Miscellaneous fixes across CLI and core modules:
- Timeline: widen TAG_WIDTH from 10 to 11 to accommodate longer event
type labels without truncation
- render.rs: save and restore LORE_ICONS env var in glyph_mode test to
prevent interference from the test environment leaking into or from
other tests that set LORE_ICONS
- logging.rs: adjust verbose=1 to info level (was debug), verbose=2 to
debug — this reduces noise at -v while keeping -vv as the full debug
experience
- issues.rs, merge_requests.rs: use infodebug! macro consistently for
ingestion summary logging
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three quality-of-life improvements to reduce log noise and improve readability:
1. logging.rs: Add CompactHumanFormat for stderr tracing output. Replaces the
default format with a minimal 'HH:MM:SS LEVEL message key=value' layout —
no span context, no full timestamps, no target module. The JSON file log
layer is unaffected. This makes watching 'lore sync' output much cleaner.
2. lock.rs: Downgrade AppLock acquire/release messages from info! to debug!.
Lock lifecycle events (acquired new, acquired existing, released) are
operational bookkeeping that clutters normal output. They remain visible
at -vv verbosity for troubleshooting.
3. ingestion/mod.rs: Add nonzero_summary() utility that formats named counters
as a compact middle-dot-separated string, suppressing zero values. Produces
output like '42 fetched · 3 labels · 12 notes' instead of verbose key=value
structured fields. Returns 'nothing to update' when all values are zero.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix MENTIONED_RE/CLOSED_BY_RE to match real GitLab format
('mentioned in issue #N' / 'mentioned in merge request !N')
- Add GITLAB_URL_RE + parse_url_refs() for full URL extraction
- Add extract_refs_from_descriptions() -> source_method='description_parse'
- Add extract_refs_from_user_notes() -> source_method='note_parse'
- Wire both into orchestrator after system note extraction
- 36 tests: regex fix, URL parsing, integration, idempotency
Adds issue:N / i:N / mr:N / m:N query syntax to bypass hybrid search
and seed the timeline directly from a known entity. All discussions for
the entity are gathered without needing Ollama.
- parse_timeline_query() detects entity-direct patterns
- resolve_entity_by_iid() resolves IID to EntityRef with ambiguity handling
- seed_timeline_direct() gathers all discussions for the entity
- 20 new tests (5 resolve, 6 direct seed, 9 parse)
- Updated CLI help text and robot-docs manifest
Moves the conn.prepare() call for fetching discussion notes outside the
per-discussion loop in collect_discussion_threads(). The SQL is identical
for every iteration, so preparing it once and rebinding parameters avoids
redundant statement compilation on each matched discussion.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements the downstream consumption of matched discussions from the seed
phase, completing the discussion thread feature across collect, CLI, and
integration tests.
Collect phase (timeline_collect.rs):
- New collect_discussion_threads() function assembles full threads by
querying notes for each matched discussion_id, filtering out system notes
(is_system = 0), ordering chronologically, and capping at THREAD_MAX_NOTES
with a synthetic "[N more notes not shown]" summary note
- build_entity_lookup() creates a (type, id) -> (iid, path) map from seed
and expanded entities to provide display metadata for thread events
- Thread timestamp is set to the first note's created_at for correct
chronological interleaving with other timeline events
- collect_events() gains a matched_discussions parameter; threads are
collected after entity events and before evidence note merging
CLI rendering (cli/commands/timeline.rs):
- Human mode: threads render with box-drawing borders, bold @author tags,
date-stamped notes, and word-wrapped bodies (60 char width)
- Robot mode: DiscussionThread serializes as discussion_thread kind with
note_count, full notes array (note_id, author, body, ISO created_at)
- THREAD tag in yellow for human event tag styling
- TimelineMeta gains discussion_threads_included count
Tests:
- 8 new collect tests: basic thread assembly, system note filtering, empty
thread skipping, body truncation to THREAD_NOTE_MAX_CHARS, note cap with
synthetic summary, timestamp from first note, chronological sort position,
and deduplication of duplicate discussion_ids
- Integration tests updated for new collect_events signature
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduces the foundation for full discussion thread support in the
timeline pipeline. Adds three new domain types to timeline.rs:
- ThreadNote: individual note within a thread (id, author, body, timestamp)
- MatchedDiscussion: tracks discussions matched during seeding with their
parent entity (issue or MR) for downstream collection
- DiscussionThread variant on TimelineEventType: carries a full thread of
notes, sorted between NoteEvidence and CrossReferenced
Moves truncate_to_chars() from timeline_seed.rs to timeline.rs as pub(crate)
for reuse by the collect phase. Adds THREAD_NOTE_MAX_CHARS (2000) and
THREAD_MAX_NOTES (50) constants.
Upgrades the seed SQL in resolve_documents_to_entities() to resolve note
documents to their parent discussion via an additional LEFT JOIN chain
(notes -> discussions), using COALESCE to unify the entity resolution path
for both discussion and note source types. SeedResult gains a
matched_discussions field that captures deduplicated discussion matches.
Tests cover: discussion matching from discussion docs, note-to-parent
resolution, deduplication of same discussion across multiple docs, and
correct parent entity type (issue vs MR).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace FTS-only seed entity discovery with hybrid search (FTS + vector
via RRF), using the same search_hybrid infrastructure as the search
command. Falls back gracefully to FTS-only when Ollama is unavailable.
Changes:
- seed_timeline() now accepts OllamaClient, delegates to search_hybrid
- New resolve_documents_to_entities() replaces find_seed_entities()
- SeedResult gains search_mode field tracking actual mode used
- TimelineResult carries search_mode through to JSON renderer
- run_timeline wires up OllamaClient from config
- handle_timeline made async for the hybrid search await
- Tests updated for new function signatures
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract shared path resolution logic from who.rs into a new
core::path_resolver module for cross-module reuse. Functions moved:
escape_like, normalize_repo_path, PathQuery, SuffixResult,
build_path_query, suffix_probe. Duplicate escape_like copies removed
from list.rs, project.rs, and filters.rs — all now import from
path_resolver.
Additionally fixes two bugs in query_expert_details() and
query_overlap() where only position_new_path was checked (missing
old_path matches for renamed files) and state filter excluded 'closed'
MRs despite the main scoring query including them with a decay
multiplier.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New module: core::file_history with resolve_rename_chain() that traces
a file path through its rename history in mr_file_changes using
bidirectional BFS (forward: old_path->new_path, backward: new_path->old_path).
Key design decisions:
- Depth-bounded BFS: each queue entry carries its distance from the
origin, so max_hops correctly limits by graph distance (not by total
nodes discovered). This matters for branching rename graphs where a
file was renamed differently in parallel MRs.
- Cycle-safe: visited set prevents infinite loops from circular renames.
- Project-scoped: queries are always scoped to a single project_id.
- Deterministic: output is sorted for stable results.
Tests cover: linear chains (forward/backward), cycles, max_hops=0,
depth-bounded linear chains, branching renames, diamond patterns,
and cross-project isolation (9 tests total).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Factor out parse_since_from(input, reference_ms) so callers can compute
relative durations against a fixed reference timestamp instead of always
using now(). The existing parse_since() now delegates to it with now_ms().
Enables testable and reproducible time-relative queries for features like
timeline --as-of and who --as-of.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move inline #[cfg(test)] mod tests { ... } blocks from 22 source files
into dedicated _tests.rs companion files, wired via:
#[cfg(test)]
#[path = "module_tests.rs"]
mod tests;
This keeps implementation-focused source files leaner and more scannable
while preserving full access to private items through `use super::*;`.
Modules extracted:
core: db, note_parser, payloads, project, references, sync_run,
timeline_collect, timeline_expand, timeline_seed
cli: list (55 tests), who (75 tests)
documents: extractor (43 tests), regenerator
embedding: change_detector, chunking
gitlab: graphql (wiremock async tests), transformers/issue
ingestion: dirty_tracker, discussions, issues, mr_diffs
Also adds conflicts_with("explain_score") to the --detail flag in the
who command to prevent mutually exclusive flags from being combined.
All 629 unit tests pass. No behavior changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add references_full, user_notes_count, merge_requests_count computed
fields to show issue. Add closed_at and confidential columns via
migration 023.
Closes: bd-2g50
User-supplied project names containing `%` or `_` were passed directly
into LIKE patterns, causing unintended wildcard matching. For example,
`my_project` would match `my-project` because `_` is a single-char
wildcard in SQL LIKE.
Added escape_like() helper that escapes `\`, `%`, and `_` with
backslash, and added ESCAPE '\' clauses to both the suffix-match and
substring-match queries in resolve_project().
Includes two regression tests:
- test_underscore_not_wildcard: `_` in input must not match `-`
- test_percent_not_wildcard: `%` in input must not match arbitrary strings
Introduces a new optional `defaultProject` field on Config (and
MinimalConfig for init output) that acts as a fallback when the
`-p`/`--project` CLI flag is omitted.
Domain-layer changes:
- Config.default_project: Option<String> with camelCase serde rename
- Config::load validates that defaultProject matches a configured
project path (exact or case-insensitive suffix match), returning
ConfigInvalid on mismatch
- Config::effective_project(cli_flag) -> Option<&str>: cascading
resolver that prefers the CLI flag, then the config default, then None
- MinimalConfig.default_project with skip_serializing_if for clean
JSON output when unset
Tests added:
- effective_project: CLI overrides default, falls back to default,
returns None when both absent
- Config::load: accepts valid defaultProject, rejects nonexistent,
accepts suffix match
- MinimalConfig: omits null defaultProject, includes when set
- Helper write_config_with_default_project for parameterized tests
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The migration runner now inserts (OR REPLACE) the schema_version row
after each successful migration batch, regardless of whether the
migration SQL itself contains a self-registering INSERT. This prevents
version tracking gaps when a .sql migration omits the bookkeeping
statement, which would leave the schema at an unrecorded version and
cause re-execution attempts on next startup.
Legacy migrations that already self-register are unaffected thanks to
the OR REPLACE conflict resolution.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a "Phase 1.5" status enrichment step to the issue ingestion pipeline
that fetches work item statuses via the GitLab GraphQL API after the
standard REST API ingestion completes.
Schema changes (migration 021):
- Add status_name, status_category, status_color, status_icon_name, and
status_synced_at columns to the issues table (all nullable)
Ingestion pipeline changes:
- New `enrich_issue_statuses_txn()` function that applies fetched
statuses in a single transaction with two phases: clear stale statuses
for issues that no longer have a status widget, then apply new/updated
statuses from the GraphQL response
- ProgressEvent variants for status enrichment (complete/skipped)
- IngestProjectResult tracks enrichment metrics (seen, enriched, cleared,
without_widget, partial_error_count, enrichment_mode, errors)
- Robot mode JSON output includes per-project status enrichment details
Configuration:
- New `sync.fetchWorkItemStatus` config option (defaults true) to disable
GraphQL status enrichment on instances without Premium/Ultimate
- `LoreError::GitLabAuthFailed` now treated as permanent API error so
status enrichment auth failures don't trigger retries
Also removes the unnecessary nested SAVEPOINT in store_closes_issues_refs
(already runs within the orchestrator's transaction context).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace .unwrap_or(), .ok(), and .filter_map(|r| r.ok()) patterns with
proper error propagation using ? and rusqlite::OptionalExtension where
the query may legitimately return no rows.
Affected areas:
- events_db::count_events: three count queries now propagate errors
instead of defaulting to (0, 0) on failure
- note_parser::extract_refs_from_system_notes: row iteration errors
are now propagated instead of silently dropped via filter_map
- note_parser::noteable_type_to_entity_type: unknown types now log a
debug warning before defaulting to "issue"
- payloads::store_payload/read_payload: use .optional()? instead of
.ok() to distinguish "no row" from "query failed"
- backoff::compute_next_attempt_at: use .clamp(0, 30) to guard against
negative attempt_count, not just .min(30)
- search::vector::max_chunks_per_document: returns Result<i64> with
proper error propagation through .optional()?.flatten()
- embedding::chunk_ids::decode_rowid: promote debug_assert to assert
since negative rowids indicate data corruption worth failing fast on
- ingestion::dirty_tracker::record_dirty_error: use .optional()? to
handle missing dirty_sources row gracefully instead of hard error
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Expert mode now surfaces the specific MR references (project/path!iid) that
contributed to each expert's score, capped at 50 per user. A new --detail flag
adds per-MR breakdowns showing role (Author/Reviewer/both), note count, and
last activity timestamp.
Scoring weights (author_weight, reviewer_weight, note_bonus) are now
configurable via the config file's `scoring` section with validation that
rejects negative values. Defaults shift to author_weight=25, reviewer_weight=10,
note_bonus=1 — better reflecting that code authorship is a stronger expertise
signal than review assignment alone.
Path resolution gains suffix matching: typing "login.rs" auto-resolves to
"src/auth/login.rs" when unambiguous, with clear disambiguation errors when
multiple paths match. Project-scoping (-p) narrows the candidate set.
The MAX_MR_REFS_PER_USER constant is promoted to module scope for reuse
across expert and overlap modes. Human output shows MR refs inline and detail
sub-rows when requested. Robot JSON includes mr_refs, mr_refs_total,
mr_refs_truncated, and optional details array.
Includes comprehensive tests for suffix resolution, scoring weight
configurability, MR ref aggregation across projects, and detail mode.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Chain: bd-jec (config flag) -> bd-2yo (fetch MR diffs) -> bd-3qn6 (rewrite who queries)
- Add fetch_mr_file_changes config option and --no-file-changes CLI flag
- Add GitLab MR diffs API fetch pipeline with watermark-based sync
- Create migration 020 for diffs_synced_for_updated_at watermark column
- Rewrite query_expert() and query_overlap() to use 4-signal UNION ALL:
DiffNote reviewers, DiffNote MR authors, file-change authors, file-change reviewers
- Deduplicate across signal types via COUNT(DISTINCT CASE WHEN ... THEN mr_id END)
- Add insert_file_change test helper, 8 new who tests, all 397 tests pass
- Also includes: list performance migration 019, autocorrect module, README updates
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three micro-optimizations with zero behavioral change:
1. timeline_collect.rs: Reorder format!() before enum construction so
the owned String moves into the variant directly, eliminating
.clone() on state, label, and milestone strings in StateChanged,
LabelAdded/Removed, and MilestoneSet/Removed event paths.
2. pipeline.rs: Use Arc<str> for doc_hash shared across a document's
chunks instead of cloning the full String per chunk. Also remove
redundant embed_buf.reserve() since extend_from_slice already
handles growth and the buffer is reused across iterations.
3. rrf.rs: Pre-allocate HashMap with combined vector+fts result count
via with_capacity() to avoid rehashing during RRF score accumulation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Comprehensive peer code review identified and fixed the following:
1. who.rs: @-prefixed path routing used `target` (with @) instead of
`clean` (stripped) when checking for '/' and passing to Expert mode,
causing `lore who @src/auth/` to silently return zero results because
the SQL LIKE matched against `@src/auth/%` which never exists.
2. db.rs: After ROLLBACK TO savepoint on migration failure, the savepoint
was never RELEASEd, leaving it active on the connection. Fixed in both
run_migrations() and run_migrations_from_dir().
3. lock.rs: Multiple acquire() calls (e.g. re-acquiring a stale lock)
replaced the heartbeat_handle without stopping the old thread, causing
two concurrent heartbeat writers competing on the same lock row. Now
signals the old thread to stop and joins it before spawning a new one.
4. chunk_ids.rs: encode_rowid() had no guard for chunk_index >= 1000
(CHUNK_ROWID_MULTIPLIER), which would cause rowid collisions between
adjacent documents. Added range assertion [0, 1000).
5. main.rs: Fallback JSON error formatting in handle_auth_test
interpolated LoreError Display output without escaping quotes or
backslashes, potentially producing malformed JSON for robot-mode
consumers. Now escapes both characters before interpolation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three defensive improvements found via peer code review:
1. lock.rs: Lock release errors were silently discarded with `let _ =`.
If the DELETE failed (disk full, corruption), the lock stayed in the
database with no diagnostic. Next sync would require --force with no
clue why. Now logs with error!() including the underlying error message.
2. filters.rs: Dynamic SQL label filter construction had no upper bound
on bind parameters. With many combined filters, param_idx + labels.len()
could exceed SQLite's 999-parameter limit, producing an opaque error.
Added a guard that caps labels at 900 - param_idx.
3. vector.rs: max_chunks_per_document returned i64 which was cast to
usize. A negative value from a corrupt database would wrap to a huge
number, causing overflow in the multiplier calculation. Now clamped
to .max(1) and cast via unsigned_abs().
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three correctness bugs found via peer code review:
1. TimelineEvent PartialEq/Ord omitted entity_type — issue #42 and MR #42
with the same timestamp and event_type were treated as equal. In a
BTreeSet or dedup, one would silently be dropped. Added entity_type to
both PartialEq and Ord comparisons.
2. discussions.rs: store_payload() was called outside the transaction
(on bare conn) while upsert_discussion/notes were inside. A crash
between them left orphaned payload rows. Moved store_payload inside
the unchecked_transaction block, matching mr_discussions.rs pattern.
3. Migration 017 created idx_issue_assignees_username(username, issue_id)
but migration 005 already created the same index name with just
(username). SQLite's IF NOT EXISTS silently skipped the composite
version on every existing database. New migration 018 drops and
recreates the index with correct composite columns.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add `lore who` command with 5 query modes answering collaboration questions
using existing DB data (280K notes, 210K discussions, 33K DiffNotes):
- Expert: who knows about a file/directory (DiffNote path analysis + MR breadth scoring)
- Workload: what is a person working on (assigned issues, authored/reviewing MRs, discussions)
- Active: what discussions need attention (unresolved resolvable, global/project-scoped)
- Overlap: who else is touching these files (dual author+reviewer role tracking)
- Reviews: what review patterns does a person have (prefix-based category extraction)
Includes migration 017 (5 composite indexes), CLI skeleton with clap conflicts_with
validation, robot JSON output with input+resolved_input reproducibility, human terminal
output, and 20 unit tests. All quality gates pass.
Closes: bd-1q8z, bd-34rr, bd-2rk9, bd-2ldg, bd-zqpf, bd-s3rc, bd-m7k1, bd-b51e,
bd-2711, bd-1rdi, bd-3mj2, bd-tfh3, bd-zibc, bd-g0d5
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Robot mode consistency improvements across all command output:
Timing:
- Every robot JSON response now includes meta.elapsed_ms measuring
wall-clock time from command start to serialization. Agents can use
this to detect slow queries and tune --limit or --project filters.
Field selection (--fields):
- print_list_issues_json and print_list_mrs_json accept an optional
fields slice that prunes each item in the response array to only
the requested keys. A "minimal" preset expands to
[iid, title, state, updated_at_iso] for token-efficient agent scans.
- filter_fields and expand_fields_preset live in the new
src/cli/robot.rs module alongside RobotMeta.
Actionable error recovery:
- LoreError gains an actions() method returning concrete shell commands
an agent can execute to recover (e.g. "ollama serve" for
OllamaUnavailable, "lore init" for ConfigNotFound).
- RobotError now serializes an "actions" array (empty array omitted)
so agents can parse and offer one-click fixes.
Envelope consistency:
- show issue/MR JSON responses now use the standard
{"ok":true,"data":...,"meta":...} envelope instead of bare data,
matching all other commands.
Files: src/cli/robot.rs (new), src/core/error.rs,
src/cli/commands/{count,embed,generate_docs,ingest,list,show,stats,sync_status}.rs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Shutdown signal improvements:
- Upgrade ShutdownSignal from Relaxed to Release/Acquire ordering.
Relaxed was technically sufficient for a single flag but
Release/Acquire is the textbook correct pattern and ensures
visibility guarantees across threads without relying on x86 TSO.
- Add double Ctrl+C support to all three signal handlers (ingest,
embed, sync). First Ctrl+C sets cooperative flag with user message;
second Ctrl+C force-exits with code 130 (standard SIGINT convention).
CLI hardening:
- LORE_ROBOT env var now checks for truthy values (!empty, !="0",
!="false") instead of mere existence. Setting LORE_ROBOT=0 or
LORE_ROBOT=false no longer activates robot mode.
- Replace unreachable!() in color mode match with defensive warning
and fallback to auto. Clap validates the values but defense in depth
prevents panics if the value_parser is ever changed.
- Replace unreachable!() in completions shell match with proper error
return for unsupported shells.
Exit code collision fix:
- ConfigNotFound was mapped to exit code 2 (error.rs:56) which
collided with handle_clap_error() also using exit code 2 for parse
errors. Agents calling lore --robot could not distinguish "bad
arguments" from "missing config file."
- Restore ConfigNotFound to exit code 20 (its original dedicated code).
- Update robot-docs exit code table: code 2 = "Usage error", code 20 =
"Config not found".
Build script:
- Track .git/refs/heads directory for Cargo rebuild triggers. Ensures
GIT_HASH env var updates when branch refs change, not just HEAD.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three fixes to the sync pipeline:
1. Atomic watermarks: wrap complete_job + update_watermark in a single
SQLite transaction so crash between them can't leave partial state.
2. Concurrent drain loops: prefetch HTTP requests via join_all (batch
size = dependent_concurrency), then write serially to DB. Reduces
~9K sequential requests from ~19 min to ~2.4 min.
3. Graceful shutdown: install Ctrl+C handler via ShutdownSignal
(Arc<AtomicBool>), thread through orchestrator/CLI, release locked
jobs on interrupt, record sync_run as "failed".
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The robot JSON envelope's meta.total_events field was incorrectly
reporting events.len() (the post-limit count), making it identical
to meta.showing. This defeated the purpose of having both fields.
Changes across the pipeline to fix this:
- collect_events now returns (Vec<TimelineEvent>, usize) where the
second element is the total event count before truncation
- TimelineResult gains a total_events_before_limit field (serde-skipped)
so the value flows cleanly from collect through to the renderer
- main.rs passes the real total instead of the events.len() workaround
Additional cleanup in this pass:
- Derive PartialEq/Eq/PartialOrd/Ord on TimelineEventType, replacing
the hand-rolled event_type_discriminant() function. Variant declaration
order now defines sort tiebreak, documented in a doc comment.
- Validate --since input with a proper LoreError::Other instead of
silently treating invalid values as None
- Fix ANSI-aware tag column padding with console::pad_str (colored tags
like "[merged]" were misaligned because ANSI escapes consumed width)
- Remove dead print_timeline_json and infer_max_depth functions that
were superseded by print_timeline_json_with_meta
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds a SAFETY comment explaining why the transmute of sqlite3_vec_init
to the sqlite3_auto_extension callback type is sound. The three
invariants (stable C-ABI signature, single-call-per-connection contract,
idempotency) were previously undocumented, which left the lone unsafe
block without justification for future readers.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Follows up on the resolve_entity_ref extraction by updating all three
pipeline stages to consume the shared helper and removing their local
duplicates (~75 lines of dead code eliminated).
timeline_seed.rs:
- Switch from local resolve_entity to shared resolve_entity_ref with
explicit Some(proj_id) scoping
- Add tracing::debug for orphaned discussion parents instead of silently
skipping them, aiding debugging when evidence notes go missing
- Use saturating_mul for the over-fetch multiplier to prevent overflow on
pathological max_seeds values
timeline_expand.rs:
- Switch from local resolve_entity_ref to shared version with None
project scoping (cross-project traversal)
- Pass Option<i64> for target_iid in UnresolvedRef construction instead
of unwrap_or(0) sentinel
- Update test assertion to compare against Some(42)
timeline_collect.rs:
- Make entity_id_column return Result instead of silently defaulting to
issue_id for unknown entity types. The previous fallback could produce
incorrect SQL queries that return wrong results rather than failing
- Replace if-let chains in collect_merged_event with exhaustive match
blocks that propagate real DB errors while gracefully handling expected
missing-data cases (QueryReturnedNoRows, NULL merged_at)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The seed, expand, and collect stages each had their own near-identical
resolve_entity_ref helper that converted internal DB IDs to full EntityRef
structs. This duplication made it easy for bug fixes to land in one copy
but not the others.
Extract a single public resolve_entity_ref into timeline.rs with an
optional project_id parameter:
- Some(project_id): scopes the lookup (used by seed, which knows the
project from the FTS result)
- None: unscoped lookup (used by expand, which traverses cross-project
references)
Also changes UnresolvedRef.target_iid from i64 to Option<i64>. Cross-
project references parsed from descriptions may not always carry an IID
(e.g. when the reference is malformed or the target was deleted). The
previous sentinel value of 0 was semantically incorrect since GitLab IIDs
start at 1.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Change detection queries (embedding/change_detector.rs):
- Replace triple-EXISTS subquery pattern with LEFT JOIN + NULL check
- SQLite now scans embedding_metadata once instead of three times
- Semantically identical: returns docs needing embedding when no
embedding exists, hash changed, or config mismatch
Count queries (cli/commands/count.rs):
- Consolidate 3 separate COUNT queries for issues into single query
using conditional aggregation (CASE WHEN state = 'x' THEN 1)
- Same optimization for MRs: 5 queries reduced to 1
Search filter queries (search/filters.rs):
- Replace N separate EXISTS clauses for label filtering with single
IN() clause with COUNT/GROUP BY HAVING pattern
- For multi-label AND queries, this reduces N subqueries to 1
FTS tokenization (search/fts.rs):
- Replace collect-into-Vec-then-join pattern with direct String building
- Pre-allocate capacity hint for result string
Discussion truncation (documents/truncation.rs):
- Calculate total length without allocating concatenated string first
- Only allocate full string when we know it fits within limit
Embedding pipeline (embedding/pipeline.rs):
- Add Vec::with_capacity hints for chunk work and cleared_docs hashset
- Reduces reallocations during embedding batch processing
Backoff calculation (core/backoff.rs):
- Replace unchecked addition with saturating_add to prevent overflow
- Add test case verifying overflow protection
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Removes module-level doc comments (//! lines) and excessive inline doc
comments that were duplicating information already evident from:
- Function/struct names (self-documenting code)
- Type signatures (the what is clear from types)
- Implementation context (the how is clear from code)
Affected modules:
- cli/* - Removed command descriptions duplicating clap help text
- core/* - Removed module headers and obvious function docs
- documents/* - Removed extractor/regenerator/truncation docs
- embedding/* - Removed pipeline and chunking docs
- gitlab/* - Removed client and transformer docs (kept type definitions)
- ingestion/* - Removed orchestrator and ingestion docs
- search/* - Removed FTS and vector search docs
Philosophy: Code should be self-documenting. Comments should explain
"why" (business decisions, non-obvious constraints) not "what" (which
the code itself shows). This change reduces noise and maintenance burden
while keeping the codebase just as understandable.
Retains comments for:
- Non-obvious business logic
- Important safety invariants
- Complex algorithm explanations
- Public API boundaries where generated docs matter
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Introduces two new modules for extracting and storing entity cross-references
from GitLab data:
note_parser.rs:
- Parses system notes for "mentioned in" and "closed by" patterns
- Extracts cross-project references (group/project#42, group/project!123)
- Uses lazy-compiled regexes for performance
- Handles both issue (#) and MR (!) sigils
- Provides extract_refs_from_system_notes() for batch processing
references.rs:
- Extracts refs from resource_state_events table (API-sourced closes links)
- Provides insert_entity_reference() for storing discovered references
- Includes resolution helpers: resolve_issue_local_id, resolve_mr_local_id,
resolve_project_path for converting iids to internal IDs
- Enables cross-project reference resolution
These modules power the entity_references table, enabling features like
"find all MRs that close this issue" and "find all issues mentioned in this MR".
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Three correctness fixes found during peer code review:
Embedding pipeline savepoint leak (HIGH severity):
The SAVEPOINT embed_page / RELEASE embed_page pattern had ~10 `?`
propagation points between them. Any error from record_embedding_error,
clear_document_embeddings, or store_embedding would exit the function
without rolling back, leaving the SQLite connection in a broken
transactional state and causing cascading failures for the rest of the
session. Fixed by extracting page processing into `embed_page()` and
wrapping with explicit rollback-on-error handling.
Dependent queue fail_job race (MEDIUM severity):
fail_job performed a SELECT followed by a separate UPDATE on the
attempts counter without a transaction. Under concurrent lock
reclamation, the attempts value could be read stale. Replaced with a
single atomic UPDATE that increments attempts and computes exponential
backoff entirely in SQL, also halving DB round-trips. Added explicit
error when the job no longer exists.
RRF duplicate document score inflation (MEDIUM severity):
If a retriever returned the same document_id multiple times, the RRF
score accumulated multiple rank contributions while the rank only
recorded the first occurrence. Moved the score accumulation inside the
`if is_none` guard so only the first occurrence per list contributes.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Targeted fixes across multiple subsystems:
dependent_queue:
- Add project_id parameter to claim_jobs() for project-scoped job claiming,
preventing cross-project job theft during concurrent multi-project ingestion
- Add project_id parameter to count_pending_jobs() with optional scoping
(None returns global counts, Some(pid) returns per-project counts)
gitlab/client:
- Downgrade rate-limit log from warn to info (429s are expected operational
behavior, not warnings) and add structured fields (path, status_code)
for better log filtering and aggregation
gitlab/transformers/discussion:
- Add tracing::warn on invalid timestamp parse instead of silent fallback
to epoch 0, making data quality issues visible in logs
ingestion/merge_requests:
- Remove duplicate doc comment on upsert_label_tx
search/rrf:
- Replace partial_cmp().unwrap_or() with total_cmp() for f64 sorting,
eliminating the NaN edge case entirely (total_cmp treats NaN consistently)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Introduce the foundational observability layer for the sync pipeline:
- MetricsLayer: Custom tracing subscriber layer that captures span timing
and structured fields, materializing them into a hierarchical
Vec<StageTiming> tree for robot-mode performance data output
- logging: Dual-layer subscriber infrastructure with configurable stderr
verbosity (-v/-vv/-vvv) and always-on JSON file logging with daily
rotation and configurable retention (default 30 days)
- SyncRunRecorder: Compile-time enforced lifecycle recorder for sync_runs
table (start -> succeed|fail), with correlation IDs and aggregate counts
- LoggingConfig: New config section for log_dir, retention_days, and
file_logging toggle
- get_log_dir(): Path helper for log directory resolution
- is_permanent_api_error(): Distinguish retryable vs permanent API failures
(only 404 is truly permanent; 403/auth errors may be environmental)
Database changes:
- Migration 013: Add resource_events_synced_for_updated_at watermark columns
to issues and merge_requests tables for incremental resource event sync
- Migration 014: Enrich sync_runs with run_id correlation ID, aggregate
counts (total_items_processed, total_errors), and run_id index
- Wrap file-based migrations in savepoints for rollback safety
Dependencies: Add uuid (run_id generation), tracing-appender (file logging)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
11 isomorphic performance fixes from deep audit (no behavior changes):
- Eliminate double serialization: store_payload now accepts pre-serialized
bytes (&[u8]) instead of re-serializing from serde_json::Value. Uses
Cow<[u8]> for zero-copy when compression is disabled.
- Add SQLite cache_size (64MB) and mmap_size (256MB) pragmas
- Replace SELECT-then-INSERT label upserts with INSERT...ON CONFLICT
RETURNING in both issues.rs and merge_requests.rs
- Replace INSERT + SELECT milestone upsert with RETURNING
- Use prepare_cached for 5 hot-path queries in extractor.rs
- Optimize compute_list_hash: index-sort + incremental SHA-256 instead
of clone+sort+join+hash
- Pre-allocate embedding float-to-bytes buffer with Vec::with_capacity
- Replace RandomState::new() in rand_jitter with atomic counter XOR nanos
- Remove redundant per-note payload storage (discussion payload contains
all notes already)
- Change transform_issue to accept &GitLabIssue (avoids full struct clone)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The sync pipeline was bottlenecked at 10 req/s (hardcoded) with
sequential project processing and no retry on rate limiting. These
changes target 3-5x throughput improvement.
Rate limit configuration:
- Add requestsPerSecond to SyncConfig (default 30.0, was hardcoded 10)
- Pass configured rate through to GitLabClient::new from ingest
- Floor rate at 0.1 rps in RateLimiter::new to prevent panic on
Duration::from_secs_f64(1.0 / 0.0) — now reachable via user config
429 auto-retry:
- Both request() and request_with_headers() retry up to 3 times on
HTTP 429, respecting the retry-after header (default 60s)
- Extract parse_retry_after helper, reused by handle_response fallback
- After exhausting retries, the 429 error propagates as before
- Improved JSON decode errors now include a response body preview
Concurrent project ingestion:
- Derive Clone on GitLabClient (cheap: shares Arc<Mutex<RateLimiter>>
and reqwest::Client which is already Arc-backed)
- Restructure project loop to use futures::stream::buffer_unordered
with primary_concurrency (default 4) as the parallelism bound
- Each project gets its own SQLite connection (WAL mode + busy_timeout
handles concurrent writes)
- Add show_spinner field to IngestDisplay to separate the per-project
spinner from the sync-level stage spinner
- Error aggregation defers failures: all successful projects get their
summaries printed and results counted before returning the first error
- Bump dependentConcurrency default from 2 to 8 for discussion prefetch
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two hardening changes to the dependent queue and orchestrator:
- dependent_queue::fail_job now propagates the rusqlite error via ?
instead of silently falling back to 0 attempts when the job row is
missing. A missing job is a real bug that should surface, not be
masked by unwrap_or(0) which would cause infinite retries at the
base backoff interval.
- orchestrator::enqueue_resource_events_for_entity_type replaces
format!-based SQL ("SELECT {id_col} FROM {table}") with separate
hardcoded queries per entity type. While the original values were
not user-controlled, hardcoded SQL is clearer about intent and
eliminates a class of injection risk entirely.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
GitLab returns null for the label/milestone fields on resource_label_events
and resource_milestone_events when the referenced label or milestone has
been deleted. This caused deserialization failures during sync.
- Add migration 012 to recreate both event tables with nullable
label_name, milestone_title, and milestone_id columns (SQLite
requires table recreation to alter NOT NULL constraints)
- Change GitLabLabelEvent.label and GitLabMilestoneEvent.milestone
to Option<> in the Rust types
- Update upsert functions to pass through None values correctly
- Add tests for null label and null milestone deserialization
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
events_db.rs:
- Removed internal savepoints from upsert_state_events,
upsert_label_events, and upsert_milestone_events. Each function
previously created its own savepoint, making it impossible for
callers to wrap all three in a single atomic transaction.
- Changed signatures from &mut Connection to &Connection, since
savepoints are no longer created internally. This makes the
functions compatible with rusqlite::Transaction (which derefs to
Connection), allowing callers to pass a transaction directly.
orchestrator.rs:
- Deleted the three store_*_events_tx() functions (store_state_events_tx,
store_label_events_tx, store_milestone_events_tx) which were
hand-duplicated copies of the events_db upsert functions, created as
a workaround for the &mut Connection requirement. Now that events_db
accepts &Connection, store_resource_events() calls the canonical
upsert functions directly through the unchecked_transaction.
- Replaced the max-iterations guard in drain_resource_events() with a
HashSet-based deduplication of job IDs. The old guard used an
arbitrary 2x multiplier on total_pending which could either terminate
too early (if many retries were legitimate) or too late. The new
approach precisely prevents reprocessing the same job within a single
drain run, which is the actual invariant we need.
Net effect: ~133 lines of duplicated SQL removed, single source of
truth for event upsert logic, and callers control transaction scope.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two bug fixes:
1. extractor.rs: The content hash was computed on the pre-truncation
content, meaning the hash stored in the document didn't correspond
to the actual stored (truncated) content. This would cause change
detection to miss updates when content changed only within the
truncated portion. Hash is now computed after truncate_hard_cap()
so it always matches the persisted content.
2. dependent_queue.rs: claim_jobs() had a TOCTOU race between the
SELECT that found available jobs and the UPDATE that locked them.
Under concurrent callers, two drain runs could claim the same job.
Replaced with a single UPDATE ... RETURNING statement that
atomically selects and locks jobs in one operation.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Automated formatting and lint corrections from parallel agent work:
- cargo fmt: import reordering (alphabetical), line wrapping to respect
max width, trailing comma normalization, destructuring alignment,
function signature reformatting, match arm formatting
- clippy (pedantic): Range::contains() instead of manual comparisons,
i64::from() instead of `as i64` casts, .clamp() instead of
.max().min() chains, let-chain refactors (if-let with &&),
#[allow(clippy::too_many_arguments)] and
#[allow(clippy::field_reassign_with_default)] where warranted
- Removed trailing blank lines and extra whitespace
No behavioral changes. All existing tests pass unmodified.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
New module src/core/dependent_queue.rs provides job queue operations
against the pending_dependent_fetches table. Designed for second-pass
fetches that depend on primary entity ingestion (resource events,
MR close references, MR file diffs).
Queue operations:
- enqueue_job: Idempotent INSERT OR IGNORE keyed on the UNIQUE
(project_id, entity_type, entity_iid, job_type) constraint.
Returns bool indicating whether the row was actually inserted.
- claim_jobs: Two-phase claim — SELECT available jobs (unlocked,
past retry window) then UPDATE locked_at in batch. Orders by
enqueued_at ASC for FIFO processing within a job type.
- complete_job: DELETE the row on successful processing.
- fail_job: Increments attempts, calculates exponential backoff
(30s * 2^(attempts-1), capped at 480s), sets next_retry_at,
clears locked_at, and records the error message. Reads current
attempts via query with unwrap_or(0) fallback for robustness.
- reclaim_stale_locks: Clears locked_at on jobs locked longer than
a configurable threshold, recovering from worker crashes.
- count_pending_jobs: GROUP BY job_type aggregation for progress
reporting and stats display.
Registers both events_db and dependent_queue in src/core/mod.rs.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
New module src/core/events_db.rs provides database operations for
resource events:
- upsert_state_events: Batch INSERT OR REPLACE for state change events,
keyed on UNIQUE(gitlab_id, project_id). Wraps in a savepoint for
atomicity per entity batch. Maps GitLabStateEvent fields including
optional user, source_commit, and source_merge_request_iid.
- upsert_label_events: Same pattern for label add/remove events,
extracting label.name for denormalized storage.
- upsert_milestone_events: Same pattern for milestone assignment events,
storing both milestone.title and milestone.id.
All three upsert functions:
- Take &mut Connection (required for savepoint creation)
- Use prepare_cached for statement reuse across batch iterations
- Convert ISO timestamps via iso_to_ms_strict for ms-epoch storage
- Propagate rusqlite errors via the #[from] LoreError::Database path
- Return the count of events processed
Supporting functions:
- resolve_entity_ids: Maps entity_type string to (issue_id, MR_id) pair
with exactly-one-non-NULL invariant matching the CHECK constraints
- count_events: Queries all three event tables with conditional COUNT
aggregations, returning EventCounts struct. Uses unwrap_or((0, 0))
for graceful degradation when tables don't exist (pre-migration 011).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
