RFR: 6177299: [Fmt-Nu] NumberFormat.getPercentInstance() does not work correctly

[Naoto Sato]

On Mon, 29 Sep 2025 21:38:28 GMT, Justin Lu <jlu at openjdk.org> wrote:

> Please review this PR which corrects an edge case bug for `DecimalFormat` parsing when a multiplier is applied.
> 
> This issue applies to any parsed Strings whose resultant double value is rounded to _9.223372036854776E18_ after the multiplier is applied. The returned value is incorrectly given as `Long.MAX_VALUE` when it should be returned as the double _9.223372036854776E18_.
> 
> For example, the String _"922,337,203,685,477,600,000%"_ is first parsed as _9.223372036854776E20_, after which the multiplier is then applied to give _9.223372036854776E18_. The original code evaluates `9.223372036854776E18 == (double)(long)9.223372036854776E18` as true, leading to the long representation returned.
> 
> The double value should first be checked if it is within the long min/max range before being checked if it can be represented as a long. Note that the check should be inclusive, as during the comparison, `Long.MAX_VALUE` is promoted to _9.223372036854776E18_. Thus _9.223372036854775E18_ correctly compares as false, and all doubles above compare as true.

Looks good. I think the tests can move to more generic NumberFormat regression tests, such as NumberRegression.java, as LenientParseTest sounds dedicated to `lenient parsing`.

-------------

PR Review: https://git.openjdk.org/jdk/pull/27563#pullrequestreview-3285988635