Compare commits
2 Commits
site-v1.1.1
...
main
| Author | SHA1 | Date | |
|---|---|---|---|
| 79830cb43f | |||
| fdc13ea06c |
@@ -90,25 +90,16 @@ jobs:
|
||||
name: windows
|
||||
path: dist/
|
||||
|
||||
# Why build-macos/build-ios don't block the working android+windows pipeline
|
||||
# when Apple secrets aren't configured yet (they aren't, as of this writing):
|
||||
# 1. release-deploy's `needs:` below is [build-android, build-windows]
|
||||
# ONLY — macOS/iOS are NOT dependencies, so release-deploy never waits
|
||||
# on them and never fails because of them.
|
||||
# 2. `continue-on-error: true` on both jobs keeps the overall workflow-run
|
||||
# status green even while compile-macos.sh hard-fails (Apple Developer
|
||||
# ID / notary secrets absent — see its fail-fast checks) — that failure
|
||||
# is real signal ("go configure the secrets"), but it shouldn't read as
|
||||
# "the release pipeline is broken" when android+windows shipped fine.
|
||||
# 3. compile-macos.sh's own default behavior is to hard-fail (not skip)
|
||||
# when its secrets are missing (macOS distribution must never ship
|
||||
# unsigned/unnotarized — see its header comment); compile-ios.sh's
|
||||
# default is to skip gracefully (exit 0) since an unconfigured iOS
|
||||
# account is a normal "not set up yet" state, not a defect. Either way
|
||||
# the job produces no dist/pangolin-macos-x64.zip, and
|
||||
# release-deploy's "Download all artifacts" step (no `name:` filter)
|
||||
# simply picks up whatever artifacts DO exist — an absent "macos"
|
||||
# artifact is not an error there.
|
||||
# build-macos/build-ios 用 `continue-on-error: true` → 即便 Apple 侧构建抖了
|
||||
# (frps 拉 sing-box 源码 / 公证超时等),也不阻塞 android+windows 发布:
|
||||
# 1. release-deploy 现在 `needs:` 全部四个平台(见下),所以它会**等**全部
|
||||
# build 完成再发布 —— 避免旧 `needs: [build-android]` 只等 android、
|
||||
# macos/ios 还没传产物就发布导致漏平台的竞态。
|
||||
# 2. continue-on-error 让 macos/ios 失败仍算「completed」满足 needs,不使整个
|
||||
# workflow 变红、不阻断已成功平台的发布;release-deploy 的「Download all
|
||||
# artifacts」(无 name 过滤)只捡实际存在的产物,缺某平台不报错。
|
||||
# 3. Apple 签名 secret 现已在 gitea 用户级配齐(DEVELOPER_ID_P12/IOS_DIST_P12/
|
||||
# APPSTORE_API_* 等),macos/ios 正常应成功;continue-on-error 只是抖动兜底。
|
||||
build-macos:
|
||||
runs-on: mac
|
||||
continue-on-error: true
|
||||
@@ -172,13 +163,24 @@ jobs:
|
||||
# No artifact upload — compile-ios.sh uploads straight to TestFlight via
|
||||
# altool (matches jiu); nothing is produced under dist/ for this job.
|
||||
|
||||
# release-deploy 只 needs build-android(唯一稳定可用的平台 floor)。windows/macos/ios
|
||||
# best-effort:各自 runner+secret 就绪则上传 artifact,release-deploy flatten 收
|
||||
# dist-raw/ 里"当时存在"的产物。windows 机离线 / Apple secret 未配 都不阻塞发版
|
||||
# (对应平台下载保留 pangolin1 上一版,待可用时下个 client-v* 追上)。
|
||||
# release-deploy 发布策略:needs 全部四平台(等它们**完成**,不抢跑),但用
|
||||
# `if` 让 mac/ios/windows 的失败不 skip 本 job —— 只要 android(稳定 floor)成功即发。
|
||||
# 背景:build-macos/ios 必须跑在 mac runner,其 Checkout(git fetch git.51yanmei.com)
|
||||
# 走 frps 隧道偶发超时挂掉(run 239 的 ios 就是 fetch fdc13ea 超时,非 iOS 构建问题;
|
||||
# run 238 同代码 fetch 碰巧成功)。这类瞬断不该阻断已成功平台的发布。
|
||||
# · needs 四平台 → download-artifact 收到当时**全部已上传**的产物(修旧
|
||||
# `needs:[build-android]` 只等 android、macos/ios 没传就发布漏平台的竞态)。
|
||||
# · continue-on-error 在 gitea 里**不**让下游 needs 在失败时继续(会 skip),故必须
|
||||
# 配 `if: always() && needs.build-android.result=='success'` 才能「等全部、失败不阻」。
|
||||
release-deploy:
|
||||
needs: [build-android]
|
||||
runs-on: mac
|
||||
needs: [build-android, build-windows, build-macos, build-ios]
|
||||
if: ${{ always() && needs.build-android.result == 'success' }}
|
||||
# ubuntu-latest = 家里 nas act_runner,与 gitea 同机/同网:Release → Forgejo 的
|
||||
# API 调用走本地不过 frps 隧道(mac runner 走 git.51yanmei.com→frps→抖,curl 超时
|
||||
# exit 28,3m3s=release_ensure 60s×3 重试耗尽)。deploy-server/deploy-site 同为
|
||||
# ubuntu-latest,其 Forgejo release 步稳定通过(server 仅 13s)。此 job 全为网络/
|
||||
# SSH 步骤(下载产物/传 Forgejo/SSH pangolin1/通知),无 mac 专属需求。
|
||||
runs-on: ubuntu-latest
|
||||
steps:
|
||||
- name: Checkout
|
||||
uses: actions/checkout@v4
|
||||
|
||||
Reference in New Issue
Block a user