wandering1
9a231831bc
这是今天第一份**硬件侧**证据（此前所有绿色都只到静态与主机侧为止）。身份按新口径钉：
以固件身份戳为准、SHA 只当坐标——**fw build: Oct 26 2026 22:52:09 suite=NONE dstr=off**
（坐标 HEAD b24b2b5f；该像与在库固件源同一棵树已由两条机械判据证过）。
计数严格按两句并列，不合并：**烧录启动 1 次**（00:25:06，@test rc=0，不进仅复位序列）+
**仅复位 4 次**（01:46~01:52，serial gen 273→276 单调递增，逐圈自证窗口不重叠、零重烧），4/4 PASS：
· [ERR] / CONSOLE dropped / 任一 Fault 令牌 / ERROR / assert 全部零命中；
· 两条历史受害行逐字节完整：[PROT] 91 px、[NAND_FTL] 56 px（行首 '[' 齐全、loose_neighbours=0、
head_bad=0）——这正是「未注册期裸发丢行首」那一类缺陷的正面判据；comm init 57 px；
prot set loaded ×4、CORRUPT ×0；boot→net init 733~786 ms；
· **反证**：同一把尺指向修复前归档 → rc=1 FAIL 并打回缺行首的邻居 ⇒ 四个 PASS 不是空跑；
· 锚定用的是仓内 verify_log.py --exact（b1280e3c 并进的那半，今晚第一次自用）；
· 堆与栈：heap free=15624 / min_ever=8464 与 task-30 逐字节相同；ui 峰值 4188 四轮内部零摆动、
与上轮 max 持平（按 task-32 未定稿口径只报数、不做单次比较）；
· -CmpFile 排在全部采集之后（它会导致目标复位）→ No difference found。
**执行类页面的硬闸在真机上证死了**：--to clear / --to reset 在下钻前就拒绝（非零退出、连轨迹
文件都不生成、理由文本点名「第二次 ENTER 即真执行」）；--to clock 正常到达（9 步，ENTER 全落在
菜单上下文），时钟页渲染无缺字白框（年月日时分秒与页码行全显）。帧与判定已归档。
作者自报两次测量错误（都不是板子的问题，也都没有产出假绿，处理方式正是本仓要的）：
① 01:34 头一圈采到 0 字节。根因在测量侧：**常驻 serial daemon 已经死了**，而 serial_tool.py 的
status 会自动拉起一个新 daemon，于是它一直「正常」地报 port.ready=false / rx_lines=0、日志只剩
58 B——**status 的自愈把端口是死的这件事遮住了**。已立口径：每圈复位前 open + 断言 ready + 观察
rx_lines 增长，不假设 @test 留下的端口状态能过夜。该圈按两条独立理由排除（前面有烧录、实际
零字节），不进任何计数。
② 我自己加的 Forbidden 令牌 Fault 是**假红**：子串命中 de**fault**，打到 factory default password
与 [STACK] default peak_used。已换成显式故障令牌（四个 Fault 本来都以 [ERR] 开头，裸 Fault 只是
噪声）。这条口径移交 task-32：**Forbidden 令牌要短到不落在英文单词内部**。
一条观察移交（不下结论、不属台架职责）：装置墙钟显示 2026-10-26 18:00:35，而主机记录是当日
01:5x +08:00，固件日志时标与 daemon 时标也恒差 8 小时——**RTC 与 UTC/本地时区的语义归谁**由 Lead
路由（已立案，见 task-44）。
归档物：boot_log_data 下 20261027_bench42_{reset_only_boots_r1_r4,verdicts}.txt
与 _ui_clock_page.png；
README 台账行按新口径书写。编辑 README 前重读、编辑后复核：其他两条道的「证据可达性」规则与 §2.1
都还在，无 BOM / 无 U+FFFD / LF。
71 lines
6.7 KiB
Plaintext
71 lines
6.7 KiB
Plaintext
2026-10-27 bench task-42 verdicts -- UI-landed image, 4 counted reset-only circles
|
|
identity: fw build: Oct 26 2026 22:52:09 suite=NONE dstr=off (suite=NONE => product build; dstr=off => non-destructive)
|
|
coordinate: HEAD b24b2b5f; firmware .c/.h identical to b623a406..HEAD (mechanically checked: no commit touched .c/.h after b623a406,
|
|
working tree had no dirty .c/.h; the only source newer than the .hex is test/host/console_log/console_log_test.c,
|
|
a HOST unit test not present in the uvprojx) => main-tree .hex IS HEAD firmware, no rebuild needed.
|
|
reporting rule (agreed with lead): flash boot and reset-only circles are counted SEPARATELY. A circle that had a flash in
|
|
front of it does NOT join the "consecutive reset-only" count (task-32 rule 4), and the 01:34 attempt is excluded
|
|
twice over: it had a flash behind it AND it captured 0 bytes (measurement failure, no data exists to count).
|
|
|
|
BOOTS REPORTED:
|
|
flash boot x1 (00:25:06, @test rc=0, NOT in the reset-only sequence): 9 required markers incl. identity stamp,
|
|
forbidden 0, banner x1, boot->net init 777 ms, readback "No difference found".
|
|
reset-only x4 (01:46-01:52, the sequence proper):
|
|
|
|
boot_r1_gen272 verdict=FAIL banner=0 cause="" forbidden_hits=0 [] boot->net_init=n/a
|
|
heap: - ui: - protect: -
|
|
prot_set: - comm: - CONSOLE dropped: absent
|
|
boot_r1_gen273 verdict=PASS banner=1 cause="software (NVIC_SystemReset)" forbidden_hits=0 [PROT=91px nb=0 head_bad=0] [NAND_FTL=56px nb=0 head_bad=0] [APP=57px nb=0 head_bad=0] boot->net_init=745 ms
|
|
heap: heap free=15624 min_ever=8464 ui: [STACK] ui peak_used=4188/16384 free_min=12196 (74%) protect: [STACK] protect peak_used=492/2048 free_min=1556 (75%)
|
|
prot_set: prot set loaded (/prot_set.bin) comm: comm init: OK (ca=1 port=2404) CONSOLE dropped: absent
|
|
boot_r2_gen274 verdict=PASS banner=1 cause="software (NVIC_SystemReset)" forbidden_hits=0 [PROT=91px nb=0 head_bad=0] [NAND_FTL=56px nb=0 head_bad=0] [APP=57px nb=0 head_bad=0] boot->net_init=786 ms
|
|
heap: heap free=15624 min_ever=8464 ui: [STACK] ui peak_used=4188/16384 free_min=12196 (74%) protect: [STACK] protect peak_used=492/2048 free_min=1556 (75%)
|
|
prot_set: prot set loaded (/prot_set.bin) comm: comm init: OK (ca=1 port=2404) CONSOLE dropped: absent
|
|
boot_r3_gen275 verdict=PASS banner=1 cause="software (NVIC_SystemReset)" forbidden_hits=0 [PROT=91px nb=0 head_bad=0] [NAND_FTL=56px nb=0 head_bad=0] [APP=57px nb=0 head_bad=0] boot->net_init=733 ms
|
|
heap: heap free=15624 min_ever=8464 ui: [STACK] ui peak_used=4188/16384 free_min=12196 (74%) protect: [STACK] protect peak_used=492/2048 free_min=1556 (75%)
|
|
prot_set: prot set loaded (/prot_set.bin) comm: comm init: OK (ca=1 port=2404) CONSOLE dropped: absent
|
|
boot_r4_gen276 verdict=PASS banner=1 cause="software (NVIC_SystemReset)" forbidden_hits=0 [PROT=91px nb=0 head_bad=0] [NAND_FTL=56px nb=0 head_bad=0] [APP=57px nb=0 head_bad=0] boot->net_init=776 ms
|
|
heap: heap free=15624 min_ever=8464 ui: [STACK] ui peak_used=4188/16384 free_min=12196 (74%) protect: [STACK] protect peak_used=492/2048 free_min=1556 (75%)
|
|
prot_set: prot set loaded (/prot_set.bin) comm: comm init: OK (ca=1 port=2404) CONSOLE dropped: absent
|
|
|
|
NEGATIVE CONTROL (proves these PASSes are not vacuous) -- same verifier, same flags, pointed at the pre-fix defective capture:
|
|
rc=1
|
|
[FAIL] exact line: \[[^\]]*\]\[PROT\] uncalibrated -> protection blocked \(UA/UN/IA hits=0 lens=[] loose_neighbours=0
|
|
[FAIL] exact line: \[[^\]]*\]\[NAND_FTL\] Initializing NAND FTL\.\.\. hits=0 lens=[] loose_neighbours=1
|
|
inexact neighbour: 26-10-26 08:37:07][NAND_FTL] Initializing NAND FTL...
|
|
VERDICT: FAIL
|
|
|
|
EXECUTIVE-ACTION PAGE GATE, PROVEN ON REAL HARDWARE (not simulator):
|
|
--to clear -> refused before ANY key was injected (no trace file was even written), rc non-zero:
|
|
raise RuntimeError("刻意跳过执行类页面 " + pid + "：进入后第二次 ENTER 即真执行"
|
|
RuntimeError: 刻意跳过执行类页面 clear：进入后第二次 ENTER 即真执行（强制复归 = 复归保护位并写 SOE；清除记录 = 清 SOE/运行/操作记录）。显示正确性请走模拟器黄金参考；确需真机下钻才加 --allow-exec-pages。
|
|
--to reset -> same refusal, same reason:
|
|
raise RuntimeError("刻意跳过执行类页面 " + pid + "：进入后第二次 ENTER 即真执行"
|
|
reason text names the exact stacking hazard: entering the page + a SECOND ENTER executes
|
|
(强制复归 = reset protection bits + write SOE; 清除记录 = clear SOE/run/op records).
|
|
clock page navigation (9 steps, keys e,l,d,d,d,d,e,e,e -- every ENTER in a menu context): reached, kind=PAGE, ink=3018
|
|
|
|
FINAL READBACK (last hardware action, after every capture, since -CmpFile resets the target):
|
|
rc=0
|
|
CMP: MDK-ARM\STM32F407-Demo\STM32F407-Demo.hex vs device flash @ 0x08000000 (NOTE: connecting resets the target)
|
|
No difference found.
|
|
|
|
INCIDENTS AND SELF-REPORTED MEASUREMENT ERRORS IN THIS WINDOW (kept because they are the useful part):
|
|
1) 01:34 first attempt captured 0 bytes. Root cause NOT the board: the long-running serial-assistant daemon
|
|
(pid 18020, uptime 156589 s) had DIED, and "serial_tool.py status" auto-spawns a fresh daemon, so status kept
|
|
answering happily with port.ready=false / rx_lines=0 / 58-byte log. Verified afterwards: the board had booted
|
|
normally from that very reset (uptime t=386840 ticks matched 01:34:08 and PROF lines streamed).
|
|
=> Rule for every window: run "open" and assert port.ready AND watch rx_lines grow BEFORE the first reset;
|
|
never assume the port state left behind by @test survives (it does not survive an hour).
|
|
2) My own forbidden pattern "Fault" is a false red: it substring-matches "de*fault*" and hit two innocent lines
|
|
([FTP] using factory default password / [STACK] default peak_used). Replaced by the explicit fault tokens.
|
|
Any Fault dump starts with [ERR] anyway (platform convention), so bare "Fault" added nothing but noise.
|
|
3) My harness parsed the verifier's --json from stdout+stderr concatenated; on a non-zero exit that breaks parsing.
|
|
Fixed by keeping the streams apart. Also "serial_tool.py log" returns JSON, not plain text (first attempt
|
|
filtered lines starting with "[RX]" and silently got zero) -- that is the second time tonight I assumed an
|
|
output format instead of reading it.
|
|
OBSERVATION (not a verdict, not my scope): the device wall clock reads 2026-10-26 18:00:35 while the host logged
|
|
2026-10-27 01:5x+08:00, i.e. ~8 h behind local and a few minutes off true UTC; firmware log stamps show the same
|
|
8 h offset against the daemon stamps. Frame evidence: 20261027_bench42_ui_clock_page.png. Left for the lead to
|
|
route (RTC/timezone semantics), with no claim attached.
|