsed date for leap-second checks. Add regression coverage for the space-separator difference, fixed-width fields, extreme-offset leap seconds, and accepted/rejected leap seconds through schema validation with Ethlo absent.

Initial implementation validation: mvn -B verify passed on JDK 17, 21, and 25: 9,105 tests per run, zero failures/errors, 30 skipped. Focused format tests passed (103 cases). Refreshed GitNexus change scan: LOW risk, seven changed files.

Review follow-up: document Ethlo's rejection of -00:00, offsets beyond ±18:00, and fractions longer than nine digits, and restore the leap-second accuracy caveat. Add five date-range boundary cases, including a written year 0000 that normalizes to UTC year -1. The final format suite passes 108 tests with zero failures/errors/skips on JDK 25. Full-suite results above apply to the initial implementation; the docs-and-tests follow-up was checked with the focused suite. GitNexus follow-up scan: LOW risk, three changed files.

Refs #1337. Companion PR for 2.x: #1339.

[stevehu]

I found no bug in #1338 or #1339, and both are safe to merge. I reported 5 low-severity findings, fewer than the 7-finding target, because I didn't want to invent more.

• Same code on both branches: The two PRs change the same lines, and the three format source files are byte-identical.
• Behavior unchanged: I fuzzed time and date-time with about 900K distinct inputs each. The new code gave exactly the same results as current master (which has Validate the date-time format without Ethlo with the time format rules (#1316) #1335 merged), and it matched my independent RFC 3339 reference with 0 mismatches. JDK 8 gave the same results as JDK 25.
• Tests: The format tests on the 2.x branch pass on JDK 8 (103 run, 0 failures). I didn't re-run the full suite. The PR descriptions report full verify runs on JDK 8 (2.x), 11, 17, 21 and 25.
• Issue Follow-ups to the date-time fallback (#1335, #1336): CHANGELOG, README and cleanup #1337 coverage: The CHANGELOG fix, the README rewrite, the space-separator note, the shared UTC normalization, the mandatory-seconds formatter and the single date parse are all in, and they are correct.

Findings:

• Worth fixing: The README and CHANGELOG now name two differences from Ethlo, the space separator and the leap-second table. They don't say that Ethlo also rejects -00:00, offsets past ±18:00 and fractions over 9 digits, which the fallback accepts. The edit also dropped the old caveat that excluding Ethlo means less accurate validation.
• Minor:
  • A new test pins Ethlo's behavior, so a dependency bump could break it.
  • Two substring allocations per call remain.
  • 24 * 60 is repeated across two files.
  • A leap second at 0000-01-01T00:59:60+01:00 shifts the UTC date into year -1, and no test covers either end of the date range.

[stevehu]

I checked the pushed diffs of #1338 and #1339. I did not re-run the tests or check CI, so I'm relying on your 108-test result on JDK 25 and Java 8.

Fixed:

• Ethlo differences: The README now says the fallback accepts -00:00, offsets beyond ±18:00 and fractions longer than nine digits, which Ethlo rejects. The space-separator note stays.
• Accuracy caveat: The old "accurate validation" wording is replaced by "less accurate leap-second validation".
• Year-boundary tests: Both ends of the range are now covered: 0000-01-01T00:59:60+00:59, 0000-01-01T00:59:60+01:00, 9999-12-31T23:59:60-00:01 and 9999-12-31T23:59:60Z.

Skipped:

• Test pinning Ethlo behavior: ethloAcceptsSpaceSeparatorButFallbackRequiresT is still in the PRs. If an itu bump changes that behavior, the test fails in an unrelated PR. I'd keep it, since it documents a stated difference.
• Substring allocations and 24 * 60: Production code is unchanged, so both remain. They were optional.