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
Decide whether 0.7.x gets a patch release that refuses Python 3.15 #516
Split from #514. Cloudfall 0.7.0, the current stable release, declares requires-python = ">=3.14" and pins ansible-core>=2.21,<2.22, which does not start on Python 3.15. So on a machine whose newest Python is 3.15, uvx --from cloudfall==0.7.0 ... or a fresh uv sync of a project pinned to 0.7.0 installs and then fails at every playbook with AttributeError: module 'dataclasses' has no attribute '_is_type'.
Options
A 0.7.1 hotfix from release/0.7 whose only change is requires-python = ">=3.14,<3.15": the resolver then picks Python 3.14 or refuses, instead of installing a release that fails
@romamo Decision needed: should 0.7.x get a patch release that refuses to install on Python 3.15?
No patch; the compatibility page says 0.7.0 needs Python 3.14, and 0.8.0 is the fix (recommended): no release work on an old line, but someone installing 0.7.0 on 3.15 still gets a broken install until they read the page
Release 0.7.1 from release/0.7 with requires-python = ">=3.14,<3.15" and nothing else: an install on 3.15 picks Python 3.14 or is refused up front; it costs one hotfix release
Reply here with a number or your own answer; nothing is released until you do.
Roman chose option 2 in chat on 2026-11-17 ("Вариант 2, патч"): release 0.7.1 from release/0.7 with requires-python = ">=3.14,<3.15" and nothing else. No open 0.8.0 milestone gives a date for the stable release, so new installs on Python 3.15 would stay broken for an unknown time.
Triage: implement, on @romamo's answer (option 2: release 0.7.1 with only the requires-python cap)
shipmill's hotfix lane builds X.Y.(Z+1) on release/X.Y (or the tag when that branch doesn't exist). It applies the diffs of the main PRs it is given, except the CHANGELOG and changelog.d/. The cap must never reach main, since main supports 3.15 through #518. So the release takes two PRs:
0.7.1: refuse Python 3.15 in requires-python #531 into release/0.7, a new branch at v0.7.0: requires-python = ">=3.14,<3.15" plus the relock. The relock drops only the 3.15 markers and wheels, and no resolved version changes. Checked locally: uv lock --check passes, and uv sync --python 3.15 is refused with "incompatible with the project's Python requirement: ==3.14.*"
Changelog entry for 0.7.1: refuse Python 3.15 on the 0.7 line #532 into main: the changelog fragment and nothing else. Its applied diff is empty, so 0.7.1 takes its entry and none of main's code. The hotfix's sync then moves the entry into main's CHANGELOG under 0.7.1 and deletes the fragment
Once both have landed, the hotfix lane runs: lane=hotfix, hotfix-prs=532, hotfix-from=v0.7.0.
#531 has landed on release/0.7, which now holds v0.7.0 plus the requires-python cap.
#532, the 0.7.1 changelog entry, is reviewed. It waits on #525: main's head is red from the rc76 publish run, and #525, the repair, waits on its security-zone question. Once #532 lands, the hotfix lane releases 0.7.1.
@romamo Decision needed: the 0.7.1 hotfix can't pass the converge check D-23 requires, because the 0.7 line predates it; how should 0.7.1 ship?
#532 landed on main (92481fc), and the hotfix lane ran with lane=hotfix, hotfix-prs=532, hotfix-from=v0.7.0 (run https://github.com/cloudfall-dev/cloudfall/actions/runs/44391818494). ci / checks passed on the release commit. converge / converge failed in "Install dependencies": release.yml runs main's converge.yml against the release commit on release/0.7, and that tree has no engine/tests/converge/project (the converge harness came after 0.7.0). It's not a flake, so a rerun fails the same way. No 0.7.1 was tagged or published.
Let converge skip a tree without the converge project (recommended): converge.yml checks for engine/tests/converge/project and, when it's missing, passes with a notice that says so, and D-23 gets an amendment limiting the rule to trees that carry the harness. 0.7.1 then ships with CI green but no Debian converge. Its only change from 0.7.0 is the requires-python cap and a relock with no resolved version changed (0.7.1: refuse Python 3.15 in requires-python #531), so the roles are the ones 0.7.0 shipped. It's a small PR on main in .github/, which is the security zone, so it comes back to you for review
Backport the converge harness to release/0.7: 0.7.1 is converged like every other release. It costs a large backport whose harness was written for 0.8's roles and may not run on 0.7's at all
Skip 0.7.1: 0.7.0 stays the last 0.7.x, and users on Python 3.15 who install it keep getting a release whose playbooks fail at start; 0.8.0 is their fix. The entry from Changelog entry for 0.7.1: refuse Python 3.15 on the 0.7 line #532 then needs removing from changelog.d/, otherwise the next rc lists a 0.7.x fix
Until this is decided, an rc cut from main lists #532's 0.7.x entry in its notes.
Reply here with a number or your own answer; shipmill takes this up on the tick after your reply.
What also waits on this answer: the failed converge in run 44391818494 is recorded against main's head (92481fc), so main reads red and the merge gate lands nothing on it (D-75). #525 and #526 (both answered "1"), and the reviewed #517 to #523, wait on it. Option 1's PR would carry a Repairs: line for that run and clear it. Options 2 and 3 leave main red until some other commit lands on it.
Three more PRs wait on this answer, held by the same red base on main@92481fc: the approved spec PRs #428 (ClickHouse) and #504 (Kibana), and #464. Each has its approval answer, and its CI is green.
Roman approved the recommended option in chat on 2026-11-19 ("все по рекомендациям", a waiting-on-me round run by Claude Code). Checked right before relaying: the item is open, still labelled needs-decision, with no newer answer. Zone rule-departure at level ask; the .github/ PR the option leads to comes back to Roman for review.
Option 1 is up as #549: converge.yml checks for engine/tests/converge/project first and, on a tree without it, passes with a notice and converges nothing. D-28 records the rule and supersedes D-23. Its Repairs: line names run 44391818494, so it can land on main's red head.
#549 changes .github/, so it waits for @romamo's review on its own question. Once it merges, the hotfix lane can run again (lane=hotfix, hotfix-prs=532, hotfix-from=v0.7.0) to cut 0.7.1.
Activity
shipmill-romamo commentedon Nov 17, 2026
@romamo Decision needed: should 0.7.x get a patch release that refuses to install on Python 3.15?
release/0.7withrequires-python = ">=3.14,<3.15"and nothing else: an install on 3.15 picks Python 3.14 or is refused up front; it costs one hotfix releaseReply here with a number or your own answer; nothing is released until you do.
shipmill-romamo commentedon Nov 17, 2026
Decision by @romamo, relayed by Claude Code
2
Roman chose option 2 in chat on 2026-11-17 ("Вариант 2, патч"): release 0.7.1 from
release/0.7withrequires-python = ">=3.14,<3.15"and nothing else. No open 0.8.0 milestone gives a date for the stable release, so new installs on Python 3.15 would stay broken for an unknown time.shipmill-romamo commentedon Nov 17, 2026
Triage: implement, on
@romamo's answer (option 2: release 0.7.1 with only therequires-pythoncap)shipmill's hotfix lane builds X.Y.(Z+1) on
release/X.Y(or the tag when that branch doesn't exist). It applies the diffs of the main PRs it is given, except the CHANGELOG andchangelog.d/. The cap must never reach main, since main supports 3.15 through #518. So the release takes two PRs:release/0.7, a new branch atv0.7.0:requires-python = ">=3.14,<3.15"plus the relock. The relock drops only the 3.15 markers and wheels, and no resolved version changes. Checked locally:uv lock --checkpasses, anduv sync --python 3.15is refused with "incompatible with the project's Python requirement:==3.14.*"Once both have landed, the hotfix lane runs:
lane=hotfix,hotfix-prs=532,hotfix-from=v0.7.0.Refuse Python 3.15 on the 0.7 line (#516) (#531)
shipmill-romamo commentedon Nov 17, 2026
#531 has landed on
release/0.7, which now holdsv0.7.0plus therequires-pythoncap.#532, the 0.7.1 changelog entry, is reviewed. It waits on #525: main's head is red from the rc76 publish run, and #525, the repair, waits on its security-zone question. Once #532 lands, the hotfix lane releases 0.7.1.
Add the 0.7.1 changelog entry for the Python 3.15 cap (#516) (#532)
shipmill-romamo commentedon Nov 18, 2026
@romamo Decision needed: the 0.7.1 hotfix can't pass the converge check D-23 requires, because the 0.7 line predates it; how should 0.7.1 ship?
#532 landed on main (92481fc), and the hotfix lane ran with
lane=hotfix,hotfix-prs=532,hotfix-from=v0.7.0(run https://github.com/cloudfall-dev/cloudfall/actions/runs/44391818494).ci / checkspassed on the release commit.converge / convergefailed in "Install dependencies":release.ymlruns main'sconverge.ymlagainst the release commit onrelease/0.7, and that tree has noengine/tests/converge/project(the converge harness came after 0.7.0). It's not a flake, so a rerun fails the same way. No 0.7.1 was tagged or published.converge.ymlchecks forengine/tests/converge/projectand, when it's missing, passes with a notice that says so, and D-23 gets an amendment limiting the rule to trees that carry the harness. 0.7.1 then ships with CI green but no Debian converge. Its only change from 0.7.0 is therequires-pythoncap and a relock with no resolved version changed (0.7.1: refuse Python 3.15 in requires-python #531), so the roles are the ones 0.7.0 shipped. It's a small PR on main in.github/, which is thesecurityzone, so it comes back to you for reviewrelease/0.7: 0.7.1 is converged like every other release. It costs a large backport whose harness was written for 0.8's roles and may not run on 0.7's at allchangelog.d/, otherwise the next rc lists a 0.7.x fixUntil this is decided, an rc cut from main lists #532's 0.7.x entry in its notes.
Reply here with a number or your own answer; shipmill takes this up on the tick after your reply.
shipmill-romamo commentedon Nov 19, 2026
What also waits on this answer: the failed converge in run 44391818494 is recorded against main's head (92481fc), so main reads red and the merge gate lands nothing on it (D-75). #525 and #526 (both answered "1"), and the reviewed #517 to #523, wait on it. Option 1's PR would carry a
Repairs:line for that run and clear it. Options 2 and 3 leave main red until some other commit lands on it.shipmill-romamo commentedon Nov 19, 2026
Three more PRs wait on this answer, held by the same red base on main@92481fc: the approved spec PRs #428 (ClickHouse) and #504 (Kibana), and #464. Each has its approval answer, and its CI is green.
shipmill-romamo commentedon Nov 19, 2026
Roman approved the recommended option in chat on 2026-11-19 ("все по рекомендациям", a waiting-on-me round run by Claude Code). Checked right before relaying: the item is open, still labelled
needs-decision, with no newer answer. Zone rule-departure at level ask; the.github/PR the option leads to comes back to Roman for review.shipmill-romamo commentedon Nov 19, 2026
Decision by @romamo, relayed by Claude Code
1
shipmill-romamo commentedon Nov 19, 2026
Option 1 is up as #549:
converge.ymlchecks forengine/tests/converge/projectfirst and, on a tree without it, passes with a notice and converges nothing. D-28 records the rule and supersedes D-23. ItsRepairs:line names run 44391818494, so it can land on main's red head.#549 changes
.github/, so it waits for@romamo's review on its own question. Once it merges, the hotfix lane can run again (lane=hotfix,hotfix-prs=532,hotfix-from=v0.7.0) to cut 0.7.1.