[MPCoreDeveloper]

Why

Sonar flagged new-code duplication in the 2.1.2 diff above the 3 % ceiling. The duplicated blocks were real: the prepared-insert path carried five copies of two algorithms, and the copies had already drifted apart in behaviour.

What changed — one algorithm per concern

Concern	Single source of truth	Copies removed
Quote/literal scanning	SqlLiteralScanner	the two inline scanners in Database.Batch.cs, behind a new TrySkipLiteral helper
VALUES splitting	ValueRangeEnumerator (ref struct, yields [start,end))	the two identical loops in ParseValues / ParseValuesToArray
? placeholder positions	ParameterBinder.ExtractPositionalParameterPositions	a byte-for-byte copy in PreparedStatement

The enumerator is a ref struct on purpose: the prepared-insert path must not allocate per row, so the dedup costs no allocation. It keeps the old semantics exactly — single-quote literals skipped as a whole through SqlLiteralScanner, () and [] (VECTOR literals) suppressed, top-level comma splits, same bounds guards.

Defect the dedup exposed, and fixed

The prepared/batch path stopped counting a tuple at the table's column count:

CREATE TABLE docs (id INTEGER PRIMARY KEY, name TEXT, score REAL);
INSERT INTO docs VALUES (1, 'a', 0.5, 21);   -- 4 values, 3 columns

It bound the first three values, discarded 21 and succeeded. The statement path has rejected that shape since [2.1.2], so the two disagreed. Surplus values now throw (InvalidOperationException; the message names the table, the tuple and both counts), and a short tuple keeps its documented meaning (missing columns fall to NULL). This was the last copy of the known-issue-7 truncation class.

Tests

• New PreparedInsertValuesParsingTests (11 cases): every case is parsed through both prepared parsers and asserted to agree — neither had any coverage before, which is how they could drift. Covers quoted commas, doubled quotes, backslashes, VECTOR literals, NULL, short tuples and the surplus-tuple rejection.
• Three fixtures had been passing because of the truncation: FixedWidthBulkUpdateTests (five DDLs realigned to the 4-column docs shape its own helper inserts into), FormatCompatPolicyTests and DefaultEngineSelectionTests. The latter two also rendered i * 0.5 as 0,5 on this nl-NL machine — two SQL values — and now format invariant, as EncryptionCoverageTests and WritePathProfilerTests already do.

Verification

• Core suite on this exact tree: 2118 / 0 failed / 0 skipped (Category!=Debug&Category!=Manual&Category!=Performance), build clean on net11.0.
• Canaries green: ReopenRoundTripMatrixTests, FormatCompatPolicyTests, SingleFileDirectoryParityTests, FixedWidthBulkUpdate/Delete/PatchTests, WritePathProfilerTests, FixedWidthInlineValueTests.Reopen_KeepsInlineAndOverflowValues.
• The §7 reproduction published in docs/sharpcoredb-known-issues.md still prints the documented fixed output for all five bound-value shapes (re-run live against this tree).
• CHANGELOG.md gains Fixed and Changed entries under [Unreleased] — the strictness is user-visible.

No API change, no format change, no durability or encryption behaviour touched.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

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

See analysis details on SonarQube Cloud

[codecov]

Codecov Report

❌ Patch coverage is 77.41935% with 14 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...c/SharpCoreDB/Database/Execution/Database.Batch.cs	78.68%	7 Missing and 6 partials ⚠️
src/SharpCoreDB/Services/PreparedStatements.cs	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!