[matt-edmondson]

Parse used CurrencyDecimalSeparator and CurrencyGroupSeparator only after it had stripped a currency symbol. Otherwise it used the number separators.

The BCL does this differently whenever the style includes AllowCurrencySymbol, which NumberStyles.Any does, and Any is what the default Parse(s, provider) uses:

• it tries the currency separators first;
• it also accepts the number separators, unless a symbol came ahead of the digits.

Some cultures write money with different separators from other numbers, such as en-DE, fr-CH and de-AT. There, PreciseNumber disagreed with decimal:

• Parse("1.25", en-DE) returned 125;
• Parse("1.5", fr-CH) threw.

Change

• A new ChooseSeparators picks a primary pair and a fallback pair. The primary pair is the currency separators when the style allows a symbol, and the number separators otherwise. The fallback pair is the number separators, used only when the style allows a symbol and none was read ahead of the digits.
• A new MatchSeparator tries the primary separator, then the fallback. The decimal separator is still tried before the group separator, and a group separator may still only follow integral digits.
• StripAffixes now reports whether the symbol was leading. A trailing symbol is read after the digits in the BCL too.

Tests

• TestParseAcceptsWhatDecimalAccepts gains two hand-built formats whose currency and number separators differ:
  • Mixed, shaped like en-DE;
  • Swiss, shaped like fr-CH.
• It gains rows for "1.5", "1.25", "1 234.5", "1,5 €", "1 234,50 CHF", "CHF 1 234.50" and "€1,5" under Any and Currency.
• TestParseRejectsWhatDecimalRejects gains "CHF 1,5": after a leading symbol, only the currency separators are read.
• TestDefaultParseReadsCurrencySeparatorsAsDecimalDoes covers the Parse(s, provider) and TryParse(s, provider, out) overloads. It uses hand-built formats, so it does not depend on ICU and runs in globalization-invariant mode.
• Without the fix, 14 of the new cases fail.

One existing test fixture changed. PreciseNumberParseCultureTests.CommaDecimal is "shaped like de-DE", but it left the currency separators at the invariant . and ,. Under Any, decimal.TryParse("1.234.567,5", NumberStyles.Any, CommaDecimal) returns false, so Parse_HonoursLeadingSignWhitespaceAndThousandsUnderNumberStylesAny was pinning behaviour the BCL doesn't have. The fixture now sets the currency separators de-DE actually uses (, and .), and the test still checks thousands grouping under Any.

The full suite passes: 604 tests. I did not run benchmarks: the change adds at most one extra separator comparison per character, and only under currency styles.

Note: #194 also edits Parse's digit loop, so whichever lands second may need a small rebase.

Fixes #195

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud