Multi-milestone work and documents on the new loop (M2.4)

- Every milestone branch is cut from the latest default branch: the next
  milestone builds on the merged previous one, and a redo after a closed PR
  drops the rejected work. Uncommitted work comes along.
- While the item's PR is open, start_milestone is refused, with the PR's
  state read from Gitea at that moment, not the last watch round.
- finish_milestone(final): earlier PRs say "Part of #N", the last "Closes
  #N"; unsaid, it's final, as before.
- close_item with the item's PR still open waits for the merge (L2: a
  closed item leaves nothing open).
- The driver's instructions cover ARCHITECTURE.md and build plans under
  docs/plans/ (D2); the harness never parses a plan.
- Parking stashes uncommitted work and pushes it to <item>-parked, put back
  on resume, instead of committing it onto the item's branch. The walk
  found a half-done review fix parked onto a PR that had merged meanwhile,
  where it was lost. Architecture D4's parking line says so.

Tests: unit tests for fresh milestone branches, the merge wait (including
a merge not yet watched), redo after a closed PR, Part of/Closes, close
waiting on an open PR, and the parked-fix regression; the walk's new
milestones_start_fresh property (fork-from-current fails 9 seeds); an e2e
two-milestone feature.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

nisse-dev/tests/test_loop_multi.py

@@ -0,0 +1,176 @@
+"""Multi-milestone work on the new loop (M2.4): each milestone starts from
+the latest default branch, the next waits for the previous PR's merge (read
+from Gitea when asked), earlier PRs are part of the issue and the last closes
+it, and the item closes only with nothing left open (D1, D2, D4). Real git on
+the throwaway build host; a fake Gitea."""
+
+from typing import Any
+
+from test_loop_milestones import (
+    RESTATE,
+    Forge,
+    ScriptedModel,
+    call,
+    git,
+    item,
+    loop_agent,
+    say,
+)
+from test_loop_milestones import world as world  # noqa: PLC0414 -- the fixture
+
+from nisse_dev.canonical import ToolCall, ToolResult
+from nisse_dev.loop.items import MERGE, WAITING, loop_of
+
+
+def milestone(path: str, final: str) -> list[Any]:
+    return [
+        call("start_milestone", name=path),
+        call("write_file", path=path, content=f"# {path}\n"),
+        call("finish_milestone", summary=f"adds {path}", title=path, final=final),
+    ]
+
+
+def tree(world: dict[str, Any], branch: str) -> list[str]:
+    bare = str(world["bare"])
+    return git("--git-dir", bare, "ls-tree", "--name-only", branch).split()
+
+
+def result_of(world: dict[str, Any], tool: str) -> str:
+    events = world["store"].transcript(item(world).id, "driver").read()
+    ids = {
+        e.event.call_id
+        for e in events
+        if isinstance(e.event, ToolCall) and e.event.tool == tool
+    }
+    results = [
+        e.event.result
+        for e in events
+        if isinstance(e.event, ToolResult) and e.event.call_id in ids
+    ]
+    return results[-1]
+
+
+def test_the_second_milestone_builds_on_the_merged_first(world: dict[str, Any]) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        milestone("a.py", final="no"),
+        milestone("b.py", final="yes"),
+        [call("close_item", summary="both done")],
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    [first] = forge.prs
+    assert forge.prs[first]["body"].endswith("Part of #14")
+    forge.merge(first)
+    agent.watch()
+    agent._schedule()
+
+    second = max(forge.prs)
+    assert second != first
+    assert "Closes #14" in forge.prs[second]["body"]
+    assert forge.files(second) == ["b.py"]  # only its own work
+    head = forge.prs[second]["head"]
+    assert "a.py" in tree(world, head)  # cut from main, with the first merged
+    assert loop_of(item(world)).milestone == 2
+
+    forge.merge(second)
+    agent.watch()
+    agent._schedule()
+    assert item(world).status == "closed"
+
+
+def test_the_next_milestone_waits_for_the_merge(world: dict[str, Any]) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        milestone("a.py", final="no"),
+        [  # a message arrives while the PR is open: try to move on
+            call("start_milestone", name="b"),
+            call("wait_for_review", note="waiting for the merge"),
+        ],
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    say(transport, world, "how's it going?")
+    assert "hasn't merged yet" in result_of(world, "start_milestone")
+    assert (item(world).position, loop_of(item(world)).wait) == (WAITING, MERGE)
+    assert len(forge.prs) == 1
+
+
+def test_a_merge_not_yet_watched_is_read_from_gitea(world: dict[str, Any]) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        milestone("a.py", final="no"),
+        milestone("b.py", final="yes"),
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    [first] = forge.prs
+    forge.merge(first)  # merged in Gitea; no watch round has seen it
+    say(transport, world, "merged, carry on")
+    assert len(forge.prs) == 2
+    assert "a.py" in tree(world, forge.prs[max(forge.prs)]["head"])
+
+
+def test_a_redo_after_a_closed_pr_drops_the_rejected_work(
+    world: dict[str, Any],
+) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        milestone("a.py", final="yes"),
+        milestone("c.py", final="yes"),
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    [first] = forge.prs
+    forge.prs[first]["state"] = "closed"  # rejected
+    agent.watch()
+    agent._schedule()
+    redo = max(forge.prs)
+    assert forge.files(redo) == ["c.py"]
+    assert "a.py" not in tree(world, forge.prs[redo]["head"])
+
+
+def test_closing_with_a_pr_open_waits_for_the_merge(world: dict[str, Any]) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        milestone("a.py", final="yes"),
+        [call("close_item", summary="done, I think")],  # too soon
+        [call("close_item", summary="done")],
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    say(transport, world, "thanks")
+    assert item(world).status == "open"
+    assert (item(world).position, loop_of(item(world)).wait) == (WAITING, MERGE)
+    [pr] = forge.prs
+    forge.merge(pr)
+    agent.watch()
+    agent._schedule()
+    assert item(world).status == "closed"
+
+
+def test_an_unsaid_final_closes_the_issue(world: dict[str, Any]) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        [
+            call("start_milestone", name="a"),
+            call("write_file", path="a.py", content="x = 1\n"),
+            call("finish_milestone", summary="adds a"),
+        ],
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    [pr] = forge.prs
+    assert "Closes #14" in forge.prs[pr]["body"]

nisse-dev/tests/test_loop_properties.py

@@ -79,7 +79,11 @@ class PolicyModel:
         return [
             call("start_milestone", name=f"m{self.n}"),
             call("write_file", path=f"f{self.n}.py", content=f"x = {self.n}\n"),
-            call("finish_milestone", summary=f"work {self.n}"),
+            call(
+                "finish_milestone",
+                summary=f"work {self.n}",
+                final=self.rng.choice(["yes", "no"]),
+            ),
         ]
 
 
@@ -308,6 +312,7 @@ class Walk:
             f"one_active (nothing left running): {where}"
         )
         self._parked_work_safe(where)
+        self._milestones_start_fresh(where)
         heads = [p["head"] for p in forge.prs.values()]
         assert len(heads) == len(set(heads)), f"one_pr_per_milestone: {where}"
         bare = self.world["bare"]
@@ -325,6 +330,29 @@ class Walk:
             )
             assert landed, f"review_fixes_land (PR {n}): {where}"
 
+    def _milestones_start_fresh(self, where: str) -> None:
+        """D4: a milestone starts from the latest default branch, so one
+        redone after a closed PR never carries the rejected work."""
+        forge: Forge = self.world["forge"]
+        bare = self.world["bare"]
+        for item in self.items():
+            mine = sorted(
+                (n, p)
+                for n, p in forge.prs.items()
+                if p["head"].startswith(f"{item.id}-milestone-")
+                and "-fix-" not in p["head"]
+            )
+            for i, (n, earlier) in enumerate(mine):
+                if earlier["merged"]:
+                    continue
+                rejected = _rev(bare, f"refs/heads/{earlier['head']}")
+                if _is_ancestor(bare, rejected, "refs/heads/main"):
+                    continue
+                for m, later in mine[i + 1 :]:
+                    assert not _is_ancestor(
+                        bare, rejected, f"refs/heads/{later['head']}"
+                    ), f"milestones_start_fresh: PR {m} carries PR {n}: {where}"
+
     def _parked_work_safe(self, where: str) -> None:
         """L1, D4: nothing of a parked item lives only on the build VM: what
         every branch of an open item that isn't on the workspace points at is
@@ -367,7 +395,17 @@ class Walk:
             for path in written:
                 if here and (root / path).exists():
                     continue
-                history = _git(bare, "log", "--all", "--format=%H", "--", path)
+                history = _git(
+                    # Full history: a parked stash keeps new files in its
+                    # third parent, which plain log simplifies away.
+                    bare,
+                    "log",
+                    "--all",
+                    "--full-history",
+                    "--format=%H",
+                    "--",
+                    path,
+                )
                 assert history, f"parked_work_safe: {path} of #{item.id} lost: {where}"
 
 

nisse-dev/tests/test_loop_queue.py

@@ -88,12 +88,14 @@ def test_switching_parks_the_work_and_resuming_finds_it(
     agent.watch()
     agent._schedule()
 
-    # Switching to 14 parked 15: its half-done work is committed and pushed.
+    # Switching to 14 parked 15: its half-done work is pushed as a stash to
+    # a side branch (new files in its third parent), never onto 15's branch.
     b_branch = loop_of(item(world, 15)).branch
-    pushed = git(
-        "--git-dir", str(world["bare"]), "show", f"refs/heads/{b_branch}:lines.py"
-    )
-    assert pushed == "L = 1"
+    bare = ["--git-dir", str(world["bare"])]
+    parked = f"refs/heads/{item(world, 15).id}-parked"
+    assert git(*bare, "show", f"{parked}^3:lines.py") == "L = 1"
+    # The branch has no commits of its own, so it wasn't pushed at all.
+    assert b_branch not in git(*bare, "branch", "--list")
     # 14's fix went to its own branch, without 15's work.
     assert loop_of(item(world, 14)).wait == MERGE
     files = git("--git-dir", str(world["bare"]), "ls-tree", "--name-only", a_branch)
@@ -105,6 +107,8 @@ def test_switching_parks_the_work_and_resuming_finds_it(
     say(transport, world, 15, "one per line")
     assert world["repo"].current_branch() == b_branch
     assert (world["repo"].root / "lines.py").read_text() == "L = 1\n"
+    # Put back, so the side branch is gone.
+    assert item(world, 15).id + "-parked" not in git(*bare, "branch", "--list")
     assert loop_of(item(world, 15)).wait == MERGE
     assert len(forge.prs) == 2
 
@@ -190,3 +194,52 @@ def test_status_shows_the_line(world: dict[str, Any]) -> None:
     assert "#15 Add count_lines: ready, 2 of 2 in line" in agent.status()
     agent.handle(InboundMessage(DON, "do #15 next"), None)
     assert "#15 Add count_lines: ready, 1 of 2 in line" in agent.status()
+
+
+def test_a_fix_half_done_when_its_pr_merges_survives_a_switch(
+    world: dict[str, Any],
+) -> None:
+    """Found by the walk: work on a review fix, parked after the PR merged,
+    was pushed onto the merged branch, where nobody would see it. Now it
+    comes back with the item and goes out in its next PR."""
+    forge: Forge = world["forge"]
+    two_issues(forge)
+    model = ScriptedModel(
+        [call("restate", text="count_words")],  # 14
+        [call("restate", text="count_lines")],  # 15
+        [  # 14: a PR
+            call("start_milestone", name="words"),
+            call("write_file", path="words.py", content="W = 1\n"),
+            call("finish_milestone", summary="words", final="yes"),
+        ],
+        [  # 14, a review comment: half a fix, then a question
+            call("write_file", path="fix.py", content="F = 1\n"),
+            call("ask_human", question="fix it in words.py too?"),
+        ],
+        [  # 15 runs meanwhile
+            call("start_milestone", name="lines"),
+            call("write_file", path="lines.py", content="L = 1\n"),
+            call("finish_milestone", summary="lines"),
+        ],
+        [call("finish_milestone", summary="the fix", final="yes")],  # 14 again
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, 14, "go")
+    [pr] = forge.prs
+    forge.comments[pr] = [Comment(7, "don", "add fix.py", "t")]
+    agent.watch()
+    agent._schedule()
+    assert loop_of(item(world, 14)).wait == ANSWER
+    say(transport, world, 15, "go")  # parks 14's half-done fix
+    forge.merge(pr)
+    agent.watch()
+    agent._schedule()  # 14 resumes; its fix goes out as a new PR
+
+    bare = ["--git-dir", str(world["bare"])]
+    merged_head = forge.prs[pr]["head"]
+    assert "fix.py" not in git(*bare, "ls-tree", "--name-only", merged_head).split()
+    newest = max(
+        n for n, p in forge.prs.items() if p["head"].startswith(item(world, 14).id)
+    )
+    assert newest != pr and forge.files(newest) == ["fix.py"]