Files
d2r-mobile/dm/verify/accept.sh
T
jhaizhouandhubeiqiao 66c3a2fb70 Fix setup on stock macOS bash, shader step false success, battery gate misreport (#4, #5, #7)
- Brace every $NAME that is directly followed by a non-ASCII character in the shell scripts: bash 3.2 in a UTF-8 locale read the first byte of the following character as part of the variable name and aborted under set -u (#4, #6).
- setup.sh and d2rm_pack.py: the shader translation step is complete only when every entry in the completion marker is ok and every library file exists; the script's exit status is no longer hidden by a pipe (#5).
- devbattery.sh / devlock.sh: a missing local timeout or ideviceinfo is reported as such and stops at once, instead of being reported as an offline device and polled for up to an hour (#7).
- smoke_mac.sh builds its package for iPhone only regardless of the project default.

Reported by @SakuraToErii. The setup.sh part of #4 was also submitted by @hubeiqiao in #6.

Co-authored-by: hubeiqiao <12777592+hubeiqiao@users.noreply.github.com>
2026-10-13 03:17:23 +08:00

238 lines
25 KiB
Bash
Executable File
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/bin/bash
# Switch 路线验收入口:⓪ 启动块清单 ① 改写等价回归 ② 存档逐字节对拍 ③ 对真值的画面闸门 ④ 单测 ⑤ 补拍积压确定性 ⑥ 全档存档事务 ⑧ iOS 构建 ⑨ 功能用例 touchchain ⑩ 局内切运行模式(①–⑩ 并行);任一项红即整体红。
# r 版本切现行之前必须跑;「完成」以它为准。
# 用法:dm/verify/accept.sh [--fast|--full] [产物目录=out/<当前代号>-keypage];缺省全档(17 页 + 存档事务 4 例),快档只抽已批准 5 页,⓪①②④⑤不变。
# 输出:每项一行读数 + 末尾「期望 → 真结果」对位表 + 总用时;日志在 $DM_ROOT/run/verify/dm/accept/<时刻>/;退出码 0 = 全绿
# 按改动范围选项(先少干活)。改动 = 本树相对 main 分叉点的已提交改动 + 未提交改动的路径并集。映射表:
# 全部路径 ⊂ { dm/ios/**、docs/**、*.md } → 只跑 ⓪ ⑧(iOS 壳与文档进不了宿主:不跑回归 / 对拍 / 闸门 / 单测 / 功能用例 / 切模式)
# 全部路径 ⊂ { dm/gfx/**、dm/ios/**、docs/**、*.md } → 跑 ⓪ ③ ④ ⑧ ⑩(图形改动不动改写产物与存档逻辑:不跑 ① ② ②B/C ⑤ ⑥ ⑨)
# 其余(dm/host、dm/verify、dm/tools、dm/data、重命名、拿不准的任何路径)、没有改动(= 主干头复核)、ACCEPT_ALL=1 → 全量
# 被跳过的项在输出与对位表里逐项列「⏭ 跳过」,不静默。
# Acceptance entry for the Switch route: (0) boot block list (1) rewrite equivalence regression (2) byte-for-byte save comparison (3) screenshot gate against ground truth (4) unit tests (5) catch-up tick backlog determinism (6) full-mode save transactions (8) iOS build (9) functional cases touchchain (10) switching operation mode in game ((1)-(10) in parallel); any failing item fails the whole run.
# Must be run before an r version is made current; "done" is defined by it.
# Usage: dm/verify/accept.sh [--fast|--full] [build output directory=out/<current codename>-keypage]; default is full mode (17 pages + 4 save transaction cases), fast mode samples only the 5 approved pages, (0)(1)(2)(4)(5) unchanged.
# Output: one readout line per item + an "expected -> actual result" side-by-side table at the end + total time; logs in $DM_ROOT/run/verify/dm/accept/<timestamp>/; exit code 0 = all green
# Items are selected by change scope (do less work first). Change = the union of paths of this tree's committed changes relative to the fork point from main + uncommitted changes. Mapping table:
# all paths ⊂ { dm/ios/**, docs/**, *.md } -> run only (0) (8) (the iOS shell and docs cannot get into the host: no regression / comparison / gate check / unit tests / functional cases / mode switch)
# all paths ⊂ { dm/gfx/**, dm/ios/**, docs/**, *.md } -> run (0) (3) (4) (8) (10) (graphics changes do not touch the rewrite build output or save logic: no (1) (2) (2)B/C (5) (6) (9))
# everything else (dm/host, dm/verify, dm/tools, dm/data, renames, any path in doubt), no change (= recheck of the main head), ACCEPT_ALL=1 -> full set
# Skipped items are listed one by one as 「⏭ 跳过」(skipped) in the output and the side-by-side table, not silently.
set -u
export LC_ALL=C
HERE=$(cd "$(dirname "$0")" && pwd)
MODE=full
case "${1:-}" in --fast) MODE=fast; shift;; --full) shift;; --*) echo "ACCEPT RED 未知档位 $1"; exit 2;; esac
OUT=${1:-$(bash "$(dirname "$0")/../tools/product.sh" path keypage)}
TR=${TRUTH_RUN:-${DM_ROOT:-$HOME/dm-work}/data/switch/truth_runs/cvd5}
# 验收入口不得过期(在 dm-gfx 树跑的 accept 落后 main 26 个验证提交——旧仓库页按时长推杆、产物 r12,偶发红无意义):
# 本树缺 main 上任何改动 dm/verify 的提交即拒跑;超前(正在改验收脚本的分支)不拦
# The acceptance entry must not be stale (an accept run in the dm-gfx tree was 26 verification commits behind main -- the old stash page pushed the stick by duration, build output r12, and the sporadic failures were meaningless):
# if this tree lacks any commit on main that changes dm/verify, refuse to run; being ahead (a branch that is changing the acceptance scripts) is not blocked
STALE=$(git -C "$HERE" log --no-merges --oneline HEAD..main -- "$HERE" 2>/dev/null | wc -l | tr -d ' ')
[ "${STALE:-0}" = 0 ] || { echo "🔴 ACCEPT 拒跑:本树 dm/verify 落后 main $STALE 个验证提交(验收入口过期,读数无意义)——先合入 main 再跑"; git -C "$HERE" log --no-merges --oneline HEAD..main -- "$HERE" | head -5; exit 1; }
# 全局锁(同一时刻只跑一路 accept——多棵工作树共用本脚本,负载叠加会拖慢并诱发时序红;串行只降噪,负载下的时序红本身仍是缺陷要修)
# 去锁进行中:共享路径已按「时刻_树名_进程号」隔离;墙钟 / 负载类判据改完后再换成按工作树的锁
# Global lock (only one accept runs at any moment -- several worktrees share this script, and stacked load slows things down and induces timing failures; serializing only reduces noise, and a timing failure under load is itself still a defect to fix)
# Lock removal in progress: shared paths are already isolated by "timestamp_treename_pid"; once the wall clock / load criteria are changed it will be replaced by a per-worktree lock
TREE=$(basename "$(git -C "$HERE" rev-parse --show-toplevel 2>/dev/null || printf '%s' "$HERE")")
LOCK=${DM_ROOT:-$HOME/dm-work}/run/verify/dm/accept.lock; ME="${ACCEPT_OWNER:-$TREE} $(git -C "$HERE" rev-parse --abbrev-ref HEAD 2>/dev/null)"
while ! mkdir "$LOCK" 2>/dev/null; do
op=$(cat "$LOCK/pid" 2>/dev/null)
if [ -n "$op" ] && ! kill -0 "$op" 2>/dev/null; then echo "接管陈旧锁(pid $op 已不在:$(cat "$LOCK/owner" 2>/dev/null))"; rm -rf "$LOCK"; continue; fi
[ "${_waited:-}" ] || { echo "等待 $(cat "$LOCK/owner" 2>/dev/null)(pid ${op:-?})的 accept"; _waited=1; }
sleep 2
done
echo $$ > "$LOCK/pid"; echo "$ME" > "$LOCK/owner"; trap 'rm -rf "$LOCK"' EXIT
mkdir -p "${DM_ROOT:-$HOME/dm-work}/run/verify/dm/accept"
L=$(mktemp -d "${DM_ROOT:-$HOME/dm-work}/run/verify/dm/accept/$(date +%m%d_%H%M%S)_${TREE}_$$.XXXXXX") || { echo "ACCEPT RED 无法创建结果目录"; exit 1; }
T0=$SECONDS
SKIP=""; SCOPE="全量"
if [ "${ACCEPT_ALL:-0}" != 1 ]; then
ROOT=$(git -C "$HERE" rev-parse --show-toplevel 2>/dev/null); MB=$(git -C "$HERE" merge-base HEAD main 2>/dev/null)
CH=""; [ -n "$ROOT" ] && [ -n "$MB" ] && CH=$( { git -C "$ROOT" diff --name-only --no-renames "$MB" HEAD; git -C "$ROOT" status --porcelain --no-renames | cut -c4-; } 2>/dev/null | sort -u | grep -v '^$') # --no-renames:改名按「删旧 + 加新」两条路径列,旧路径在别的目录时不会被漏掉 | --no-renames: a rename is listed as two paths, "delete old + add new", so the old path is not missed when it is in another directory
[ -n "${ACCEPT_SCOPE_CH:-}" ] && CH=$(printf '%s\n' "$ACCEPT_SCOPE_CH" | tr ' ' '\n') # 干跑范围判定用:直接给改动路径清单(空格或换行分隔),配 ACCEPT_SCOPE_ONLY=1 只打印 SCOPE 行就退出 | for dry-run scope decisions: give the list of changed paths directly (space or newline separated), and with ACCEPT_SCOPE_ONLY=1 only print the SCOPE line and exit
# dm/ios 下被编进 Mac 宿主(含宿主单测、触屏功能用例)的源文件,改了等于改宿主:清单从两份 Makefile 里现取,不手写
# (dm_touch_controls.c 的改动被判成「只在 dm/ios」跳过了 ⑨,把主干的 skill_main 用例弄红)
# Source files under dm/ios that are compiled into the Mac host (including host unit tests and touch-screen functional cases): changing them equals changing the host; the list is taken live from the two Makefiles, not written by hand
# (a change to dm_touch_controls.c was judged "only in dm/ios" and skipped (9), turning the skill_main case on main red)
HOSTDEP=$(grep -hoE 'ios/Sources/[A-Za-z0-9_]+\.(\[ch\]|[chm])' "$HERE/../host/Makefile" "$HERE/../host/tests/Makefile" 2>/dev/null | sed -E 's/^(.*)\.\[ch\]$/\1.c\n\1.h/' | sed 's#^#dm/#' | sort -u)
CH=$(printf '%s\n' "$CH" | while IFS= read -r f; do [ -n "$f" ] || continue; if printf '%s\n' "$HOSTDEP" | grep -qxF "$f"; then echo "dm/host/(宿主也编) $f"; else echo "$f"; fi; done)
if [ -n "$CH" ]; then
if ! printf '%s\n' "$CH" | grep -qvE '^(dm/ios/|docs/)|\.md$'; then SKIP="1 2 3 4 5 6 7 9 10"; SCOPE="改动只在 dm/ios 与文档"
elif ! printf '%s\n' "$CH" | grep -qvE '^(dm/gfx/|dm/ios/|docs/)|\.md$'; then SKIP="1 2 5 6 7 9"; SCOPE="改动只在 dm/gfx(及 dm/ios、文档)"; fi
fi
fi
echo "SCOPE ${SCOPE}$([ -n "$SKIP" ] && echo ";跳过项序号 ${SKIP}(ACCEPT_ALL=1 强制全量)")"
[ "${ACCEPT_SCOPE_ONLY:-0}" = 1 ] && exit 0
# 并行子项都调用增量make;先在锁内只构建一次,避免冷树多个链接器争写同一个dmhost/dSYM。
# The parallel sub-items all invoke an incremental make; build once first inside the lock, so that on a cold tree several linkers do not compete to write the same dmhost/dSYM.
if make -s -C "$HERE/../host" > "$L/build.log" 2>&1; then
echo "BUILD OK 宿主预构建完成,后续子项只做增量核对"
else
echo "宿主预构建失败;运行子项未启动,现场保留" > "$L/KEEP"
cat "$L/build.log"
echo "ACCEPT RED 宿主预构建失败 日志 $L"
exit 1
fi
item() { # item <序号> <名> <期望> <日志> <读数行正则> <读数行数> <命令…>:结果写 $L/<序号>.row(首行 rc,次行表格行) | item <number> <name> <expected> <log> <readout line regex> <readout line count> <command...>: the result is written to $L/<number>.row (first line rc, second line the table row)
local k=$1 name=$2 want=$3 log=$4 kl=$5 kn=$6; shift 6; local s=$SECONDS
case " $SKIP " in *" $k "*) printf '0\n| %s | %s | ⏭ 跳过:%s | 0 秒 |\n' "$name" "$want" "$SCOPE" > "$L/$k.row"; echo "⏭ ${name}(跳过:${SCOPE})"; return 0;; esac
"$@" > "$L/$log" 2>&1; local rc=$?
local got=$(grep -E "$kl" "$L/$log" | tail -$kn | tr '\n' ' ') mark
[ $rc = 0 ] && mark="✅" || mark="🔴"
printf '%s\n| %s | %s | %s %s | %s 秒 |\n' "$rc" "$name" "$want" "$mark" "${got:-(无读数行,见 ${log})}" "$((SECONDS - s))" > "$L/$k.row"
echo "$mark ${name}($((SECONDS - s)) 秒)"
}
item 0 "⓪ 启动块清单 bootlist.tsv" "存在、构建号与 main.cuts.tsv 一致、与仓库 dm/data/bootlist_<构建号>.tsv 逐字节相同、sha 自洽、块数 > 0、逐块与资源包一致" bootlist.log '^BOOTLIST' 1 "$HERE/bootchk.sh" "$OUT"
# ①②③ 互相独立,并行(accept ≤ 2 分钟):regress 吃 CPU、host_savecmp 一局宿主、shotgate 4 路宿主
# (1)(2)(3) are independent of each other and run in parallel (accept <= 2 minutes): regress eats CPU, host_savecmp is one host run, shotgate is 4 hosts in parallel
regress_both() { # ① 被测产物全套 + 出包产物(iOS 发给玩家的 touch)结构核对(触屏私有门 4 条登记缺口,accept 只核 keypage 一直没暴露) | (1) structure check of the full set of build output under test + the packaging build output (the touch one iOS ships to players) (4 registration gaps in the touch-screen private gates went unexposed because accept only checked keypage)
local rc=0 touch_out=${DM_TOUCH_OUT:-$(bash "$(dirname "$0")/../tools/product.sh" path touch)}
"$HERE/regress.sh" "$OUT" || rc=1
"$HERE/regress.sh" --struct-only "$touch_out" || rc=1
return $rc
}
item 1 "① 改写等价回归 regress(+ 出包产物 touch 结构核对)" "被测产物三模块 STRUCT/EQUIV OK、GATEFN 全过;touch 三模块 STRUCT OK、GATEFN 全过;两行 REGRESS OK" regress.log '^(STRUCT|EQUIV|REGRESS|GATEFN)' 13 regress_both &
savecmp_gate() { # host_savecmp + 宿主日志红线(只比存档字节、没看日志,缺着色器 20 行【红】连报 5 次绿);$1 = 本项日志名 | host_savecmp + host log a failure line line (it only compared save bytes and did not look at the log, so a missing shader with 20 lines of 【红】(red) was reported green 5 times in a row); $1 = this item's log name
bash "$HERE/../tools/host_savecmp.sh" "$TR" "$OUT"; local rc=$?
local hd=$(sed -n 's/^局目录 //p' "$L/${1:-savecmp.log}" | head -1) r
r=$("$HERE/logred.sh" "$hd/host.log"); local lrc=$?
echo "HOSTLOG 宿主日志红线 $(echo "$r" | cut -f1) 行 $(echo "$r" | cut -f2)($hd/host.log)"
[ $lrc = 0 ] || { [ $lrc = 2 ] && echo "HOSTLOG 找不到宿主日志,判红"; rc=1; }
# 预取命中只报不判:预取是性能优化,命中数取决于最低 QoS 预取线程跑了多久,负载下实测 5270→3244、2148,
# 截止点无论取「到选角」还是整局都和墙钟赛跑;清单对不对是静态性质,归 ⓪ bootchk(逐块核资源包)
# Prefetch hits are only reported, not judged: prefetch is a performance optimization, and the hit count depends on how long the lowest-QoS prefetch thread got to run, measured under load 5270->3244, 2148,
# so whether the cutoff is taken "at character select" or over the whole run it is a race against the wall clock; whether the list is correct is a static property and belongs to (0) bootchk (checks the resource package block by block)
local hit=$(grep -o 'DM_TACT_PREFETCH:命中 [0-9]*' "$hd/host.log" | tail -1 | grep -o '[0-9]*$')
echo "PREFETCH 预取命中 ${hit:-无读数} 块(只报不判)"
return $rc
}
item 2 "② 存档逐字节对拍 host_savecmp $(basename "$TR")" "逐拍 6 格一致、正对照一致、负对照不一致、宿主日志无红线" savecmp.log 'SELFCMP|^SAVECMP|结论|^HOSTLOG|^PREFETCH' 6 savecmp_gate savecmp.log &
gatebc() {
local rc=0 rule_out=${DM_RULE_OUT:-$(bash "$(dirname "$0")/../tools/product.sh" path rules)} self_out=${DM_LOGICSELF_OUT:-$(bash "$(dirname "$0")/../tools/product.sh" path touch)}
if [ ! -f "$HERE/rulegate.sh" ] || [ ! -f "$HERE/logicself.sh" ]; then
echo "RULEGATE TEST FIXTURE 未装载脚本"
echo "LOGICSELF TEST FIXTURE 未装载脚本"
return 0
fi
bash "$HERE/rulegate.sh" 400 "$rule_out" || rc=1
bash "$HERE/logicself.sh" "$self_out" 230 || rc=1
return $rc
}
item 7 "②B/C 逻辑闸门 rulegate + logicself" "RULEGATE 400 逐拍六组一致;LOGICSELF 230 逐拍自身基线一致" logic_gate.log '^(RULEGATE|LOGICSELF)' 8 gatebc &
# ⑤ 确定性回归:拍 400 入口人为睡 300 ms 造补拍积压,冻结拍仍须落在 401、存档正绿负红
# (5) Determinism regression: sleep 300 ms artificially at the entry of tick 400 to build a catch-up tick backlog; the frozen tick must still land on 401, and the save must pass in the positive case and fail in the negative case
lag_gate() { # ⑤:savecmp_gate + 射程(证明不是空跑——宿主日志「下一拍时限 A → B」的增量须 ≥ 200;注入时实测 281、不注入 28) | (5): savecmp_gate + coverage (proves it is not an empty run -- the increment of the host log 「下一拍时限 A → B」(next tick deadline A -> B) must be >= 200; measured 281 with injection, 28 without)
savecmp_gate savecmp_lag.log; local rc=$?
local hd=$(sed -n 's/^局目录 //p' "$L/savecmp_lag.log" | head -1)
local inc=$(grep -o '下一拍时限 [0-9]* → [0-9]*' "$hd/host.log" | tail -1 | awk '{print $4 - $2}')
if [ -n "$inc" ] && [ "$inc" -ge 200 ]; then echo "LAGRANGE 积压已造出:下一拍时限增量 ${inc}(≥ 200)"; else echo "LAGRANGE 空跑:下一拍时限增量 ${inc:-无读数}(须 ≥ 200)"; rc=1; fi
return $rc
}
item 5 "⑤ 补拍积压下冻结拍确定性 DM_TEST_LAG=400 $(basename "$TR")" "冻结拍 401(目标 401)不带【红】、正对照一致、负对照不一致、宿主日志无红线、积压射程增量 ≥ 200" savecmp_lag.log 'SELFCMP|^SAVECMP|结论|^HOSTLOG|^LAGRANGE|^PREFETCH' 7 env DM_TEST_LAG=400 bash -c "$(declare -f savecmp_gate lag_gate); HERE=$HERE TR=$TR OUT=$OUT L=$L lag_gate" &
func_cases() {
local case_dir name total green=0 status elapsed note log detail reported
local cases=() pending=()
while IFS= read -r -d '' case_dir; do
[ -f "$case_dir/script.txt" ] && [ -f "$case_dir/assert.py" ] || continue
# 用例目录里有 pending 文件 = 待接入(如等宿主改动才能进时限):不跑,但必须列出,不静默跳过;首行写原因
# A pending file in a case directory = waiting to be wired in (e.g. waiting for a host change before it can fit the time limit): not run, but must be listed, not silently skipped; the first line gives the reason
if [ -f "$case_dir/pending" ]; then pending+=("$(basename "$case_dir"):$(head -1 "$case_dir/pending")"); continue; fi
cases+=("$(basename "$case_dir")")
done < <(find "$HERE/touchcases" -mindepth 1 -maxdepth 1 -type d -print0 | sort -z)
total=${#cases[@]}
if [ "$total" -eq 0 ]; then
echo 'FUNC 0/0'
return 1
fi
# 令牌池限并发(FUNC_JOBS,缺省 2;写法同 shotgate):8 例同时起时每例的启动加载被拖长 4–5 倍(mini 实测单跑 2.8 s 撤读取画面、
# 8 路并行 11–14 s),时限 20 s 卡的是并行启动而不是用例本身;快进后每例单跑 4–5 s,2 个一批、本项总墙钟约 30 s;mini(12 核)上背景 6 宿主 + 3 个一批仍有 15–22 s,故取 2
# 功能用例跑在 iOS 包同一份界面覆盖层上:起跑前生成一次(幂等,约 3 秒),各例共用;不依赖 ⑧ iOS 构建的先后
# A token pool limits concurrency (FUNC_JOBS, default 2; written the same way as shotgate): with 8 cases started at once, each case's startup loading is stretched 4–5 times (measured on the second Mac: 2.8 s to take down the loading screen when run alone,
# 11–14 s with 8 in parallel), so the 20 s limit was catching the parallel startup rather than the case itself; after fast-forward each case takes 4–5 s alone, and with batches of 2 the total wall clock of this item is about 30 s; on the second Mac (12 cores) with 6 background hosts + batches of 3 it is still 15–22 s, hence 2
# Functional cases run on the same UI overlay as the iOS app: generated once before starting (idempotent, about 3 seconds) and shared by all cases; does not depend on the ordering of (8) iOS build
python3 "$HERE/../ios/tools/gen_romfs_overlay.py" "${DM_ROMFS_SRC:-${DM_ROOT:-$HOME/dm-work}/data/switch/3.3.94012/Update/RomFS}" "$L/romfs_overlay" > /dev/null 2>"$L/func_overlay.err" || { echo "FUNCFAIL 界面覆盖层生成失败:$(tail -1 "$L/func_overlay.err")"; echo 'FUNC 0/0'; return 1; }
export DM_ROMFS_OVERLAY="$L/romfs_overlay"
local sem="$L/func.sem.$$"; mkfifo "$sem" || return 1; exec 8<>"$sem"; rm -f "$sem"
for _ in $(seq 1 "${FUNC_JOBS:-2}"); do echo >&8; done
for name in "${cases[@]}"; do
read -r -u 8
(
log="$L/func_${name}.log"
local started=$SECONDS rc=0
timeout 120s env DM_TOUCH_RUNROOT="$L/touchchain/$name" bash "$HERE/touchchain.sh" "$name" > "$log" 2>&1 || rc=$?
reported=$(awk '/^TOUCHCHAIN / {print $(NF - 1)}' "$log" | tail -1)
elapsed=${reported:-$((SECONDS - started))}
status=RED
[ "$rc" -eq 0 ] && [ "$elapsed" -le "${FUNC_LIMIT_S:-60}" ] && status=OK # 判对错只看断言;墙钟只作卡死兜底(60 s)。20 s 硬线在多线并行 accept 时反复误红,用例单跑 4–5 s | right or wrong is judged by assertions only; wall clock is only a backstop for hangs (60 s). The hard 20 s line repeatedly failed spuriously when several accept runs ran in parallel; a case run alone takes 4–5 s
printf '%s %s\n' "$status" "$elapsed" > "$L/func_${name}.row"
echo >&8
) &
done
wait; exec 8>&-
for name in "${cases[@]}"; do
read -r status elapsed < "$L/func_${name}.row"
note=''
[ "$elapsed" -gt 20 ] && note='(慢:> 20 s,本机负载高?)'; [ "$elapsed" -gt "${FUNC_LIMIT_S:-60}" ] && note="(超卡死兜底 ${FUNC_LIMIT_S:-60} s)"
printf 'FUNCCASE %s %s %s 秒%s\n' "$name" "$status" "$elapsed" "$note"
if [ "$status" = OK ]; then
green=$((green + 1))
else
log="$L/func_${name}.log"
detail=$(grep -E '局目录:|断言失败:' "$log" | tail -2 | tr '\n' ' ')
[ -n "$detail" ] || detail="日志:$log"
printf 'FUNCFAIL %s %s\n' "$name" "$detail"
fi
done
printf 'FUNC %s/%s\n' "$green" "$total"
[ ${#pending[@]} -eq 0 ] || { printf 'FUNCPENDING 待接入 %s 例:' "${#pending[@]}"; printf '%s;' "${pending[@]}"; echo; }
[ "$green" -eq "$total" ]
}
item 9 "⑨ 功能用例 touchchain" "自动收集 script.txt + assert.py 用例;各例并行;断言通过(墙钟 60 s 只作卡死兜底)" func.log '^(FUNCCASE|FUNC|FUNCFAIL|FUNCPENDING)' 1000 func_cases &
PAGES=()
[ "$MODE" = fast ] && PAGES=(camp_e stash wild charselect skilltree)
PAGE_NAMES=${PAGES[*]:-$(cd "$HERE/shotgate_pages" && ls *.json | sed 's/\.json$//' | tr '\n' ' ')}
item 3 "③ 画面闸门 shotgate ${MODE}(${PAGE_NAMES})" "每页对真值绿(结构 / 非黑 / 网格 / 命名区域 / 应有元素 / 宿主日志无【红】)" shotgate.log 'SHOTGATE_ALL' 1 "$HERE/shotgate.sh" "$OUT" ${PAGES[@]+"${PAGES[@]}"} &
unit_tests() { # ④ 宿主与图形单测(E:SHOT 合入后 gfx 单测链接失败,accept 却绿):任何失败或构建 / 链接错误即红 | (4) host and graphics unit tests (after E:SHOT was merged the gfx unit test failed to link, yet accept was green): any failure or build / link error is a failure
local rc=0
"${DM_ROOT:-$HOME/dm-work}/build/cache/dm/venv/bin/python" "$HERE/ryu/test_selfcmp.py" > "$L/unit_selfcmp.log" 2>&1 && echo "UNIT selfcmp 通过" || { echo "UNIT selfcmp 失败(见 unit_selfcmp.log)"; rc=1; }
"${DM_ROOT:-$HOME/dm-work}/build/cache/dm/venv/bin/python" "$HERE/ryu/test_plan.py" > "$L/unit_plan.log" 2>&1 && echo "UNIT plan 通过" || { echo "UNIT plan 失败(见 unit_plan.log)"; rc=1; }
make -j8 -C "$HERE/../host/tests" run > "$L/unit_host.log" 2>&1 && echo "UNIT host 通过" || { echo "UNIT host 失败:$(grep -m1 -iE 'error|fail|失败|红' "$L/unit_host.log" | cut -c1-140)"; rc=1; }
make -j8 -C "$HERE/../gfx/tests" run > "$L/unit_gfx.log" 2>&1 && echo "UNIT gfx 通过" || { echo "UNIT gfx 失败:$(grep -m1 -iE 'error|fail|失败|红|undefined' "$L/unit_gfx.log" | cut -c1-140)"; rc=1; }
# iOS 触摸层单测(这一套从没接进闸门,回城钮改版后红了一天没人看见)
# iOS touch layer unit tests (this set was never wired into the gate check, and after the town portal button redesign it was red for a day without anyone seeing it)
make -C "$HERE/../ios/tests" test > "$L/unit_ios.log" 2>&1 && echo "UNIT ios 通过" || { echo "UNIT ios 失败:$(grep -m1 -E 'FAIL|Error|error' "$L/unit_ios.log" | cut -c1-140)"; rc=1; }
# accept 自己的接线单测(加 ⑧ 时漏过替身,test_accept 4 例红而 accept 全绿);它用替身起 accept,不跑真宿主,几秒
# accept's own wiring unit test (when (8) was added a stand-in was missed, test_accept had 4 failure cases while accept was all green); it starts accept with stand-ins, does not run the real host, and takes a few seconds
(cd "$HERE" && python3 -m unittest -q test_accept) > "$L/unit_accept.log" 2>&1 && echo "UNIT accept 通过" || { echo "UNIT accept 失败:$(grep -m1 -E '^(FAIL|ERROR):' "$L/unit_accept.log" | cut -c1-140)"; rc=1; }
return $rc
}
item 4 "④ 单测 selfcmp + plan + make -C dm/host/tests run + dm/gfx/tests run + dm/ios/tests test + test_accept" "六套全过、无构建 / 链接错误" unit.log '^UNIT' 6 unit_tests &
ios_build_gate() {
local s=$SECONDS raw="$L/iosbuild-command.log"
DM_NOINSTALL=1 DM_IOS_DD="$HERE/../../.iosdd" bash "$HERE/../ios/run.sh" build > "$raw" 2>&1
local rc=$? elapsed=$((SECONDS - s))
if [ "$rc" = 0 ]; then
echo "IOSBUILD OK 用时 $elapsed 秒"
else
echo "IOSBUILD RED 用时 $elapsed 秒"
grep -m1 -E 'error:' "$raw" || true
fi
return "$rc"
}
item 8 "⑧ iOS 构建(只编不装)" "DM_NOINSTALL=1 工作树派生目录构建成功;首个 error 可见" iosbuild.log '^IOSBUILD (OK|RED)' 2 ios_build_gate &
# ⑩ 局内切运行模式(真机过热切掌机即崩、各闸门都不切模式没抓到):进局后 主机 → 掌机,进程须活着继续出帧
# (10) Switching operation mode in game (a real device overheating and switching to handheld crashed immediately, and no gate check switched modes so it was not caught): after joining a run, docked -> handheld, and the process must stay alive and keep producing frames
item 10 "⑩ 局内切运行模式 opmode" "切换前已进局;取走通知 30/31;活到 timeout 无 CRASH;切换后 ≥ 300 帧;宿主日志无红线" opmode.log '^OPMODE (OK|RED)' 1 bash "$HERE/opmode.sh" "$OUT" &
# 事务四局串行本身占约 90 秒,列入全档;与其余项目独立,原来的六项判据保持不变。
# The four transaction runs in series take about 90 seconds by themselves and are put in the full mode; they are independent of the other items, and the original six criteria stay unchanged.
[ "$MODE" = full ] && item 6 "⑥ 存档事务 save_txn_regress" "4/4 注入命中;已提交文件同期望版;校验和自洽;残缺工作目录 / 拒绝日志负对照成立" save_txn.log '^✓|^✗|^结论|^SAVE_TXN' 6 bash "$HERE/../tools/save_txn_regress.sh" "$OUT" &
wait
allrc=0; rows=()
ITEMS=(0 1 2 3 4 5 7 8 9 10); [ "$MODE" = full ] && ITEMS+=(6)
for k in "${ITEMS[@]}"; do { read -r rc; read -r row; } < "$L/$k.row"; [ "$rc" = 0 ] || allrc=1; rows+=("$row"); done
[ "$allrc" = 0 ] || echo "ACCEPT $MODE 红;本目录保留,不删不覆盖" > "$L/KEEP"
echo; echo "| 项 | 期望 | 真结果 | 用时 |"; echo "|---|---|---|---|"; printf '%s\n' "${rows[@]}"
echo; echo "ACCEPT $([ $allrc = 0 ] && echo OK || echo RED) 档位 $MODE 产物 $(basename "$OUT") 总用时 $((SECONDS - T0)) 秒 日志 $L"
exit $allrc