Agent-mode Go rollback and remove after `go mod tidy` drop the replace but not restore the module's go.sum lines, so the next default `go build` fails with "missing go.sum entry" while rollback exits 0 with no warning
維護者通常 1 天內回覆
還沒有人認領這個 Issue。
評估
研究方向
Start in crates/socket-patch-core/src/patch/redirect/golang_local.rs at remove_go_redirect and follow go_mod_edit::drop_replace_entry in crates/socket-patch-core/src/vendor/go_mod_edit.rs. Compare the hosted-takeover handling in crates/socket-patch-core/src/vendor/golang.rs:364-377, then reproduce apply → go mod tidy → rollback/remove with the hermetic setup. Done means rollback and remove restore the module's go.sum entries or clearly report the required follow-up, and the default go build succeeds.
由索引模型根據 Issue 內容生成。
描述
[agent] Found by the scheduled Go modules bug-hunt routine (ledger #317).
Summary
In agent mode, apply points the patched module at .socket/go-patches/<module>@<version>/ with a directory replace. Go doesn't need go.sum lines for a directory replace, so go mod tidy removes the original module's h1: lines from go.sum. docs/ecosystems.md says "the wiring survives go mod tidy", so running tidy after apply is a supported workflow.
rollback (and remove <purl>) then drops the replace and the copy but leaves go.sum as tidy left it. The result is a go.mod that requires the module with no go.sum entry. The next go build with default flags (-mod=readonly) fails with missing go.sum entry for module providing package …. Rollback still reports status: success with warnings: [].
Impact
After a routine socket-patch rollback or remove, every CI build and every fresh go build/go test is red until someone runs go mod tidy or go get by hand. Nothing in the output says this is needed. Other paths already handle this:
- The hosted leg re-resolves and restores the original module's go.sum lines.
- The hosted→vendored takeover warns that
go mod tidywill be needed aftervendor --revert(vendor/golang.rs~L364).
The agent leg does neither.
Repro (hermetic, Linux, go 1.24.7)
SP=/path/to/target/release/socket-patch
W=$(mktemp -d); M=example.com/upstream; V=v1.0.0
mkdir -p $W/stage/$M@$V $W/proxy/$M/@v $W/c/.socket/blobs
printf 'module %s\n\ngo 1.21\n' $M > $W/stage/$M@$V/go.mod
printf 'package upstream\n\nfunc Greeting() string { return "PRISTINE" }\n' > $W/stage/$M@$V/lib.go
printf 'package upstream\n\nfunc Greeting() string { return "PATCHED" }\n' > $W/patched.go
echo "{\"Version\":\"$V\"}" > $W/proxy/$M/@v/$V.info; cp $W/stage/$M@$V/go.mod $W/proxy/$M/@v/$V.mod
(cd $W/stage && zip -qrD $W/proxy/$M/@v/$V.zip $M@$V)
export GOMODCACHE=$W/modcache GOPROXY=file://$W/proxy GOSUMDB=off GOTOOLCHAIN=local
cd $W/c
printf 'module example.com/consumer\n\ngo 1.21\n\nrequire %s %s\n' $M $V > go.mod
printf 'package main\n\nimport (\n\t"fmt"\n\t"%s"\n)\n\nfunc main() { fmt.Println("OUT:", upstream.Greeting()) }\n' $M > main.go
go mod download $M@$V && go mod tidy # go.sum: 2 lines
gsha(){ python3 -c "import hashlib,sys;d=open(sys.argv[1],'rb').read();print(hashlib.sha256(b'blob %d\0'%len(d)+d).hexdigest())" "$1"; }
B=$(gsha $W/stage/$M@$V/lib.go); A=$(gsha $W/patched.go); cp $W/patched.go .socket/blobs/$A
cat > .socket/manifest.json <<J
{"patches":{"pkg:golang/$M@$V":{"uuid":"4d5e6f70-8192-4a1b-8c2d-0123456789ab","exportedAt":"t","files":{"lib.go":{"beforeHash":"$B","afterHash":"$A"}},"vulnerabilities":{},"description":"","license":"","tier":""}},"setup":{"manual":["golang"]}}
J
$SP apply # exit 0, replace written
go mod tidy # supported per docs; go.sum is now empty
go build -o /dev/null . # OK (PATCHED)
$SP rollback # exit 0, "Rolled back packages: pkg:golang/example.com/[email protected]"
go build -o /dev/null . # exit 1: main.go:5:2: missing go.sum entry for module providing package example.com/upstream
$SP remove pkg:golang/example.com/[email protected] instead of rollback gives the same result. Without the go mod tidy step, both pass (go.sum keeps its 2 lines).
Expected vs actual
- Expected: CLI_CONTRACT.md "Rollback command contract (v5.0)" says a bare
rollback"restores the SYSTEM to unpatched". docs/ecosystems.md "Go: directory replaces and go.sum" says the wiring "survivesgo mod tidy". So after apply → tidy → rollback, the project should build exactly as it did before apply. That means either restoring the module'sh1:and/go.mod h1:lines (as the hosted leg does) or, at minimum, a warning plus a non-silent outcome telling the user to rungo mod tidy. - Actual: go.mod's
requireis left with no go.sum lines. Rollback/remove exit 0 withwarnings: [](--json), and the next defaultgo buildfails.
Matrix (Linux; each cell run 2× on go 1.24.7, 1× on the others)
| go | apply → rollback (no tidy) | apply → tidy → rollback | apply → tidy → remove |
|---|---|---|---|
| 1.22.12 | pass | fail | fail |
| 1.24.7 | pass | fail | fail |
| 1.26.8 | pass | fail | fail |
| macOS / Windows | untested (probe branches currently blocked) | untested | untested |
Vendored mode (vendor --revert after tidy) is untested here: it needs the mock patch service. The same drop-without-go.sum path looks likely.
Tested on main 61cfb9b (CLI 4.0.0). There have been no Go code changes since 4.0.0, so this isn't a regression in the 4.x line.
Suspect code
crates/socket-patch-core/src/patch/redirect/golang_local.rs:316(remove_go_redirect) only callsgo_mod_edit::drop_replace_entry(crates/socket-patch-core/src/vendor/go_mod_edit.rs:193) and removes the copy. go.sum is never reconciled and no warning is raised. Compare the hosted-takeover warning atcrates/socket-patch-core/src/vendor/golang.rs:364-377.
- 主要語言
- Rust
- 星號
- 8
- 分支
- 0
- 平均合併
- 15 小時 39 分鐘
- 30 天內合併 PR
- 104
環境準備
- 沒有 Dockerfile 或 Docker Compose 檔案
- 沒有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
SocketDev/socket-patch 的其他 Issue
-
agent:triaged bug bughunt pm:pipenv priority:p1
難度 2/5 1-3 小時 新手友好度 83/100
SocketDev/socket-patch#744 · 1 則留言 ·
維護者通常 1 天內回覆
-
agent:triaged bug bughunt pm:cargo priority:p2
難度 2/5 1-3 小時 新手友好度 84/100
SocketDev/socket-patch#651 · 3 則留言 ·
維護者通常 1 天內回覆
-
Vendored Hatch runs a `hatch` executable planted in the scanned project可能已有人在做 @mikolalysenko 於 2 天前認領。 未關閉agent:claimed agent:triaged arch-audit bug pm:hatch priority:p1
難度 2/5 半天 新手友好度 88/100
SocketDev/socket-patch#613 · 3 則留言 ·
維護者通常 1 天內回覆
-
Patch blob and diff downloads buffer the whole response body with no size cap可能已有人在做 @mikolalysenko 於 2 天前認領。 未關閉agent:claimed agent:triaged arch-audit bug priority:p3
難度 2/5 1-3 小時 新手友好度 84/100
SocketDev/socket-patch#571 · 5 則留言 ·
維護者通常 1 天內回覆
-
agent:triaged bug bughunt pm:composer priority:p2
難度 2/5 1-3 小時 新手友好度 90/100
SocketDev/socket-patch#515 · 1 則留言 ·
維護者通常 1 天內回覆
查看 SocketDev/socket-patch 的全部 Issue
相似的 Issue
-
area/cli parity
難度 2/5 1-3 小時 新手友好度 74/100
維護者通常 1 天內回覆
-
A-Picking A-UI C-Bug D-Trivial S-Ready-For-Implementation
難度 2/5 1-3 小時 新手友好度 72/100
bevyengine/bevy#26029 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 75/100
-
ai_p3 comp-protocols
難度 2/5 1-3 小時 新手友好度 74/100
ClickHouse/ClickHouse#123884 ·
維護者通常 1 天內回覆
-
state:needs triage
難度 2/5 1-3 小時 新手友好度 64/100
zed-industries/zed#65146 ·
維護者通常 1 天內回覆