Repository navigation
Sashiko v.0.9.5 Release
Highlights of main changes since 0.9.4:
- O(1) Prometheus Telemetry and 4-Tab Statistics Dashboard:
- Replace full-table statistics scans with four bounded O(1) layers:
lock-free runtime counters/histograms with a zero-allocation
tracing MetricsLayer, trigger-maintained live state and
BugAccess-scoped subsystem gauges, multi-resolution SQLite rollups
with resumable background backfill (migration 024), and
pre-extracted per-review pipeline summaries - Expose GET /metrics (Prometheus text 0.0.4 with optional
server.metrics_token auth) and /api/v1/{query,query_range,
query_batch,labels,label/{name}/values,metadata} backed by a
bounded PromQL parser/evaluator, covering indexes, singleflight
caches, per-IP/global rate limiting, concurrency shedding, and
query complexity bounds - Redesign the Web UI #/stats view into four flat Sashiko-native tabs
(Project, Bugs, Health, Pipeline) featuring weekly Clean Patches %
and Issues per Patch by Revision (v1 vs v2+) with upstream release
markers, cumulative Open vs Fixed bug trends, 60-day subsystem fix
facets, stale git remote tables, outgoing email baselines, and
stage discovery/overlap funnels
- Replace full-table statistics scans with four bounded O(1) layers:
- Linux Performance Specialist Review Stage and Benchmarking:
- Add an optional performance specialist analysis stage to
linux_patch_review (available in daemon and local sashiko review,
including --stages performance and --cheap) with second-order
side-effect tracing and anti-noise guardrails - Add expect_clean neutral benchmark support, per-stage
false-positive tracking, and the balanced 40-commit
benchmark_performance.json suite (20 upstream regressions + 20
soak-verified clean commits)
- Add an optional performance specialist analysis stage to
- Consolidated Series Summaries and Opt-In Failure Notification Emails:
- Add patchset-level email_outbox support (migration 023) and replace
N per-patch positive emails on multi-patch series with a single
consolidated cover-letter summary while keeping per-patch inline
replies for patches with findings or related pre-existing bugs - Add opt-in failed_review ("none" | "author" | "all") mailing-list
policy with strict cross-posting isolation to notify authors when a
series exhausts retries in Failed To Apply or Failed state without
leaking raw git stderr or embargoed findings
- Add patchset-level email_outbox support (migration 023) and replace
- Per-Subsystem Baseline Rules and Late Cover-Letter Requeue:
- Support ordered per-mailing-list baselines rules (url, branches,
subject_prefixes, paths) in projects//mailing_lists.toml
with specificity ranking and scheme-agnostic remote deduplication,
configured for bpf, damon, linux-ext4, netdev, mptcp, linux-nfs,
intel-wired-lan, and linux-mm - Automatically requeue Failed To Apply patchsets when a late 0/N
cover letter arrives carrying base-commit, prerequisite-*, version,
or stable backport metadata
- Support ordered per-mailing-list baselines rules (url, branches,
- Review Engine, Maintainer Notes, AI Providers, and Maintenance:
- Record BestEffort stage failures in failed_stages and review_map
so sashiko review exits 3 with "Review incomplete" (and --agent
status "incomplete"), and return all validation violations at once
from concerns validators across all projects - Add Linux DTS and DT/OF maintainer notes (dts.md, of.md) and expand
dt-bindings.md with property ordering, clock items, and subject
checks; track gfs2@lists.linux.dev - Disable idle TCP keepalive and TCP_USER_TIMEOUT on Claude HTTP
sockets, index the local AI response cache on created_at to avoid
full-table startup scans, and isolate git_cmd test runs from host
~/.gitconfig URL rewrites - Update PRESS.md with Google Cloud Blog and Phoronix v0.9.4
coverage, update lettre to 0.11.24 (RUSTSEC-2026-0560 through
2026-0565), and bump clap (4.6.8) and uuid (1.29.3)
- Record BestEffort stage failures in failed_stages and review_map