fix: preserve unsupported year tokens in format (#3015)

[Ryokki]

Summary

This changes format() so unsupported year tokens are preserved literally instead of falling through to the timezone-offset fallback.

Closes #3015.

Root Cause

REGEX_FORMAT matches Y{1,4}, but format() only implements YY and YYYY.

Previously, unmatched tokens fell through to:

matches(match) || zoneStr.replace(':', '')

That fallback was intended for ZZ, but it also affected unsupported tokens such as Y and YYY, causing them to be formatted as timezone offsets.

Changes

• handle ZZ explicitly in format()
• preserve unsupported matched tokens literally
• add regression coverage in display.test.js for:
  • Y
  • YYY

[iamkun]

Perhaps it would be better to update the REGEX_FORMAT regex constant to YYYY|YY to address this issue at the source.
A copy of the current REGEX_FORMAT could be preserved in src/plugin/duration/index.js.
This would fix the problem at its root and keep the implementation simpler and cleaner.
The current test cases could remain as they are.

[Ryokki]

Thanks, I agree that tightening REGEX_FORMAT is the better fix here, since it resolves the mismatch at the source instead of handling it later in format().

I also agree that src/plugin/duration/index.js may need to keep its own copy of the current regex, because its supported year tokens differ from core format().

One point I want to clarify is the expected behavior for YYY once the core regex is narrowed to YYYY|YY.

My current preference is to treat YYY as YY token plus a literal Y, rather than preserving the whole sequence as literal YYY. That seems like the simpler rule: supported tokens are matched normally, and only the unsupported remainder falls back to literal text.

If you agree with that interpretation, I can adjust the implementation and tests in that direction.

[iamkun]

Thanks for the clarification. I agree that interpreting YYY as YY + a literal Y makes sense. It keeps the rule simple: supported tokens are parsed normally, and the remaining characters fall back to literal text. That behavior looks reasonable to me.

[Ryokki]

Thanks for confirming. I’ve updated the implementation accordingly.

[iamkun]

THX