Thin p2p command over the memory-sync service: publish <key> <value|@file>,
read <key>, and status (merged index summary + backend identity). Binary
values are hex-encoded in JSON output; config resolves from [memory_sync]
with --backend/--path/--node-id overrides.
Typed MemorySyncDaemonConfig parsed from the [memory_sync] TOML section
(enabled, node_id, backend, path, sync_interval_ms) plus
createMemorySyncService() which resolves filesystem/s3 backends and
constructs the service. No YAMS_* env knobs.
Factor the post-setup 'what next / how to configure' guidance into pure
formatLaterCommand/formatInitSummary builders so each decision branch's
help is testable without stdout, and add a focused test asserting the
later-command and ready-summary text. Output is byte-identical.
Portability allowlist line numbers updated for the shift.
Factor the alreadyInitialized && !force_ branch out of the monolithic
execute() into a private handleAlreadyInitialized() helper. Behavior
preserved; setup decision/try nesting reduced. Portability allowlist
line numbers updated for the shift.
Daemon-facing service owns the backend + sync loop and reconciles on a
configurable interval via a stoppable worker thread. Typed
MemorySyncConfig (nodeId, syncIntervalMs); start/stop idempotent and
stop noexcept-safe. Backend injected so filesystem and S3/R2 resolution
stay with the caller.
Publish memory records as content-addressed blobs plus a
version-vector-tagged index record per write; sync lists the index,
pulls records, and LWW-merges to a deterministic winner per logical
key. Backend injected (filesystem / S3-R2 interchangeable).
Version vector (element-wise max merge, dominates/concurrent partial
order) plus hybrid timestamp and LWW resolution: causally-later writes
win, concurrent writes resolve deterministically by (ts, origin). JSON
serialization via nlohmann. Header-only, no library deps.
Stress-tagged cases ran in the default unit lane because only ~[slow]
was filtered. Tag them [slow] and add ~[slow]/[slow] splits to the five
binaries (mcp_submodule, work_coordinator, daemon_background,
plugin_infrastructure, hybrid_search) so they run in the slow suite.
Stress cases with 20 threads / 200 iterations run in the default unit
lane because only ~[slow] is filtered. Tag them [slow] so they move to
the storage_submodule_slow suite.
Catch Windows/Linux/macOS-only defect classes before push: raw
std::jthread, sqlite3_open* via bare path.c_str(), and raw getenv
reads outside the config boundary. Wired into pre-commit and the
unit/policy suite; 575 reviewed sites allowlisted.
copy_file of the migrated WAL-mode template plus -wal/-shm sidecars
corrupts WAL replay on Windows (stale -shm), losing kg_nodes. VACUUM
INTO produces a self-contained single-file copy; a probe verifies the
KG schema survived the clone.
_putenv_s removes a variable for "" and returns EINVAL for NULL, so the
CRT environment cannot represent an explicitly empty value; the boundary
collapsed unset and empty into one state. Read/write through the Win32
environment (SetEnvironmentVariableA/GetEnvironmentVariableA) instead,
keeping a best-effort _putenv_s sync so legacy std::getenv readers stay
coherent, and treat the empty-vs-unset read race fail-closed.
Rebuild macOS TSan outputs so compiler instrumentation and compiler-rt come from one Apple toolchain generation. Preserve daemon fixture dependencies until ServiceManager teardown so sanitizer lanes do not exercise a test-only use-after-free.
Reuse admitted relations and dense splitter state while preserving deterministic identities, centroids, and the 64-document safety cap.
Add corpus-shaped benchmarks so construction regressions stay measurable.
Profiling running daemon: bit-by-bit CRC32 top CPU consumer (2289
samples/15s during grep). Same algorithm/polynomial, byte-identical
output (oracle test 0..512 len), stored checksums stay valid.
Applies slice-by-8 everywhere crc ran:
- yams::common::crc32 (+ crc32Update seeded variant for mid-stream)
- compression::calculateCRC32/updateCRC32 delegate to it
- storage-local duplicate copy removed (dead scope)
- WAL entry checksum copy removed
Also restructures sizeof arithmetic in WAL size math so opengrep does
not misread it as pointer arithmetic.
Same 6-field time filter logic copy-pasted in GetCommand and ListCommand
(~72 lines each). Extract to yams::cli::applyDocTimeFilters + DocTimeFilters,
wire both commands, add unit test.
The same string helpers were copy-pasted across CLI and service code:
trim/trimCopy in 5 files, escapeRegex in 3, hasWildcard in 2, and the
regex meta-char table in 2. Move canonical versions to
yams::common::string_utils.h (trimCopy, hasWildcard, escapeRegex,
isRegexMetaChar) and wire all call sites to them; daemon_helpers'
trim_copy now delegates. Add a Catch2 test for the shared helpers.
Kept isTransientMetadataError and annotateResultRelations untouched:
they have semantically different implementations and need a separate
review, not a mechanical merge.
The add-sandbox smoke test (and CLI startup generally) can fail with
'database is locked' when a prior async batch commit holds the reference
database longer than ReferenceDB's short retry budget, which CI runners
hit consistently while local runs pass. Retry the whole init (schema +
migrations) with a bounded, growing backoff (up to ~12s total) so
transient contention no longer fails startup.
Adds a Catch2 test that holds a write lock across init and releases it
mid-retry.
Xcode 16.4's Apple libc++ lacks std::jthread; the deadline test used
it directly, breaking the macos-arm64 build. Use the repo's existing
yams::compat::jthread shim (native alias when available, std::thread
fallback otherwise).
- replace GCC/Clang-only __builtin_{add,sub,mul}_overflow calls with
yams/core/checked_arithmetic.h helpers (addOverflow/subOverflow/
mulOverflow/narrowOverflow); MSVC lacks these builtins, breaking the
Windows build introduced by #61
- port the reviewed helper + unit tests from bugfix/review-findings
- detect floating-point std::from_chars availability at compile time
(some Apple libc++ versions only provide the integral overload) and
fall back to a strict strtod parser; template indirection keeps the
missing overload from being diagnosed in the discarded branch
- add config_parse_utils and checked_arithmetic test targets
* refactor(tuning): prune dormant runtime knobs
Remove unused tuning controls and adaptive search budget state. Extract testable ONNX slot reconciliation and search telemetry hydration seams, and correct the Apple Silicon sysctl value width.
* refactor(vector): retire dormant backends
Remove unused FAISS, model-infrastructure, VectorIndexManager, and TurboQuant paths while keeping Simeon PQ and float-vector storage.
Migrate unused V2.2 quantized schemas transactionally to V2.3. Databases containing non-default quantized formats or packed-code blobs fail closed with recovery guidance and no mutation.
* refactor(plugins): remove legacy SDK tooling
Remove the Ghidra and SDK submodules, their bespoke tests, and stale scan exclusions. Retain generic PluginProcess coverage against the repository mock fixture.
* build(vector): remove stale FAISS overrides
* refactor(ipc): remove dormant FSM snapshots
* refactor(ingest): unify bounded KG enqueue
* fix(plugins): harden allocation contracts
Keep ABI outputs null or fully initialized when allocations fail, and represent empty object payloads without relying on malloc(0).
* fix(storage): define empty object reads
Return a null buffer with zero length for successful empty-object reads instead of treating malloc(0) as an allocation contract.
* refactor(storage): trim through string views
* test: prune dormant harness helpers
Remove unused readiness and fixture helpers, and keep benchmark client-count doubling in a wide integer domain.
* chore(cpp): initialize analyzer-facing locals
* build(simeon): advance analyzer cleanup
* refactor(config): drop dead daemon client keys
* refactor(ipc): inject daemon client transport
Socket and in-process execution now share DaemonClient request dispatch, so CLI fallbacks do not need alternate service paths.
* refactor(retrieval): route CLI through daemon
Search and grep now use the same daemon handlers over socket or in-process transport. Live grep remains an explicit freshness inspection.
* refactor(graph): move CLI scope into daemon
Graph CLI requests now share daemon handlers across socket and in-process transports. CWD filtering occurs before pagination so scoped totals and pages stay correct.
* build: isolate test-only runtime controls
* refactor(metadata): decouple lock telemetry
* fix(services): return on request timeout
* fix(retrieval): enforce workspace scope
* fix(graph): scope nodes before pagination
* fix(grep): drop binary result lines
* fix(daemon): bound checkpoint wait
* fix(repair): use KG health checks
* fix(storage): always verify compressed CRC
* perf(vector): skip unchanged PQ snapshots
* build(zyp): exclude bench from default build
* build: protect Meson files from formatter
* refactor: remove unreachable production APIs
* refactor(daemon): remove dormant runtime state
* fix(vector): harden daemon provider calls
Timed-out coroutines must own their client and requests. Provider discovery must also fail instead of returning synthetic embeddings in production.
* refactor(storage): remove unusable adapters
* fix(cli): unify in-process add requests
* fix(tsan): preserve async worker progress
* test(cli): isolate daemon-dependent suites
* fix(repair): expose queue backoff progress
