fix: match by Unicode codepoint so a lone surrogate cannot match half a pair [patch]

[matt-edmondson]

Fixes #71

What was wrong

Fuzzy.ContainsCore and Fuzzy.CalculateScoreCore advanced one UTF-16 code unit at a time, with no awareness of surrogate pairs. An unpaired surrogate in the pattern was therefore an ordinary matchable character and could match the corresponding half of an unrelated supplementary-plane character in the subject:

Fuzzy.Contains("x\U0001F601y", "\uD83D"); // → true, before this change

The fix

Both loops now advance one Unicode codepoint at a time:

• CodepointLengthAt reports whether an index starts a well-formed surrogate pair (2 code units) or not (1). An unpaired surrogate is a codepoint of its own, so it is never treated as half of a neighbouring character.
• CodepointsEqual compares the codepoints at two indices, treating codepoints of differing code-unit length as unequal — that length check is what stops the spurious half-match.
• The scoring loop tracks the best matched codepoint by index and length rather than by char, so a surrogate pair is scored as one character instead of two.

char.IsHighSurrogate / char.IsLowSurrogate are used rather than System.Text.Rune, which is not available on the netstandard2.0 / netstandard2.1 legs.

Behaviour notes

• BMP text is unchanged. Every BMP character occupies a single code unit, so both loops iterate exactly as they did before — same matches, same scores.
• Supplementary-plane codepoints are compared exactly, which is also the existing behaviour: char.ToLowerInvariant operates on single UTF-16 code units and has no case mapping to apply to a surrogate. Case folding for supplementary-plane characters is out of scope here.
• No public API change.

This issue was previously triaged as a follow-on to #70; that issue was resolved by PR #72 via NFC normalization rather than codepoint enumeration, so this gap remained and is addressed here on its own terms.

Testing

Added a Surrogate Pair Tests region to FuzzySearch.Test/FuzzyTests.cs covering the issue's acceptance criteria — an unpaired high surrogate, an unpaired low surrogate, and the scoring overload against a subject containing U+1F601 adjacent to other matchable characters, plus regression guards that a pair still matches itself, that an unpaired surrogate still matches the same unpaired surrogate, that a whole pair does not match an unpaired surrogate in the subject, and that two emoji sharing a high surrogate are not confused.

Verified by reverting FuzzySearch/Fuzzy.cs to main with the new tests in place: 3 of the new tests fail (Contains_LoneHighSurrogatePattern_DoesNotMatchHalfOfASurrogatePair, Contains_LoneLowSurrogatePattern_DoesNotMatchHalfOfASurrogatePair, Contains_WithScore_LoneHighSurrogatePattern_IsNotReportedAsPresent). With the fix restored, all 42 tests pass, and dotnet build -c Release succeeds with 0 warnings across all five target frameworks.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Rcu8VsXN5GQsBmuvn6NQ6y

---

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