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

Comment 1, line 1

//! Codex boundary tests: directories/headers marked synthetic, under V17 rules.

//! Cover gaps in M0 real samples: duplicate final records, subagents, positive cache writes,

//! tool/user messages without usage and absent response_id. Unknown versions try the latest parser

//! with V17/V30 compatibility metadata; unknown formats are rejected.

// Three scanned events include a duplicate final; ingestion deduplicates them into two calls.

// Unregistered version with unchanged shape tries the latest built-in parser, persists usage

// and compatibility metadata; V17/V30 does not reject an unlisted version alone.

// Valid compatibility data counts one call with input=10, output=5, total=15.

// Persist compatibility in event parse_basis, file active_compat and detection JSON.

// Repeated scans add no usage; compatibility metadata preserves deduplication.

// Version is absent but session_meta.id identifies the Agent's session.

// V30 selects the latest built-in parser with absent format_version.

// Additional optional envelope/payload/usage keys do not break required structure;

// valid records are persisted under the V30 optional-field scenario.

// An unknown version yielding no events plus structure diagnostics is incompatible;

// retain old results without advancing the cursor or committing events/aggregates.

// Retry next round, including after parser updates (V30).

// Empty ingestion_checkpoints permits a fresh attempt next round.

// Show both latest_fallback and structure diagnostics, distinguishing rejection from an empty success.

// Partly usable data: one valid usage record and one with an invalid required field type.

// Persist independently validated records with compatibility metadata; report the gap (V30).

// Synthetic positive cache-write data checks all six mapped fields, as in synthetic-cache-write.

// cached+write exceeding input leaves uncached unknown with a diagnostic; do not truncate values.