[CyberNativeAI]

ISOFormatParser accepts 1 to 4 digits of fractional seconds. A timestamp with 5 or more digits, such as the 6 that Python's datetime.isoformat() writes whenever microseconds are non-zero, does not match the ISO pattern, and other parsers pick up its pieces. There is no error, only a wrong result. On 2.10.3, run with TZ=UTC unless noted:

chrono.parse("2024-03-10T10:00:00.123456+05:30")
// two results: "2024-03-10T10:00:00" as local time (offset and fraction dropped)
// and a separate "05:30" on the reference day
// expected: one result, 2024-03-10T04:30:00.123Z

chrono.parse("2024-03-10T10:00:00.12345-05:00")
// one result: a time range "10:00:00.12345-05:00" on the reference day

chrono.parseDate("2024-03-10T10:00:00.123456Z")
// the "Z" is not consumed, so 10:00 is read as local time:
// 14:00Z under TZ=America/New_York, 01:00Z with timezone "JST"

The same happens inside running text ("job finished at 2024-03-10T10:00:00.123456+05:30 with status ok"). Timestamps with no fraction or with 3 digits parse correctly.

The parser's header cites the W3C date/time profile, where the fraction is "one or more digits", and RFC 3339 has time-secfrac = "." 1*DIGIT. new Date() parses all of the inputs above to the expected instants.

Change

• The fraction group in PATTERN becomes \\d+.
• The millisecond component takes the first 3 digits, right-padded: parseInt(fraction.padEnd(3, "0").slice(0, 3)). Without this line, .123456 would be assigned 123456 ms.

This millisecond line is the same as the one in open PR 671, which fixes the short-fraction case (.1 read as 1 ms) but keeps the 4-digit limit. So this change also makes .1 give 100 ms and .1234 give 123 ms; credit for that part goes to PR 671. The two branches merge cleanly in either order and give the same tree.

Tests

Three new tests at the end of test/en/en_inter_std.test.ts, one test(...) per input: 6 digits with +05:30, 5 digits with -05:00, and 9 digits with Z.

• TZ=UTC npx jest --ci: without the parser change only the three new tests fail (753 passed, 3 failed); with it, 756 passed. Merged with the branch of PR 671: 832 passed.
• A separate probe of 5, 6, 7 and 9 digits with Z, +05:30 and -05:00, against a Date reference and a "JST" reference, under TZ=UTC and TZ=America/New_York: 25 of 37 rows wrong on 2.10.3 and none with this change. The 12 rows with no fraction or 3 digits are correct both before and after.
• prettier --check is clean on both changed files.

Prepared by CyberNative AI LLC, an AI-run company, using AI agents.