Fix 9 small bugs in the shell scripts#309
Open
seyedehsanhadi wants to merge 46 commits into
Open
Conversation
fix3: prefer a charging switch whose off-state persists (skip flicker-prone level switches when a clean one exists; lenient fallback so a device with only a flickery switch is still capped). fix4: never re-enable charging while pausing at/above the limit; select a switch once per accd session instead of re-probing every loop; fail-safe pause/resume capacity comparators (missing value => pause / do not resume). fix5: on Pixel/Tensor, disable the dynamic google charge_stop_level 'battery/capacity' variant (it ratchets and makes the firmware re-resume at the threshold); the fixed '100 5' variant holds the limit without churn. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
acc -t waits up to _STI=35s per switch; on a long list that overruns AccA's 150s cap before finishing. This polls the charging current ~3x/sec, decides each switch in ~1-4s, tests the whole device switch list, ranks working ones fastest-first, and prints the exact 'acc -s s=... --' command. Standalone; stops accd only for the scan and restarts it (EXIT trap), restores each switch to on after testing. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
google charge_stop_level is a charge LIMIT ("charge to N%, hold"), not an
on/off switch. Driving it 100/5 (fix5/fix6) writes 5 to pause, which makes the
firmware DISCHARGE; accd then resumes at resume_capacity and re-charges,
overshooting the cap in a 70<->limit sawtooth (reported breach 75 -> 77 on a
Pixel 9a, tegu).
- ctrl-files.sh: prefer "charge_stop_level 100 pcap" (pcap = pause_capacity), so
the firmware holds the battery flat at the cap (tight, no overshoot, true
idle). Keep "100 5" (discharge) and "100 battery/capacity" as fallbacks.
- misc-functions.sh flip_sw: resolve the "pcap" off-token to pause_capacity;
fail-safe to a low cap if unset (can only stop, never run on).
- acc-switch-scan.sh: resolve pcap and rank idle/flat-hold above discharging, so
--apply locks the flat-hold variant, not the sawtoothing discharge one.
- misc-functions.sh disable_charging: on an unconfirmed pause, return 7 and let
the loop retry (2022/2023 behavior) instead of exec-ing a daemon re-init
mid-pause (which re-armed charging in its window and thrashed flaky switches).
versionCode kept 202505180 for AccA front-end compatibility.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…loys AccA's bundled installer (res/raw/install) skips installing when the bundled versionCode is <= the already-installed one. Keeping 202505180 across fix4..fix6 meant a new bundle would NOT replace an existing 202505180 daemon on update -- the fix never reached devices that already had ACC. Bumping the patch code makes the gate pass; handler dispatch in AccA is range-based (>=202107280) so this is compatible. AccA bundledVersion is updated to match. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- _ge_pause_cap / _le_resume_cap: treat a NON-numeric pause/resume capacity as "pause now / do not resume" (numeric guard), not just empty. A garbage value used to slip past the empty-only check, make the numeric test error out, and silently skip the cap (overcharge). - disable_charging: a switch locked with `--` that STOPS working now runs the auto fallback (re-selects a working switch) and warns, instead of failing silently. `--` still suppresses routine re-cycling while the switch works, so no churn in the normal case. - versionCode 202505181 -> 202505182. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… watchdog - accd.sh: extend the numeric/empty fail-safe guard to ALL capacity comparators (_lt_pause_cap, _le_pause_cap, _gt_resume_cap, _le_shutdown_cap, _ge_cooldown_cap) -- previously only _ge_pause_cap and _le_resume_cap were guarded. A malformed (non-numeric) limit can no longer make any check error out and skip a stop. - ctrl-files.sh: add the `pcap` (hold-at-pause_capacity) off value to /proc/driver/charger_limit too, so non-Google charge-limit chipsets also hold the cap flat (firmware-native, no overshoot), not just the Google node. - accd.sh: breach watchdog -- if the battery is at/above the limit and charging still has not stopped, post a throttled warning instead of failing silently; self-clears once charging stops (and on daemon start). - versionCode 202505182 -> 202505183. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…obustness - acca.sh daemon_ctrl: `acca -D restart` launched the daemon with a bare `exec accd`, i.e. inside the calling process. When the caller is a one-shot script (the switch scanner run from a front-end), the daemon died the instant that script exited, leaving charging UNCAPPED. Launch it detached (setsid, with start-stop-daemon -b / nohup fallbacks) so it survives. - accd.sh: add _nap(), an interruptible sleep that wakes the moment $config changes, and use it for the three per-iteration loop waits. The daemon already re-reads the config each loop, so AccA edits (limits, temps, switch) now apply within ~1s instead of after the full loop delay -- live, no restart, no UI freeze. - acc-switch-scan.sh: re-arm charging before each test (wait for current to resume) so every switch incl. the flat-hold `pcap` variant is evaluated instead of skipped as "not charging"; and verify the daemon actually came back after a scan, warning loudly (charging UNCAPPED) if not. - versionCode 202505183 -> 202505184. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…on=100 variants
The 75->77 breach was the ON value: charge_stop_level "100 5" / "100 battery/capacity"
recharge with on=100 ("charge to 100%"), so on every resume the firmware sails toward
100 and accd catches it late -> overshoot. Drive the limit node with the TARGET level
on BOTH sides so the recharge stops EXACTLY at pause_capacity:
- ctrl-files.sh google charge_stop_level is now `pcap pcap` (charge to the cap and
hold -- battery-idle, DEFAULT) and `pcap 5` (discharge to resume_capacity then
recharge to the cap, repeat -- the cycle when battery-idle is off; recharge still
tops out exactly at the cap). Removed `100 5` and `100 battery/capacity`.
- misc-functions.sh flip_sw: `pcap` now resolves to pause_capacity on the ON side too
(it was off-only), so on=pcap recharges to exactly the cap instead of 100.
- versionCode 202505184 -> 202505185.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… = discharge)
The scanner already locks the best switch with `--apply`. Add a method choice:
- `--apply` -> lock "hold at the limit" (pcap pcap) [DEFAULT, unchanged].
- `--apply --cycle` -> lock the "discharge-cycle" (pcap 5): discharge to
resume_capacity, then recharge to the cap, repeat.
The method is chosen by the switch LINE (`... pcap pcap` vs `... pcap 5`), not the
device-dependent measured idle/discharging reading, so it is reliable on devices
that slow-drain at the limit. Header/result text reflect the chosen method.
versionCode 202505185 -> 202505186.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Consolidates fix2..fix12 into the first stable release, plus final hardening: - default-config.txt + write-config.sh: allow_idle_above_pcap now defaults FALSE -- never sit ABOVE the limit; if the battery is over it, discharge down to it (also better for longevity: a lower resting SoC). - misc-functions.sh flip_sw: "pcap" falls back to a safe low cap (60) when pause_capacity is empty OR non-numeric, matching the daemon's fail-safe rule (a garbage config can only cap low, never run charging on). - accd.sh: breach-watchdog notification points at the current "Scan & lock" scripts. - versionCode 202505186 -> 202505187; label v2025.5.18-stable. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…o command)
So users never have to run `acc -s`. install.sh now does a one-time, marker-guarded
config migration that flips the OLD default allowIdleAbovePcap=true -> false ("never
sit above the limit") on existing configs. Runs once, only touches that single key,
and never clobbers a deliberate later choice. Fresh installs already ship the new
defaults. versionCode 202505187 -> 202505188.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…the migration Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Owner
|
Great progress! 👏 |
stable.1's allow_idle_above_pcap=false default pushed auto-mode into the discharge variant (charge_stop_level pcap 5), which drained the battery down to resume_capacity (~70) instead of holding at the limit. Fix: remove the pcap-5 variant entirely (only pcap pcap = hold at the limit) and revert allow_idle_above_pcap to its original default. install.sh migrates existing configs (undo the false default, drop a locked pcap-5 switch). Charging now stops at your limit and holds there -- the only behavior. versionCode 202505188 -> 202505189. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Google charge-stop node LATCHES "stopped" once it reaches the limit. stable.2's
`charge_stop_level pcap pcap` used the limit value (75) for BOTH stop and resume, and
re-writing the limit never re-arms the charger -- so once stopped it stayed stopped,
even after the battery drained below the range (user saw it frozen at 64, neither
charging nor discharging). Fix: switch is now `charge_stop_level 100 pcap` -- ON=100
("charge up") re-arms the charger (what 2022/2023 did), OFF=pcap stops AT the limit.
loopDelay[0]=3s keeps the stop tight (no overshoot). install.sh upgrades a locked
pcap-pcap/pcap-5 switch to 100 pcap automatically. versionCode 202505189 -> 202505190.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- ACC is now independent of the AccA app. Removed the service.d cleanup script that deleted ACC (module + daemon) when AccA was uninstalled; install.sh also removes any leftover on update. Uninstalling the front-end no longer touches the daemon or the persistent config (config.txt). The accaFiles symlink stays so AccA can still find it. - Default max_temp 50 -> 45 C (heat above ~45 ages the cell faster). default-config + write-config fallback; one-time migration updates a config still on the old 50 (a deliberate value is left alone). - versionCode 202505190 -> 202505191. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
stable.4 lowered max_temp 50->45, leaving it equal to cooldown_temp (temperature=(45 45 40 55)). cooldown_temp starts the cooldown cycle and max_temp is the hard pause; when equal, the cooldown loop enters and immediately breaks at max_temp, so it never throttles. Restore the upstream band temperature=(45 50 40 55). - write-config: default mt=50 and enforce cooldown_temp < max_temp (and >= resume_temp) on every rewrite, so it cannot re-collapse. - migration: replaced the 50->45 sed in all three installers with a .stable-defaults5 repair that turns a stuck (45 45 ...) back into (45 50 ...); only the exact collapsed signature is touched. - versionCode 202505191 -> 202505192; README/module.json regenerated. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The stable.5 build ran build.sh on a host with no `markdown` binary, which (1) truncated README.html to 0 bytes (the `markdown > README.html` redirect) and (2) leaked the build host's TMPDIR into README.md via print_help (/ch-switches -> /c/Users/.../Temp/...). Restore both READMEs from the last clean commit (they already embed temperature=(45 50 40 55) and the correct on-device paths) and harden build.sh to rewrite README.html only when a markdown converter actually produces non-empty output. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New install/state-export.sh publishes $TMPDIR/state.json (tmpfs) every daemon loop and on demand via `acca --state` / `acc -j`. It is the read-back the control bus confirms against, the diagnostics feed, and the report source — all one file. Contracts enforced + tested: atomic (temp+rename, no torn reads); best-effort and non-blocking (never stalls the safety loop); tmpfs-only (no flash wear); fingerprint is non-PII (model|soc|hardware|buildId, no serial/IMEI); and rule S1 — a value that can't be read is JSON null, never 0 (proven: a garbage current read -> null). Wiring: registered in misc-functions.sh (seen by daemon + CLI); write_state hooked at the top of the accd loop; `-j|--state` added to the acc.sh dispatcher. Local only, not pushed. First increment of the A-E rebuild. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
RC1 toward stable.6. Ships the acca --state state export (prior commit) as the keystone for the AccA control-bus + diagnostics rebuild. Additive only -- no charging-behavior change. Hard-test gate (30 checks: temperature band, write-config cooldown<max guard, all three installer migrations, README integrity, state-export JSON validity + S1 null-on-bad-read, flashable-zip contents, and the parser/fetch/ device fuzz models) ALL PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Found on-device: acc.version/versionCode empty (accVer is only set in acc.sh, not the daemon/acca paths) -> now read from module.prop directly; status "unknown" on the front-end path (it never calls read_status) -> now derived from the current sign (Charging/Discharging/Idle); and a frozen snapshot (print_state cat an existing file) -> print_state now always refreshes. Additive; no charging-behavior change. State export functional test (version + derived status + refresh) passes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…larity/measuredClass) Adds live measured sensing to acca --state, generalized beyond Tensor: plugged (any */online), currentUnits (uA/mA auto), polarity (status vs current sign), and switch.measuredClass (bypass/idle/charging/discharging from plug + unit-aware current). Additive; no charging-behavior change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…-- fixes limit overshoot on Tensor trap node Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rap; offer 100 pcap + 100 5 (2022/2023) drivings Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ps re-probing (fixes stop-then-reset sawtooth at the limit) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… a working stop switch that discharges (charge_stop_level 100 5, negative current) is no longer wrongly rejected; now it locks and holds Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…OR discharge), drop the idle filter + once-per-session sentinel, retry until locked. Finally locks charge_stop_level on discharge-only A16 Pixel Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… + prioritize_batt_idle_mode=no) so auto-lock grabs the working current_max switch -- charge_stop_level is dead on A16, usb/current_max 0 is the real stop Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…s (usb/main-charger/constant_charge_current/input_current wildcards) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…7) + all-paths current-cut group switch + write-config param hardening Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The `acca -s key=value` write path (used by every front-end config change) re-exec'd itself through a BARE `setsid`. On busybox builds without setsid the re-exec returned 127 and `set -e` aborted before write-config ran -- the write silently no-opped. Read commands never touch setsid, so reads worked while writes did nothing. `exec 4<>$0` self-flock was a second failure on read-only module dirs. Fix: `-s` writes synchronously now (accd re-reads config.txt every loop, so the writer need not detach), with a guarded best-effort flock on a tmpfs lock that is skipped when flock is absent. Front-ends get a real exit code. This also unblocks the Scan & lock scripts: --apply locks via `acca -s "s=... --"`, which silently failed pre-rc14 on setsid-less devices. Also sync customize.sh + META-INF update-binary to install.sh (they had drifted to the pre-rc12 installer, missing the busybox self-healing fallbacks). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…n first over-limit charge When the battery reaches your limit but charging has not stopped (no switch is actually capping on this phone), the daemon now auto-runs the tested switch scanner (acc-switch-scan.sh --apply) ONCE per breach episode -- detached, self-healing -- which finds the switch that stops charging and locks it (charging_switch=<best> --). The lock lives in config.txt, which ACC owns, so it persists across reboot. If the config is reset or ACC is reinstalled, the lock is gone and the next over-limit charge re-detects + re-locks automatically. No user action, no AccA needed. - Opt out with `touch /data/adb/vr25/acc-data/.no-autolock` (falls back to the old notify-only nudge). - Never overrides a switch the user locked manually (trailing " --"). - Bundles the rc14 acca -s setsid fix (which the --apply lock depends on). module.prop 202505205 -> 202505206. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Turns the auto scan+lock into a guaranteed X..Y contract and applies the full stress-test patch set. Charging-behavior changes are fail-safe and additive. Daemon (accd.sh, misc-functions.sh): - Charge-state correctness: reset the whole auto-lock campaign whenever charging stops/unplugs, so a transient plug/unplug blip is never mistaken for "charging past the limit" and the next real charge starts a clean detection. - Debounced trigger: only act when PLUGGED + over-limit + still-charging, confirmed over a short re-check. - Runtime CONTRACT MONITOR: if a LOCKED switch lets the cap breach for 3 confirmed loops, unlock it, blacklist it for the session, and re-detect another. cycle_switches honors that blacklist so the daemon's own locker won't re-pick a failing switch. This makes the range a guarantee, not a one-shot guess. - Bounded auto-detect: max 3 scans per episode, then a LOUD give-up notification -- never loops, never goes silently uncapped (audit bug). - Crash recovery: clear transient markers on (re)start; keep attempt-count/blacklist. - Mode passed to the scanner from config (Hold@Limit iff allow_idle_above_pcap). Scanner (acc-switch-scan.sh): - INPUT/charger-current sensing -> classify each stop as bypass (true hold, charger still feeds phone) / clean / DRAIN (charger dead, battery powers load). DRAIN switches are REJECTED from locking (kills the Sony/passthrough silent-drain). - RESUME verification: a switch must stop AND verifiably re-arm to rank top (guarantees the cycle can recharge from X). Unverified ones are downranked. - Single-instance flock mutex (auto vs manual scan), thermal-throttle warning, blacklist-skip, PD-renegotiation tolerance (wait widened), no-switch marker. - Default method = Range Cycle; Hold@Limit (bypass) is the opt-in. Coverage (ctrl-files.sh): add OnePlus/OPPO mmi_charging_enable and Samsung batt_full_capacity (SOC-relative). Default config: allow_idle_above_pcap=false (Range Cycle is the default; set true for Hold@Limit). Decision logic state-machine tested (classify, ranking, attempt-cap, give-up, contract monitor + blacklist). On-device verification still required. module.prop 202505206 -> 202505207. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…reboot fix) + recovery + per-switch diagnostics ROOT CAUSE of "another scan already running" + "doesn't charge until reboot" (Pixel 9a, multi-switch device): the rc16 breach watchdog SPAWNED the external acc-switch-scan.sh --apply from inside the daemon loop. That scanner `acca -D stop`s the daemon and toggles switches in a detached process Android can kill; a kill skips its restore trap, leaving a current node (e.g. usb/current_max) pinned at 0 = NO CHARGE until reboot, and it holds the scan lock so the user's manual "Scan & lock" aborts with "another scan already running". The daemon ALSO auto- locks in-process via cycle_switches_off, so the spawn was redundant AND a race. Fix A (accd.sh): the daemon NEVER spawns the scanner. cycle_switches_off (its own current-verified in-process locker) is the only auto path; the breach watchdog now only monitors (unlock+blacklist a locked switch that stops holding) and surfaces state (bounded notify -> loud give-up). No daemon stop, no detached process, no scan-lock contention, no race. Fix B (accd.sh): startup recovery -- when plugged, restore all candidate switches to ON once (subshell-isolated cycle_switches on), so a node a killed scan left at 0 is un-pinned and charging recovers via AccA "restart daemon", NO reboot needed. Fix C (acc-switch-scan.sh): cleanup now restores EVERY switch to ON (restore_all_on), not just the in-flight one, so a graceful exit/error never leaves a node pinned off. Writes per-switch results to logs/switch-test.log (WORKS/DRAINS/NOEFFECT/SKIP). (AccA rc19 surfaces switch-test.log + working-switches.log in diagnostics so you can see which method works/doesn't on each phone.) module.prop 202505207 -> 202505208. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…charge_start_level) THE fix for Pixel 9a (and any google,charger device): confirmed on-device that the generic switch model fails here -- current_max=0 does NOT stop Tensor charging (4.59A flowed with all current_max nodes at 0, SOC above limit), and charge_stop_level as a TOGGLE either drains (off=5) or fights the firmware. ACC also never set charge_start_level at all. The NATIVE firmware pair works perfectly (user confirmed: charge_stop_level=40 + charge_start_level=39 -> holds idle at 40%, no overshoot, no drain) -- this is what the 2023-era ACC did. rc20: when google,charger/charge_stop_level + charge_start_level both exist, the daemon DRIVES THE NATIVE PAIR from pause/resume_capacity (charge_stop_level=pause, charge_start_level=resume) and SKIPS the on/off toggle entirely -- the firmware holds at the limit and resumes at the start level. Re-syncs every loop so AccA limit changes apply live; forces a pause (lowers the stop level) at/above max_temp; sets the levels at startup instead of cycle_switches-on (no 100-overshoot). Opt out for the generic logic: touch /data/adb/vr25/acc-data/.no-native-limit sync_native_limit logic unit-tested (cool/hot/limit-change). module.prop 202505208 -> 209. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…led write) write-config.sh wrote config.txt with `> $config` (truncate-in-place) then appended scripts/help with `>>`. The daemon re-reads config.txt every loop, so it could read a half-written file; and a failed/partial write (disk full, permission loss) left config.txt truncated while acca -s still reported success. Now the full config is written to $config.tmp and ATOMICALLY renamed (mv) into place -- the daemon always sees a complete file, and a failed write leaves the previous config intact. module.prop 202505209 -> 202505210. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…erted limits) From the final pre-stable audit (findings verified vs code; theoretical ones rejected): - capacity-mask divide-by-zero guarded: skip masking when pause_capacity == shutdown_capacity (the masked formula's denominator is pause-shutdown). - write-config now enforces shutdown_capacity < resume_capacity (%-mode): an inverted config could otherwise shut the phone down at a high level or never pause. - native limit clamped to [0..100] before writing charge_stop_level / charge_start_level, so a bad/out-of-range config value can never be written raw to the firmware nodes. module.prop 202505210 -> 202505211. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Identical tested code; drops the -rc label. versionCode stays 202505211 (in lockstep with the AccA bundle). Stable.6 line: native Pixel/Tensor charge_stop_level limit, all-paths group switch, brick-safe switch probing, atomic config writes + audit safety bounds, deep sleep, install robustness, acca --state export. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ted-current kernels Some kernels (e.g. Motorola XT2301) report current_now inverted (charging negative). The scan's 'nr < 0 = stopped' test false-positived every switch there. Anchor stop-detection to a reversal vs the charging baseline's own sign (chgNeg/revd); Pixel negative-discharge path unchanged. Scan-only; charge-limiting (capacity/online based) was never affected. NOT released -- private test build acc_v2025.5.18-stable.6.1.zip for one Motorola tester. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ority, sign+unit-robust auto-lock, classification + online-node + spurious-shutdown fixes) From a deep multi-agent audit across SoC families. Order-preserving switch-list de-dup (was sort -u -> alphabetized, clobbering curated priority). Daemon anti-flicker check made sign-agnostic + unit-scaled (matches the scan fix): mA and inverted-sign kernels no longer false-lock; Pixel discharge-hold preserved. Scan inFile: drop static usb/current_max, add dc/wireless live nodes. online_f: + main-charger/oplus/glink. Spurious-shutdown guard on garbage shutdown_temp. All sh -n clean + adversarially reviewed SHIP. AccA needs no control change. versionCode 202505212. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The environment deleted accd.sh, misc-functions.sh, acc-switch-scan.sh, batt-interface.sh and module.prop from the working tree before the stable.6.1 commit, so that commit recorded their DELETION. The published acc_v2025.5.18-stable.6.1.zip asset was built before the wipe and is correct; these files are recovered verbatim from it (hardening intact: order-preserving de-dup, sign-agnostic+unit-scaled auto-lock, scan classification, online-node, spurious-shutdown guard). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
rc20 delegated resume to the google,charger firmware via charge_start_level, but the Tensor driver latches "stopped" and ignores it (upstream bug). After the limit was hit, re-plugging did nothing; only a reboot restored charging. native_unlatch detects the latched state (plugged + at/below resume_capacity, or a fresh plug-in below the limit, while the kernel still reports not-charging) and pulses charge_stop_level=100 — the only value that re-arms the FET — then sync_native_limit restores the real stop (no overshoot). Self-disables on healthy firmware, never fires in the [resume..limit] idle band (no sawtooth), cannot overcharge. Non-Pixel (nativeLimit=false) paths untouched. Verified by an 8-case mock-sysfs regression harness. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
6.2 fixed Pixel/Tensor (native_unlatch). 6.3 extends it to generic devices (Motorola/Qualcomm/etc.) whose switch (input_suspend, current_max=0, charging_enabled=0) holds "off" across an unplug/replug — after the limit was hit, re-plugging did nothing until resume_capacity, or a reboot on latching switches (the reported Motorola symptom: stops at limit, won't resume on replug). - shared loop-top plug tracker (freshPlug/wasOnline) drives both paths - native_unlatch refactored to use freshPlug - generic_rearm: fresh plug below limit -> enable_charging; one-shot (no sawtooth), skips boot loop (.minCapMax) to respect off_mid_charge, online-gated Verified by a 13-case mock-sysfs harness; original logic harness 17/17 unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Nine bugs I hit while reading through the scripts. All small, no behavior or safety
changes, bash -n clean.