Preserve leading fractional zeros in plural operands - #1342

Is there a related issue, or how did you come across this? Could you show another compliant CLDR implementation producing v = 3, w = 3?

Thanks for asking. I don't have a related issue to link. I'm working on a project that automates finding and submitting bug-fix PRs and compatibility issues to GitHub projects with active contributor communities. This PR came out of that workflow, which checked extract_operands() against the CLDR operand definitions and ran local regression tests.

The specific discrepancy is that Decimal('0.001').as_tuple() has digits=(1,) and exponent=-3, so counting the stored fractional digits loses the leading fractional zeros.

I have now independently checked ICU4J 78.1 on Java 21. It produces v=3, w=3 for 0.001. Here is the reproducer, using ICU's number formatter to produce the operands:

import java.math.BigDecimal;
import com.ibm.icu.number.NumberFormatter;
import com.ibm.icu.number.Precision;
import com.ibm.icu.text.PluralRules;
import com.ibm.icu.text.PluralRules.Operand;
import com.ibm.icu.util.ULocale;

class CheckOperands {
    public static void main(String[] args) {
        var rules = PluralRules.forLocale(new ULocale("lv"));
        for (String s : new String[]{"0.001", "0.00100", "0.011", "0.11", "0.000"}) {
            var input = new BigDecimal(s);
            var formatted = NumberFormatter.withLocale(ULocale.ROOT)
                .precision(Precision.fixedFraction(Math.max(0, input.scale())))
                .format(input);
            var operands = formatted.getFixedDecimal();
            System.out.printf("%s: v=%.0f w=%.0f f=%.0f t=%.0f lv=%s%n",
                s, operands.getPluralOperand(Operand.v),
                operands.getPluralOperand(Operand.w),
                operands.getPluralOperand(Operand.f),
                operands.getPluralOperand(Operand.t), rules.select(formatted));
        }
    }
}

With the ICU4J 78.1 JAR, run java -cp icu4j-78.1.jar CheckOperands.java:

0.001: v=3 w=3 f=1 t=1 lv=one
0.00100: v=5 w=3 f=100 t=1 lv=other
0.011: v=3 w=3 f=11 t=11 lv=one
0.11: v=2 w=2 f=11 t=11 lv=zero
0.000: v=3 w=0 f=0 t=0 lv=zero

The explicit fraction precision preserves the visible digits of the input, including trailing zeros, matching the Decimal inputs in the PR. The operand inspection APIs emit deprecation warnings in ICU4J 78.1; the example above ran successfully. Plural selection uses select(FormattedNumber).

All five operand/category results match this PR. I also reran tests/test_plural.py: 49 passed.

The original investigation, this cross-check, and this reply were prepared with Codex assistance.