desktop/src-tauri/crates/core/tests/kimi_code_incremental_v12.rs

Comment 1, line 1

//! Kimi Code (A12, M4) V12: repeat deduplication, appended reads and persisted reconciliation,

//! partial lines, truncation rescans, bounded batches with resumption and renamed-file detection.

/// Base file: three lines, metadata and two turn records

/// {100,50,400} and {200,60,800}, with totals 550/1060.

/// Appended reads add only new records and preserve cumulative reconciliation across rounds.

/// Echo starts at zero; the appended echo still covers only part of the record sum.

// Base has no step.end: two usage records and zero echo give echo_subset.

// Append one usage.record and a step.end echo matching the first existing usage record.

// Persisted sums: three records 550+1060+1270=2880 versus echo 550, giving echo_subset.

/// Leave a final line without newline unread; consume it after the newline is appended.

// Metadata and the first record are complete; the last record has no newline.

/// Truncation increments generation and rescans; stored history survives source truncation

/// under V12, as for Codex. Rescanning adds only events with new keys.

// Rewrite with metadata and one different usage record.

// Keep two historical events and add one new event despite source truncation.

/// Stop at a complete line on reaching the line limit; resume from the consumed-line cursor.

// Limit this round to two lines: metadata and first record; status budget_exhausted.

// Reconcile only at EOF; this limited round produces no reconciliation row.

// The next normal-limit round reads the remaining record and completes.

/// Rename the unchanged stream while keeping wire.jsonl and changing its agent directory.

/// file_identity matches the stream, preventing duplicate events.

// Change the agent directory; the unchanged stream remains discoverable.