Repository navigation
fence: go.sum completeness across the multi-module tree (#976) - #985
Merged
Merged
Conversation
The #975 class: a merge changed one module's import surface (#938's nip19 in package config_manager) and left three importers' go.sum incomplete; every gate that ran stayed green because go.sum completeness was nobody's beat — check-deps-sync.py covers go.mod Require sync only, and per-PR verification builds the PR's touched modules, not every importer. Only the 2026-10-16 night battery caught it, hours after merge. tests/contract/check-gosum-completeness.sh probes every module under src/ (glob discovery, never enumeration) with the exact failing command — go build ./... from inside the module dir (#961), GOWORK=off and -mod=readonly so nothing rescues an incomplete go.sum — and fails loudly naming the modules, quoting the missing-entry lines, with the go-mod-tidy fix hint. ~2s warm for all 18 modules; test files are the battery's beat, keeping the fence per-push cheap. Negative controls, both proven: the hermetic --selftest strips the go.sum lines of a dependency chosen from the module's own dep graph (rot-proof) and requires the probe to go red; and the live one — on current main the fence is RED on exactly merchant_types, upstream_detector and upstream_session_manager until #975 lands (the documented sequencing: #975 first, or accept the temporary red). Wired into release-check's deps leg and both test.yml twins' deps-and- imports jobs (the twins have drifted on main — pre-existing, noted in the PR, not touched here).
Collaborator
Author
|
Evidence review (#976 fence — independently verified on x280, live negative control included):
Ready for approving review; tracked on #965. |
This was referenced Oct 17, 2026
Open
Closed
Collaborator
Author
Local CI gate — PR #985: fence: go.sum completeness across the multi-module tree (#976)Run of
Verdict: 2 leg(s) RED — 4 green. What this gate does not run (no vacuous green)
failing-leg log tailscontract-fences — tailgo-battery — tail |
Collaborator
Author
|
Stamp context (this seat): both reds are the documented tree-age classes — battery = the pre-#975 go.sum trio (this branch adds the fence for that class; the fix is #975, so its own base carries the breakage by design), contract-fences = the pre-#1000 doc-facts/desync drift. Fittingly, the fence this PR ships goes green the moment #975 lands (proven in the release-prep dry-run: 18/18 inside release-check with both merged). |
Collaborator
Author
Local CI gate — PR #985: fence: go.sum completeness across the multi-module tree (#976)Run of
Verdict: 2 leg(s) RED — 4 green. What this gate does not run (no vacuous green)
failing-leg log tailscontract-fences — tailgo-battery — tail |
This was referenced Oct 20, 2026
Collaborator
Author
Evidence review — 13-criteria pass (verified at head, with a bonus live negative)Fence mechanism verified twice over at this head:
Shape review: probe = the exact failing command ( Merge note (same class as #1039): rebase onto current main first — the three carriers are fixed there (#1013/#1032/#1030), so the fence goes green on its own tree; without the rebase it ships red-on-self. After merge: this is the fence the #975/#1013 history justifies, and the natural negative-control home for the cold-cache class #1013's diagnosis named (warm-cache CI passing while cold vet fails — worth one seeded-failure test in the follow-up the body's boundary statement already scopes). Verdict: mechanically ready post-rebase. Human gate. |
c03rad0r
approved these changes
Oct 26, 2026
c03rad0r
left a comment
c03rad0r
left a comment
Contributor
There was a problem hiding this comment.
Technical review — head 5eb29a7. The fence is right and it has already earned its keep; approve with a sequencing condition.
The shape is exactly the check-ssid-format pattern done well: probe the real failing command (GOWORK=off GOFLAGS=-mod=readonly go build ./...) from inside each glob-discovered module, categorize the failure, and refuse to hide a non-go.sum red behind the category. The hermetic --selftest is the part most fences skip: deleting the last go.sum line did not fail (not every sum is build-required), so the fixture derives its victim from go list -deps — the control is load-bearing.
Verified here:
bash tests/contract/check-gosum-completeness.sh --selftest
selftest ok: stripping 'github.com/ImVexed' go.sum lines in a temp copy of config_manager fails the probe
It also caught a real main break, which I fixed: current main still fails the probe in src/upstream_detector and src/upstream_session_manager (missing btcsuite/btcd/btcec/v2 resolution from go-nostr v0.51.12; #975 was closed unmerged). I opened #1120 with go mod tidy in both modules — all 18 modules now build with GOWORK=off -mod=readonly, so the fence lands green.
Condition: the branch still points at an older main, so at its head the fence is red (3 modules, incl. merchant_types). Rebase onto current main after #1120 lands and re-run the fence before merge — otherwise the PR that adds the fence is the one that turns the lane red.
Disposition: land (after #1120 + rebase).
c03rad0r
added a commit
that referenced
this pull request
Oct 26, 2026
* fix(deps): complete go.sum for the two go-nostr carriers (#975) `GOWORK=off GOFLAGS=-mod=readonly go build ./...` fails in src/upstream_detector and src/upstream_session_manager: go-nostr v0.51.12 imports btcsuite/btcd/btcec/v2 and the carriers' go.sum lacks the entries that resolve it (btcutil v1.1.6). The night battery caught it; no per-PR gate did, which is what #985's fence is for. `go mod tidy` in both modules adds the missing sums and drops requires the tidy graph no longer needs. Verified all 18 module builds green with GOWORK=off and -mod=readonly, and deps-sync + gonuts-pin fences still pass. * changelog: fragment for #1120 --------- Co-authored-by: felixfelix-bot <felixfelix-bot@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
tests/contract/check-gosum-completeness.sh— the fence for the #975 class: a merge can change one module's import surface (e.g. #938 adding nip19 topackage config_manager) and leave every importer's go.sum incomplete.go buildthen fails withmissing go.sum entry, but only for that module — and no gate owned that beat (check-deps-sync.py= go.mod Require sync; per-PR verification builds the PR's touched modules, not every importer). Only the 2026-10-16 night battery caught it, hours after merge.Shape (per #976's proposal, and the repo's fence patterns)
go build ./...from inside each module dir (ci: go test fails in git worktrees — 'main module does not contain package' error #961 discipline), withGOWORK=off GOFLAGS=-mod=readonly— no workspace rescue, no re-tidy rescue; each module resolves against its own go.sum.go.modundersrc/(18 today; a new module must not need a fence edit).missing go.sum entrylines, prints thego mod tidyfix hint to stderr; non-gosum build failures are reported separately (still red — a fence that hides a red behind a category is decoration).go testlegs — that boundary is stated in the header).release-check.shdeps leg (the per-push shape) + bothtest.ymltwins'deps-and-importsjobs, one identical step each.Negative controls — both proven
--selftest): temp-copies a real module, strips the go.sum lines of a dependency chosen from that module's owngo list -depsoutput (so the fixture cannot rot with dep churn), and requires the probe to go red. Verified:selftest ok: stripping 'github.com/ImVexed' go.sum lines in a temp copy of config_manager fails the probe. (A naive "delete the last go.sum line" fixture was tried first and did not fail — not every sum line is build-required; the graph-derived victim is the fix.)src/merchant_types,src/upstream_detector,src/upstream_session_manager— the fix(deps): restore battery-green main — bech32 go.sum entries for #938's nip19 import surface #975 breakage, named, with thebtcsuite/btcd/btcutil/bech32missing-entry lines quoted, exit 1. Full transcript in the verification section below. This is the fence working as designed: main stays red until fix(deps): restore battery-green main — bech32 go.sum entries for #938's nip19 import surface #975 lands — sequencing note for the owner: land fix(deps): restore battery-green main — bech32 go.sum entries for #938's nip19 import surface #975 first and this PR's lanes go green, or land this first and accept the temporary red (either order proves the fence).Evidence (this branch)
bash -nclean;release-check.shsyntax-clean with the new leg.Notes
test.ymltwins are documented byte-identical (.ngit/README.md) but have substantially diverged on main (the.githubtwin carries a happy-path job and more). This PR added the identical step to both; reconciling the twins is its own change.