You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fm-pr-merge refuses a protected branch with required_status_checks: null as unreadable #7343
bin/fm-pr-merge.sh's classic required-checks reader treats any non-object required_status_checks value as unreadable, including an explicit null.
A branch protected only by review (or other non-check) rules can report protected: true with protection.required_status_checks: null. That is readable emptiness for classic required checks - there are none - not an unreadable producer that must refuse the merge.
Expected
protected: true with a readable protection object whose required_status_checks is explicitly null -> no classic required checks; the merge is not refused for that source.
protected: true with no protection object, a protection object missing the required_status_checks key, or a non-null non-object value -> still unreadable; refuse.
Actual
Any non-object shape, including explicit null, is refused as an unreadable branch protection summary.
Notes
Split out of #4932 so the null-checks fix can land separately from the Treehouse pool-roots work.
Speaking as Kun's firstmate: thank you @mehulbhagwani for splitting this out of #4932. A narrow issue is much easier to land.
Checked on main 9b1ada04.github_read_required_contexts in bin/fm-pr-merge.sh (around line 898) refuses any non-object .protection.required_status_checks, so an explicit null is treated as an unreadable summary, as you describe. A missing protection object and a missing key take that same refusal path.
Existing PR:#7347 (fix(bin): treat null classic required checks as empty, Fixes #7343) is exactly the split you describe. It treats an explicit null inside a readable protection object as no classic required checks. It still refuses a missing protection, a missing required_status_checks key, and a non-null non-object value, and it adds tests for all of those. That PR is triaged on its own thread.
One thing that would help #7347: I sampled live github.com payloads for review-only protected default branches (facebook/react, golang/go). Both return an object ({"checks":[],"contexts":[],"enforcement_level":"off"}), not null. Could you share the forge, host, and version (for example GHES x.y) or the API call where you saw required_status_checks: null? This change loosens a merge refusal, so a real payload is the evidence that turns it from plausible into proven. It would also cover #7347's open live-validation scenario.
contract-class
restore, if the null shape is real. The refusal path exists for summaries that can't be read. A readable protection object that explicitly says "no classic checks" falls outside that promise, and rulesets are still read separately.
VISION.md per-rule
One captain, one interface: aligns. A false "unreadable" refusal is noise the captain has to unpick.
Authority is explicit: aligns. This only matters on paths where merge authority was already granted, and every other unreadable shape still refuses.
Scripts own the mechanics: aligns. This is an exact jq shape decision.
Activity
kunchenguid commentedon Oct 15, 2026
Speaking as Kun's firstmate: thank you @mehulbhagwani for splitting this out of #4932. A narrow issue is much easier to land.
Checked on main
9b1ada04.github_read_required_contextsinbin/fm-pr-merge.sh(around line 898) refuses any non-object.protection.required_status_checks, so an explicitnullis treated as an unreadable summary, as you describe. A missingprotectionobject and a missing key take that same refusal path.Existing PR: #7347 (
fix(bin): treat null classic required checks as empty,Fixes #7343) is exactly the split you describe. It treats an explicitnullinside a readableprotectionobject as no classic required checks. It still refuses a missingprotection, a missingrequired_status_checkskey, and a non-null non-object value, and it adds tests for all of those. That PR is triaged on its own thread.One thing that would help #7347: I sampled live github.com payloads for review-only protected default branches (
facebook/react,golang/go). Both return an object ({"checks":[],"contexts":[],"enforcement_level":"off"}), notnull. Could you share the forge, host, and version (for example GHES x.y) or the API call where you sawrequired_status_checks: null? This change loosens a merge refusal, so a real payload is the evidence that turns it from plausible into proven. It would also cover #7347's open live-validation scenario.contract-class
restore, if the
nullshape is real. The refusal path exists for summaries that can't be read. A readable protection object that explicitly says "no classic checks" falls outside that promise, and rulesets are still read separately.VISION.md per-rule
nullshape is a documented or observed forge semantic. That's the evidence asked for above.Outcome: existing-pr → #7347. This stays open until that PR lands.
fix(bin): treat protected+null classic required checks as empty
fix(bin): treat protected+null classic required checks as empty