Repository navigation
🐛 dco: accept GitHub-noreply sign-offs from the same login and waive 687c3cb - #15488
Conversation
…87c3cb The post-merge trailer check failed v5 687c3cb (squash of #15315): GitHub wrote the PR author's private-email identity 91499691+odttlnt@users.noreply.github.com while the branch's Signed-off-by stayed sachin.naik@ibm.com. Same person; pre-merge DCO passed because the commits API resolves that address to @odttlnt. - check-dco-trailers.sh: same_person_noreply_evidence accepts a noreply author signed off under another address when the inspected branch's own history (DCO_SAME_PERSON_WINDOW, default 500) shows the link: authored-under-signoff-email (same name authored/committed under the sign-off address) or login-and-email-cooccur (both identities on one commit). No GitHub API call. Reported as an informational same-person-noreply line with the evidence commit, counted in the summary; a waiver on an accepted commit is STALE-WAIVER as before. DCO_SAME_PERSON_EVIDENCE_REFS adds refs for local checks. - lib-dco-identity.sh: github_noreply_login helper (pure). - Waive 687c3cb: v5 has no prior commit under the work address (only the PR branch commit 97a705c), so the heuristic has nothing to cite yet. - Self-tests for both evidence kinds, the three negatives (unknown address, different name, repeated ambiguous shape), window=0, waiver fallback, stale-waiver, evidence refs and config errors. - CONTRIBUTING.md / v5-sync-policy.md document the rule. Fixes #15401 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of No conflicting open PRs or same-file overlaps were found. |
|
@clubanderson is auto-promoted for contributor-lane merging on this PR: the last 5 merged PRs by this author in this hive's repositories have no revert or sentinel-alert evidence. CI, reviewer verdict, diff fences, soak, holds and branch protection still apply. |
There was a problem hiding this comment.
Decision needed: @hanthor or @Danathar (CODEOWNERS for .github/workflows/**) should decide whether a display-name match against a prior commit under the sign-off address is acceptable DCO policy for protected v5 history, since the author name on a squash is contributor-controlled.
This touches .github/workflows/dco-post-merge.yml and the DCO compliance checker, so it needs a code-owner's call rather than mine. The code itself I read and exercised; I found no defect.
Verified facts
687c3cb0onv5is authoredSachin Naik <91499691+odttlnt@users.noreply.github.com>withSigned-off-by: SACHIN NAIK <sachin.naik@ibm.com>, exactly as the waiver comment states.git log v5 --author=sachin.naik@ibm.comreturns nothing else, so there is no local evidence yet and the per-SHA waiver is the only way the heuristic would not cover it.src/scripts/test-check-dco-trailers.shon this head: 40 cases pass. The same test file against thev5checker fails six new cases (same-person noreply squash … not accepted,co-occurrence squash … not accepted,evidence ref was not consulted), so the tests do gate the behaviour.github_noreply_login(src/scripts/lib-dco-identity.sh:112-133) rejects a non-numeric id prefix and an empty login;same_person_noreply_evidence(src/scripts/check-dco-trailers.sh:255-302) returns 1 when the sign-off is itself a noreply address, skips the commit under test, and the evidence refs default to the inspected branch only, so a PR cannot smuggle its own evidence in CI.CONTRIBUTING.mdandsrc/docs/v5-sync-policy.mddescribe the new acceptance and the waiver fallback consistently with the script.
The policy question (not a bug, and the PR body is candid about it)
The authored-under-signoff-email rule at check-dco-trailers.sh:270-274 accepts when some earlier commit in the last 500 was authored or committed under the sign-off address by the same lower-cased name. On a GitHub squash the author name is the PR author's GitHub display name, which the contributor sets freely. So the heuristic's strength is "this login's display name equals a name that has previously committed under that address" — meaningful against mistakes, weak against intent. The pre-merge probot check is already spoofable by a local git config, so this does not lower the floor, but it is a documented widening of what the post-merge monitor accepts without a human waiver, on a protected branch. That is the decision for an owner.
Side note for anyone running the harness locally: it inherits GIT_AUTHOR_EMAIL/GIT_COMMITTER_EMAIL from the environment; with a [bot] address exported every fixture commit is skipped as a bot and the suite fails spuriously. Unsetting them makes it pass. Pre-existing, not introduced here.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
af61945
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @clubanderson, head
Hive did not add the |
3f2fe4d
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @clubanderson, head
Hive did not add the |
99600d2
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @clubanderson, head
Hive did not add the |
There was a problem hiding this comment.
Decision needed: @hanthor should decide whether the post-merge DCO checker may accept a noreply-authored commit with a real-address sign-off on login-and-email-cooccur evidence (check-dco-trailers.sh:284-299), given that a cherry-pick with -s by a noreply-identity maintainer creates exactly that co-occurrence for two different people.
Routing to a second approver: this relaxes DCO identity matching (a licensing/provenance gate) and the author is an approver. The waiver itself is in the documented format and the workflow diff touches only env (comments + one SHA appended to DCO_WAIVED_COMMITS), no permissions.
What I verified at head 99600d2:
src/scripts/check-dco-trailers.sh:278-282—authored-under-signoff-emailrequires the earlier commit's author (or committer) name to equal the current author name and its email to equal the sign-off address. That is a reasonable same-person link; the PR's own test run (evidence=authored-under-signoff-email:97a705cac) uses this branch.src/scripts/check-dco-trailers.sh:284-299—login-and-email-cooccurdoes not check the name. It accepts when any earlier commit in the window has the sign-off address as author/committer email and either (a) the login's noreply as the other of author/committer, or (b) the login's noreply among that commit'sSigned-off-bytrailers (:290-298).
Mechanism for a false link via (b): maintainer M (git identity <id>+M@users.noreply.github.com) runs git cherry-pick -s on contributor B's commit. The result has author = B's work address and trailers Signed-off-by: B, Signed-off-by: M-noreply. Any later commit authored as M-noreply carrying a Signed-off-by: B <work address> — a squash that inherited B's trailer, which is precisely the shape #15401 is about — is then accepted as same-person with evidence=login-and-email-cooccur, even though M and B are different people. Path (a) has the same property for co-authored/committed-by pairs (author B, committer M-noreply).
Consequence: the checker would silently accept a sign-off from someone other than the author in exactly the case DCO exists to catch, printing an informational line instead of FAIL. Severity medium — it needs a prior co-occurrence in a 500-commit window, which is common on this repo's branches (maintainer cherry-picks between v4/v5/v6 are routine).
Suggestion, not a demand: drop sub-path (b) and require the name match in (a) too, or keep login-and-email-cooccur but report it as a soft STALE-WAIVER-style notice while still FAILing. I have not changed anything; the test file (test-check-dco-trailers.sh) would need a case for the cherry-pick shape either way.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
@clubanderson this PR is held at Why:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. |
There was a problem hiding this comment.
Decision needed: @clubanderson should decide whether authored-under-signoff-email (display-name equality + prior commit under the sign-off address) is strong enough evidence for the post-merge DCO monitor, or whether only login-and-email-cooccur should auto-accept.
This edits .github/workflows/dco-post-merge.yml and loosens the DCO compliance monitor, so I am routing it rather than calling it fine. What I verified, then the one question.
Verified
src/scripts/test-check-dco-trailers.shpasses on the PR head (99600d2) when run withGIT_AUTHOR_*/GIT_COMMITTER_*unset: all 12 new casesok, including "same ambiguous shape is not evidence",WINDOW=0disables, and stale-waiver on an accepted commit. (With those env vars set they overridegit config user.*in the fixtures and the harness reports a false failure — pre-existing harness trait, CI unaffected.)- Workflow change is a waiver append of
687c3cb0plus the rationale block (dco-post-merge.yml:292-309), in the format of the existing entries; no permission, trigger, or step changes. lib-dco-identity.sh:113-137github_noreply_loginis pure string parsing; the pre-merge gates are untouched.- Body matches diff; the three-run table is reproducible from the script's flags.
The decision (medium, check-dco-trailers.sh:276-280)
authored-under-signoff-email accepts the noreply author when any of the last 500 commits has the sign-off address as author/committer and the same author name (case-insensitive). The name on a squash commit is the PR author's GitHub display name, which any account can set to anything. So the rule links <id>+login@users.noreply to a historical contributor's address on nothing stronger than a display-name match — e.g. a new account named "Good Author" opening a PR signed off as good@example.com would pass this monitor once merged. The second rule (login-and-email-cooccur, :282-298) is sound: it requires the login's own noreply to appear on a commit under that address.
Mitigations already in place: this is post-merge, a maintainer had to merge, and each acceptance prints evidence=<kind>:<sha> so it is auditable. Whether that is enough for a compliance monitor is a policy call, not a code defect — hence the question above. If the answer is "name match is fine", nothing else here blocks landing.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @clubanderson whether the local-history heuristic in check-dco-trailers.sh (name-equality link at src/scripts/check-dco-trailers.sh:275-279) is an acceptable DCO policy relaxation; unchanged since the previous review.
Delta re-review (99600d2 → 59f6068, correctness): the only change is merge commit 59f6068 absorbing v5 #15498 (src/pkg/dashboard/api_campaigns_jam_live_test.go, already on v5). gh pr diff --name-only shows the PR's seven files are byte-identical to the previously reviewed head. No new correctness findings; the prior finding at src/scripts/check-dco-trailers.sh:275-279 is still present and still requires a human policy call.
— hive: agent=ci-maintainer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @clubanderson: whether authored-under-signoff-email (name + address equality in branch history, no API) is an acceptable identity link for post-merge DCO; prior security finding is unchanged at 386429b.
Delta re-review (99600d2 → 386429b, correctness): the two new commits are both Merge branch 'v5' pulling in already-landed #15498 and #15478 (dashboard/tracker files only). No PR-authored line changed; the PR's diff against v5 is still the 7 DCO files. No correctness findings in the delta.
Prior finding status: src/scripts/check-dco-trailers.sh:278-282 — still present, identical code (previously cited at :276; line shift is from the base merge). Correctness-wise the rule does what the PR body says; whether that evidence is strong enough is the policy call above.
— hive: agent=adjudicator-3 backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @clubanderson: accept the authored-under-signoff-email name-equality rule as a DCO identity link on a protected-branch gate (unchanged since the prior review).
Delta re-review (99600d2 → 386429b, intent-alignment): both new commits are Merge branch 'v5' (59f6068, 386429b); every changed file is from the v5 side (api_campaigns_jam_live_test.go, tracker CSS/JS/HTML, changelog.d/fixed-15451-…) and none of the PR's seven DCO files were touched. Nothing in the delta changes what this PR does for #15401. The waiver + rationale for 687c3cb satisfies the issue's stated close condition.
Prior finding — still present: src/scripts/check-dco-trailers.sh:277-278 — authored-under-signoff-email accepts when %an of a prior commit equals the current commit's author name and %ae/%ce equals the sign-off address. Author name is set by whoever makes the commit, so the link is only as strong as a self-asserted display name. This is a governance choice about what the post-merge gate treats as identity evidence; the delta does not alter it.
— hive: agent=ci-maintainer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @clubanderson: is a local name-equality link an acceptable identity proof for the post-merge DCO monitor (authored-under-signoff-email), given both inputs are contributor-controlled?
Delta re-review (99600d2 → 386429b, security perspective): the delta is only the two v5 merge commits (#15478 tracker-row pill helper, #15498 jam-live test). stageTrackerPRPillsHtml (src/pkg/dashboard/static-src/js/tracker-row.js:1009-1011) escapes baseRef via esc() and passes lineage through exactly as the two call sites did before; no new trust boundary. No security findings in the delta.
Prior finding — still present (unchanged by the delta): src/scripts/check-dco-trailers.sh:276-281. authored-under-signoff-email accepts a noreply-authored squash when any commit in the last DCO_SAME_PERSON_WINDOW commits has author/committer email == sign-off email and author/committer name == this commit's author name (case-insensitive). On a GitHub squash the author name is the PR author's profile display name and the sign-off address is whatever they put on the branch commits — both chosen by the contributor. A login that sets its display name to an existing contributor's name and signs off with that contributor's address matches an earlier commit by the real person and is reported same-person-noreply instead of mismatched-signoff. The login-and-email-cooccur branch (:283-296) does not have this weakness, since it requires <login>'s own noreply to appear on the evidence commit.
Mitigant to weigh: this is the post-merge monitor; the pre-merge squash-attribution gate (CONTRIBUTING.md:231-245) rejects a sign-off naming a different human, so the question is whether the monitor may be weaker than the gate. If not, the concrete tightening is to drop the name-equality kind and keep only login-and-email-cooccur (plus the per-SHA waiver for first-time cases, which the PR already documents at CONTRIBUTING.md:272-274):
- if { [ "$ev_ae_lc" = "$sp_signoff_lc" ] && [ "$ev_an_lc" = "$sp_author_name_lc" ]; } ||
- { [ "$ev_ce_lc" = "$sp_signoff_lc" ] && [ "$ev_cn_lc" = "$sp_author_name_lc" ]; }; then
- printf 'authored-under-signoff-email:%s' "$(git rev-parse --short "$ev_sha")"
- return 0
- fi
-— hive: agent=adjudicator-4 backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
Decision needed: @clubanderson should decide whether the authored-under-signoff-email evidence kind — which turns on a display-name match that the PR author sets themselves — is an acceptable basis for auto-accepting a noreply-author / foreign-sign-off squash that today lands on a human's desk via the monitor.
Trust-policy change (post-merge DCO acceptance rules plus a workflow-side waiver), so I am routing rather than judging. Facts I verified by reading src/scripts/check-dco-trailers.sh and test-check-dco-trailers.sh at 386429b:
What the code does. For a commit whose author is <id>+<login>@users.noreply.github.com and whose sign-off is some other address, same_person_noreply_evidence (check-dco-trailers.sh:253-305) scans the branch's last 500 commits and accepts on either of:
authored-under-signoff-email(:272-276): an earlier commit authored/committed under the sign-off address whose author/committer name equals this commit's author name, case-insensitive. Nothing else is compared.login-and-email-cooccur(:278-295): an earlier commit under the sign-off address that also names<login>'s noreply in its identities or sign-offs. This one ties the login to the address and looks sound to me.
Negative tests exist for "same address, different name" and "two commits of the same ambiguous shape" (test-check-dco-trailers.sh, the pass lines at the hunks around +388 and +463), which confirms the name is the discriminator in path 1.
The open question. On a squash merge the author name is the PR author's GitHub profile name, which the account holder sets freely. So path 1 means: a contributor whose login is X can set their display name to match an existing v5 contributor, author and sign off their branch commits as that contributor's address (probot DCO compares strings only), and after squash the post-merge check resolves same-person-noreply … evidence=authored-under-signoff-email:<sha> and passes. Before this PR that commit was mismatched-signoff and the hourly monitor filed it for disposition (:375-394 is the new branch that now intercepts it). I am not claiming this is exploitable in practice — the squash-attribution gate and reviewer eyes sit in front of it — only that path 1 removes the human from a case it currently reaches, on the strength of a self-asserted field. Whether that trade is right is a maintainer call. Path 2 alone would resolve the motivating 687c3cb case once @odttlnt's next commit lands, at the cost of one more waiver cycle.
The 687c3cb waiver itself (dco-post-merge.yml:292-309) matches the shape documented in the existing entries, and its author is the repository owner, so there is no question about who is disposing it.
Docs (CONTRIBUTING.md, src/docs/v5-sync-policy.md, changelog fragment) describe the behaviour accurately as implemented.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
merged by Hive trusted-author auto-merge: author @clubanderson holds role owner and GitHub permission admin on repo; CI green at 386429b. |
|
Thank you for your contribution! Your PR has been merged. We'd love to hear how your experience was: share feedback |
Fixes #15401
Two failure modes behind the red monitor
The commit itself (
687c3cb0, squash of 🐛 fix: stop filing the pre-restart kick-undeliverable warning as a fleet issue #15315). GitHub's squash wrote the PR author's private-email identitySachin Naik <91499691+odttlnt@users.noreply.github.com>while the branch'sSigned-off-by: SACHIN NAIK <sachin.naik@ibm.com>carried over. Same person (GitHub user odttlnt, id 91499691); pre-merge probot DCO and the squash-attribution gate both passed because the commits API resolves the work address to @odttlnt. Protected v5 cannot be re-signed, so the SHA is waived inDCO_WAIVED_COMMITSwith a rationale block (format as 🌱 ci: 3 dashboard shuffle slices on v5; waive dropped squash sign-off #13904):The checker could not reason about this shape at all. It accepted a noreply sign-off for the authoring login, but a noreply author with a real-address sign-off was always
mismatched-signoff.check-dco-trailers.shnow hassame_person_noreply_evidence: when the author is<id>+<login>@users.noreply.github.com, the sign-off is accepted if the inspected branch's own history (DCO_SAME_PERSON_WINDOW, default 500 commits) shows the link —authored-under-signoff-email(same name authored/committed under that address before) orlogin-and-email-cooccur(that address and<login>'s noreply appear on one earlier commit). No GitHub API call. Accepted commits print an informationalsame-person-noreply <sha> … evidence=<kind>:<sha>line instead of FAIL and are counted in the summary; a waiver on an accepted commit is reportedSTALE-WAIVERas before. A commit of the same ambiguous shape is deliberately not evidence.DCO_SAME_PERSON_EVIDENCE_REFSlets a local run add refs (e.g. the PR branch). Pre-merge probot/prow/squash-attribution semantics are unchanged;lib-dco-identity.shonly gains the pure helpergithub_noreply_login.Local verification against
origin/v5(50-commit window)1 waived; 0 same-person-noreply→passed (with recorded waivers), rc 0FAIL 687c3cb0… mismatched-signoff— no prior commit undersachin.naik@ibm.comexists on v5, so the heuristic has nothing to cite and the waiver is the right instrument todayDCO_SAME_PERSON_EVIDENCE_REFS=<PR #15315 branch>same-person-noreply 687c3cb0… evidence=authored-under-signoff-email:97a705cac, rc 0Once the contributor's next commit lands on v5 under the work address, the monitor will report the waiver as
STALE-WAIVERand it can be removed.Tests / docs
src/scripts/test-check-dco-trailers.sh: 12 new cases (both evidence kinds; unknown address, different-name address and repeated-shape negatives;DCO_SAME_PERSON_WINDOW=0; waiver fallback; stale-waiver; evidence refs; config errors). Existing cases unchanged..github/scripts/test-dco-waiver-integrity.pypasses with the new entry (16 waivers).src/docs/v5-sync-policy.mddocument the rule;changelog.d/fixed-dco-postmerge-noreply-identity.md.