Preserve TUN settings and restore proxy operation after fallback.
Deliver fallback notices after UI subscription and stop retrying
deterministic bundle refusals. Publish fallback state atomically
and avoid conflicting service startup notifications.
* fix(profile): virtualize editor lists for rule/proxy/group editors
* refactor(list): remove feedbackClone from sortable item and list integration
* refactor(profile): use type instead of export type for category
* refactor(ui): rename measureElementRef to measureElement in sortable components
linked-issue-check only exempted drafts and the no-issue-needed
label, so member and collaborator PRs failed the linkage gate while
the slop review itself skipped them via skip-roles (#7944).
Skip OWNER, MEMBER, COLLABORATOR, and bot authors as well, reading
the association from the event payload so the permission-free gate
design holds. External contributors still need an issue reference.
Manual update-check failures surfaced as red error toasts that read as
an app fault. Add a warning severity to the notice service and use it
in both check entry points alongside a console.warn; automatic checks
and install failures keep their existing behavior.
* fix(traffic): avoid scientific notation in traffic display near 1000 threshold
When a traffic value in any unit (KB, MB, etc.) falls in the 999.5–1000
range, toPrecision(3) rounds it to 1000 but outputs scientific notation
"1.00e+3" instead of plain "1000". Check Math.round(dat) >= 1000 instead
of dat >= 1000 to catch this rounding boundary.
Signed-off-by: Alec <alechron@outlook.com>
* fix(traffic): avoid scientific notation in formatConnectionTraffic
Signed-off-by: Alec <alechron@outlook.com>
---------
Signed-off-by: Alec <alechron@outlook.com>
Add `mips` to the TUN stack switch so the stack shipped in mihomo
v1.19.31 is selectable from the GUI. Older cores reject the value,
which the changelog entry discloses.
The service installer now admits the staged core copies through Windows
Firewall, so TUN system/mixed stacks pass traffic in service mode again
without users re-allowing the new path by hand.
Refs #7915
Limit GUI authority to saved TUN fields while preserving custom overrides.
Sync provider caches after service startup, reloads and updates.
Publish copies atomically with session, runtime and cache metadata checks.
Removes zero-consumer surface verified by repo-wide grep: the
reference-count render machinery, the never-called clearData chain,
unread snapshot fields (rangeMinutes, reason), dead refresh callbacks
on the connection hooks, the write-only wsRef sync, and the worker's
unreachable DEFAULT_CONFIG placeholder (now definite assignment under
the init-first client contract). No behavior change; the compressed
sampler path, samplerStats diagnostics, and lastTimestamp perf
contract stay. Net -120 lines across six files.
The traffic pipeline re-rendered twice per second (worker snapshot
plus a 1 s now-tick), cloned the 10-minute and 60-minute arrays
through every snapshot, and re-filtered them on the main thread.
Snapshots now carry only the range slice the graph renders, a 1 s
worker-side heartbeat re-broadcasts only when points age out of the
window, and snapshot ingestion pauses while the window is hidden.
Perf harness A/B: hidden-phase webview CPU 2.45% -> 0.77%, main
0.90% -> 0.37%, webview RSS about -3 MB; visible phases unchanged.
The secondary wait only polled the notify endpoint, which a restarting instance has already shut down before spawning its successor, so a child that lost the spawn-vs-exit race could never acquire the freed lock and died after 20s behind a startup-failure dialog. Retry try_lock_instance on every poll so the child promotes as soon as the old process exits. Double-launch handoff and wedged-holder timeout behavior are unchanged.
The v2.5.4 notes had grown to 97 free-form entries exposed verbatim in
the in-app update dialog. Distill them to 58 one-theme-per-line entries
capped at ~60 CJK characters with no internal terminology, and move the
writing rules into template/Changelog.md so every cycle starts from them.
Release plumbing now agrees on the --- divider: extract_update_logs.sh
stops there like updatelog.mjs, so file-only content never reaches
release bodies, while updater.mjs appends a per-tag See More link to the
update notes it generates. scripts-workflow/group_platforms.sh locks
platform entries under emoji platform headings, wired into pre-commit,
release and autobuild CI, and the changelog bump script; AGENTS.md points
agents at the template rules.
Every settings toggle reparsed the profile and chain files, ran four
Boa interpreters, serialized the runtime config twice, and re-ran
mihomo -t over bytes the core had just accepted. Builtin and template
scripts now run without Boa, custom scripts bind the config through
JSON.parse and take ownership instead of per-script clones, the config
serializes once, and mihomo -t is skipped unless the bytes, core
build, or geo assets changed. Heavy parses and serializations moved
to spawn_blocking. Large-subscription toggles drop from hundreds of
milliseconds to ~140ms with no new retained state.
* fix(ui): use utf-8 encoding for inline SVG icons to support non-latin characters
* chore(test-item): re-add data-sortable-handle to TestItem box
* docs: update changelog
* fix(macos): discard a core cache the current user cannot write
* docs(changelog): add the unwritable core cache fix
* docs: update changelog to include Linux in cache.db ownership fix
---------
Co-authored-by: oomeow <oomeow@outlook.com>
* feat(service): hand upgraded cores to the service through the installer
The service now only executes administrator-staged core copies, so an
in-app core upgrade has to deliver the new bytes there before the
restart asks for them; otherwise a service-mode restart keeps running
the previous core while the UI reports the new version.
After publishing the downloaded core, service-mode upgrades invoke the
elevated service installer with --install-core <path> --sha256 <hex>.
The digest is computed over the published file and rides the elevated
command line, which no other local account can alter, so a file
swapped after hashing is refused rather than staged. Sidecar-mode
upgrades run the file directly and prompt for nothing.
Rollback distinguishes staging that never ran, was refused, or
succeeded: only a successful staging obliges the rollback to restore
the approved copy too (deliberately not gated on the running mode,
which reads NotRunning after the very crash being rolled back), and a
refused staging restores the file so a retry does not see the new
version on disk and conclude there is nothing to do.
Requires clash-verge-service-ipc >= 2.6.7. The pinned dependency and
the packaged service release must move together with this change, or
the old installer silently treats --install-core as a plain install.
* feat(installer): attest bundled cores to the service at install time
prebuild computes sha256 digests of the downloaded mihomo sidecars
into packages/windows/core-hashes.nsh (gitignored), included through
the NSIS installerHooks path. After unpacking and before starting the
service, the installer stages both cores into the service's approved
directory with those digests, so what gets attested is exactly what
this installer shipped: a core swapped on disk in a user-writable
install directory is refused, not published. Failure is logged and
does not abort the install; a later elevated service install can
stage the cores again.
* chore(deps): require clash-verge-service-ipc v2.6.8
v2.6.8 is the first service whose installer understands
--install-core --sha256; the staging this branch introduces is inert
against anything older, because the old installer treats the flag as a
plain service installation and reports success. The service release and
this pin therefore have to ship in the same app release.
* fix(service): heap-allocate the core hashing buffer
clippy::large_stack_arrays denies the 64 KiB stack array.
Two months of usage data (56 evaluated runs, 1372 AI credits) show a
median of 22 credits and 6 tool turns per review, with 97% of billed
context being cache tokens anchored on the prompt itself.
Cap each run at 12 turns and 60 credits so only runaway sessions are
stopped, and stop triggering on reopened, which carries no new
implementation evidence; workflow_dispatch remains available for
manual re-reviews.
Compact the policy body from 367 to 177 lines by merging the signal
lists with the weighting/escalation sections that repeated them, with
no verdict, label, rerun-stability, or output-contract semantics
dropped. Add a tool-call budget for normal evaluations and a fast path
for fork-sync and placeholder PRs, which were 29% of measured
credits.
Codify the rules agents and contributors are judged by: a tracked
AGENTS.md instruction contract, a Contribution Expectations section in
CONTRIBUTING, and a minimal PR template. Align the AI-slop review
policy with them - linkage substance, test/defensive-code padding,
write-access skips, disclosure as transparency - and add two
tamper-resistant checks: an ai-slop:high merge gate with a linked-issue
requirement (no-issue-needed exemption), plus a docs-consistency anchor
check guarding the cross-references.
* refactor(log): migrate logging to tracing-estuary pipeline
Swap flexi_logger + clash-verge-logger wiring for the tracing-estuary
crate: buffered file sink with explicit panic/exit drains, live level
switching, and tokio-trace/chrome-trace behind crate features.
AsyncLogger moves in-house (VecDeque ring); file format and rotation
stay unchanged.
* fix(log): reset sidecar writer in place and re-pin service-ipc v2.6.7
Rebuilding the sidecar FileLogWriter on config changes leaked flexi's
unstoppable flusher thread plus an fd each time; the writer is now flushed
and reset in place, matching what the service crate does.
get_clash_logs_by_service drops the CompactString conversion layer now that
service-ipc v2.6.7 returns plain Vec<String>, and AsyncLogger is renamed
LogRing to match.
* refactor(log): span-based observability and log noise cleanup
Instrument spans carry the structure flat lines never could: the core-start
spine (mode, readiness/owner generations, pid, config file), proxy apply and
clear routes with guard generations, service IPC sessions with refusal codes
and outcomes, profile-switch terminal outcomes, subscription fetch strategies,
core upgrades with per-proxy version probes, backup flows, timer tasks, config
generation uids, and the startup setup flow; running-mode changes emit one
guarded transition event each.
~90 hand-written noise lines are gone or demoted (entry+success pairs, polled
getter echoes, phase brackets, window sub-steps, platform install chatter);
errors keep their anyhow chain, operation name, and item id, and swallowed
retry results surface per attempt.
pnpm 12.3.4 deterministically breaks every Windows CI build: the install inside the tauri beforeBuildCommand fails importing into the pnpm store with os error 32 (ERROR_SHARING_VIOLATION). Reproduced on fresh runners with caches wiped; 12.3.3 was green on the same pipeline. Re-pin to 12.3.3 until upstream ships a fix.
wonfen