Vendored Go undo (vendor --revert, rollback, remove) stays silent about a stale vendor/modules.txt, so every build fails with "inconsistent vendoring"
维护者通常 1 天内回复
已经有一个关联 PR 被合并了。
- #1778 来自 @mikolalysenko —— 已合并
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 30/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 领域
- build-system, cli
调研方向
Start in crates/socket-patch-cli/src/commands/vendor.rs at run_revert, compare its forward warning path with go_consumer_sync::audit_warnings, then trace the vendored branches in rollback.rs and remove.rs. Read the apply --check / vendor --check handling to see why a stale vendor/modules.txt is not detected once no replacement entry remains. Done means all three undo paths emit go_vendor_modules_txt_out_of_sync and a later check fails until go mod vendor/go work vendor is run; check open PR #1778 before duplicating work.
由索引模型根据 Issue 内容生成。
描述
[agent] Found by the scheduled Go modules bug-hunt routine (ledger #317).
Summary
Take a Go project with a committed vendor/ directory that was patched in vendored mode, with go mod vendor re-run as the forward warning asks. Undoing the patch with vendor --revert, rollback or remove <purl> drops the go.mod replace and exits 0 with status success. None of them warns go_vendor_modules_txt_out_of_sync, and vendor --check and apply --check both exit 0 afterwards. But vendor/modules.txt still records the removed replace, so the next default build fails:
go: inconsistent vendoring in …/c:
example.com/[email protected]: is marked as replaced in vendor/modules.txt, but not replaced in go.mod
Agent mode handles the same situation correctly. After apply → go mod vendor → rollback, the CLI warns go_vendor_modules_txt_out_of_sync and apply --check exits 1 with "still records a socket-patch replacement … run go mod vendor". This is the #343 fix. Only the vendored undo paths miss it.
Impact
Repos that commit vendor/ are the main audience for vendored mode (air-gapped and frozen CI). There, every vendored undo leaves the default build broken, and nothing tells the user to run go mod vendor. This includes the cleanup a user does when moving away from vendored mode. It's a rewrite that makes the next frozen build fail, while the CLI reports success and its own audits pass.
Expected vs actual
docs/ecosystems.md, "Files go derives from go.mod": apply, vendor, rollback and scan --mode hosted warn with the command that does the regeneration, and apply --check / vendor --check report the disagreement as drift until it is run. The warning covers a vendor/modules.txt that "does not record a socket replace, or still records one that was removed".
- Expected:
vendor --revert,rollbackandremovewarngo_vendor_modules_txt_out_of_syncwithgo mod vendor(orgo work vendor), as agentrollbackdoes, and a later check fails until the regeneration is done. - Actual: no warning (rollback only reports
reinstall_required), exit 0, and both checks exit 0.
Repro
This uses a hermetic HTTP GOPROXY and a local mock of the patch API. POST /v0/orgs/<org>/patches/package returns a granted tarball: an example.com/[email protected]/ module zip with a patched lib.go, plus sha512 integrity, in the vendor/golang.rs mount_go_granted shape. GET …/patches/view/<uuid> returns the record.
export GOFLAGS= GOENV=off GOTOOLCHAIN=local GOPROXY=http://127.0.0.1:$PORT/proxy GOSUMDB=sum.invalid.example
API="--api-url http://127.0.0.1:$PORT --org test-org --api-token fake"
# consumer: go 1.21, require example.com/upstream v1.0.0, go.sum has the upstream lines
go mod vendor && git add -A && git commit -qm vendored-dir
socket-patch get $UUID --mode vendored --yes $API # exit 0, warns go_vendor_modules_txt_out_of_sync (correct)
go mod vendor && go run . # OUT: PATCHED
git add -A && git commit -qm patched
socket-patch vendor --revert --json # exit 0, status success, warnings: []
socket-patch vendor --check # exit 0
socket-patch apply --check --offline # exit 0
go run . # exit 1: inconsistent vendoring (above)
# Same result from `git checkout . && git clean -fd` and then
# socket-patch rollback --offline --json -> exit 0, warnings: [reinstall_required]
# socket-patch remove pkg:golang/example.com/[email protected] --offline --json -> exit 0, warnings: []
# Control, agent mode on the same project shape:
# apply --offline; go mod vendor; rollback --offline --json -> warnings: [go_vendor_modules_txt_out_of_sync]
# apply --check -> exit 1 "still records a socket-patch replacement ..."
The same happens in a go.work workspace after go work vendor, both with use . at the root and in a non-root member (use ./a). The non-root member case overlaps #1766, which is agent mode. Its fix, #1767, only changes the audit in go_consumer_sync.rs, and vendored undo never calls that audit.
OS × version
| OS | go | vendor --revert | rollback | remove |
|---|---|---|---|---|
| Linux | 1.22.12 | silent, build broken | silent, build broken | silent, build broken |
| Linux | 1.24.7 (3 runs) | silent, build broken | silent, build broken | silent, build broken |
| Linux | 1.26.8 | silent, build broken | silent, build broken | silent, build broken |
| macOS / Windows | — | untested (the code path is OS-independent) |
It isn't a regression: the v5.0.1 release (npm @socketsecurity/[email protected]) behaves the same. Tested on main b787454.
Suspect code
crates/socket-patch-cli/src/commands/vendor.rs:4417(run_revert) never callsgo_consumer_sync::audit_warnings. The forward path does, atvendor.rs:4013-4026(#343, #618).crates/socket-patch-cli/src/commands/rollback.rs:3319-3329only audits when a resultis_local_go(the agent go-patches leg), so the vendored leg is skipped.crates/socket-patch-cli/src/commands/remove.rshas no Go consumer audit at all.apply --check/vendor --checknever look at a stale modules.txt when no vendored or manifest entry is left, so they pass after the revert.
No probe runs: the sandbox is Linux-only, and the go/modules.txt logic involved is OS-independent.
- 主要语言
- Rust
- 星标
- 9
- 派生
- 2
- 平均合并
- 20 小时 47 分钟
- 30 天内合并 PR
- 681
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
SocketDev/socket-patch 的其他 Issue
-
agent:triaged arch-audit priority:p3 refactor
难度 2/5 1-3 小时 新手友好度 83/100
SocketDev/socket-patch#1860 ·
维护者通常 1 天内回复
-
agent:triaged arch-audit priority:p3 refactor
难度 2/5 1-3 小时 新手友好度 78/100
SocketDev/socket-patch#1838 ·
维护者通常 1 天内回复
-
arch-audit refactor
难度 4/5 3-5 天 新手友好度 54/100
SocketDev/socket-patch#1993 ·
维护者通常 1 天内回复
-
arch-audit refactor
难度 3/5 1-2 天 新手友好度 66/100
SocketDev/socket-patch#1992 ·
维护者通常 1 天内回复
-
agent:triaged arch-audit priority:p3 refactor
难度 4/5 3-5 天 新手友好度 35/100
SocketDev/socket-patch#1987 · 1 条评论 ·
维护者通常 1 天内回复
查看 SocketDev/socket-patch 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 66/100
riverqueue/river#1576 ·
维护者通常 1 天内回复
-
bug
难度 2/5 1-3 小时 新手友好度 88/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 74/100
stadelmanma/fitparse-rs#64 ·