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
Kernel: a merge read while an issue is held or declined is a Hold, not done (#227) #237
Part of #3. From the Maintainer's ruling on #227 (2026-10-20): the Controller dequeues the Factory's pull request on a person's hold or close, so a merge never lands while an issue is held or declined; the issue Lifecycle and tools/check_lifecycle.py stay as they are.
The Kernel's boundary case.kernel/issue.py reads a merge of a linked pull request in the present life as done (derive, the merged row), before it reads a Hold or a close. A merged fact read while the issue is held (a person's factory:hold, or hands-off) or declined (closed by a person, or reopened by someone without write access) is the platform doing what the Controller should have prevented. It is not done: it is a Hold, with a reason code that names the contradiction, so a person sees it.
To do.
A boundary case in tests/test_issue.py for each of the three: merged under factory:hold, merged under hands-off, merged after a person's close; and one for a merge after an outsider's reopening (reopened_last_by: other).
The fix in kernel/issue.py, with the new reason code in kernel/terms.py, its feed entry in config/feed.json, and a counter-example if a new field is read.
The issue world (sim/issues.py) keeps making no merge while held or declined; once the Kernel holds this, the world may make one on a fault seed and check the Hold.
The seeds that found it: 20 (held, then the queue merged), 22 (closed, then the queue merged), 99 (reopened by an outsider, then a colleague merged their own), in #232.
Activity
Kernel: a merge read while an issue is held or declined is a Hold, no…