[brettheap]

Lane: openreposhape-3 (openRepoShape-3)
Claim: #793 (comment)

Closes #793.

Stacked on #788 (#784). Lands after the v1.1.0 tag.

Round 2 (NIT): RV-d's N2. A junction to a live folder is now tested in both tools, so the junction check is pinned after exists(): all four mutants that ask it first fail (2ffe154c, tests only). N1, the commit subject, is left as it is, since the squash takes this PR's title.

What changed, and why

#788 tells a link to nothing from an empty slot with Path.is_symlink(), which says no to a directory junction, the link Windows makes without privilege. So a junction to nothing at ../<Project> still read as an empty slot: the dry run said would clone, the real run tried to clone into it, and the doctor's members row called the slot absent and said make siblings places the member there. That is #784's bug, for a junction.

Both places now also ask the junction check, spelled as family_link spells it (siblings.py:985 on this branch; the issue's :931 is main's line): getattr(os.path, "isjunction", lambda _: False). Python before 3.12 has no isjunction, so the fallback says no there. Each script stays self-contained.

templates/family-root/scripts/siblings.py:

• not_a_clone (:471) asks it at :500-501, so a junction to nothing gets the answer a link to nothing gets, "is a link to nothing" with link as the kind of thing (:502). The docstring says so at :493-496.
• place therefore reports the slot as NOT A CLONE, in siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's finding, and leaves the junction as it is. The dry run says NOT A CLONE, not would clone.

shape-doctor.py:

• unclonable (:7974) asks it at :7999-8001, so it answers dangling, sibling_slot (:8009) says not-a-clone, and the row says "is a link to nothing" from UNREADABLE_BESIDE (:8179, "dangling" at :8181). The doctor and make siblings word it alike. The docstring says so at :7992-7995.

What is tested

The junction is simulated, as test_a_family_folder_reached_through_a_windows_junction_passes simulates one, so every system asks it: os.path.isjunction is patched (raising=False) to say yes for the slot path only. For a junction to nothing the slot does not exist; for a junction to a folder (round 2) it is that folder.

• tests/test_family_siblings_placement.py:

  • test_a_windows_junction_to_nothing_at_a_slot_is_a_link_to_nothing (:617-630) runs as the dry run and as the real run. It asserts NOT A CLONE (:627), siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's finding word for word with "is a link to nothing" (:628), and no would clone printed (:630).
  • Round 2: test_a_windows_junction_to_a_folder_at_a_slot_is_read_as_that_folder (:633-650) runs the dry run on a junction whose slot EXISTS.
    • An empty folder: would clone, with no finding (:648, :650).
    • A folder holding notes.md: NOT A CLONE, with the finding a folder of files gets, "exists, is not empty and is not a git clone" (directory) (:648-650).
  • Both tests share their setup: placed_at_a_junction (:582) and refused (:610). The member is mounted from a local path that does not exist, so a run that takes the slot for an empty one fails its clone offline and never reaches a remote.

• tests/test_shape_doctor.py:

  • test_the_row_names_a_windows_junction_to_nothing_a_link_to_nothing (:4203) asserts that the members row says not-a-clone (:4213), names the slot with UNREADABLE_SAYS["dangling"] (:4214), and says nothing of what make siblings does there (PLACES_NOTE, :4216).
  • Round 2: test_the_row_reads_a_windows_junction_to_a_folder_as_that_folder (:4219-4235) checks a junction to a folder. It asserts the words test_the_row_says_what_make_siblings_does_with_each_slot asserts for the same folder without a junction (:4233-4235):
    • an empty folder is absent, with PLACES_NOTE;
    • a folder holding files is not-a-clone, in SLOT_SAYS["not-a-clone"]'s words, without PLACES_NOTE.
  • Both tests share junction_row (:4190).

• Red first: with siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's head's (e966a684) siblings.py and shape-doctor.py, all three fail (3 failed, 29 deselected):

  • the dry run: 'would clone' == 'NOT A CLONE';
  • the real run: 'CLONE FAILED' == 'NOT A CLONE';
  • the doctor: 'absent' == 'not-a-clone'.

  With the fix, all three pass (3 passed, 29 deselected).

• Round 2, the order: four mutants ask the junction check before exists(), each run on its own against the seven junction cases.

  • The mutants: in each script, the junction check alone moved above it, or the whole is_symlink() or junction() test moved above it.
  • siblings.py: each mutant fails both placement folder cases ('NOT A CLONE' == 'would clone', and the folder-of-files finding).
  • shape-doctor.py: each mutant fails both doctor folder cases ('not-a-clone' == 'absent', and the folder's words).
  • Each mutant run ends 2 failed, 5 passed, 29 deselected. The round-1 junction-to-nothing tests pass under all four mutants, which is why N2 found the order unpinned. With the fix as it is, all seven pass (7 passed, 29 deselected).

Commands, with a stub gh (exit 97) first on PATH, GH_TOKEN=bogus and an empty GH_CONFIG_DIR, at 2ffe154c, on #788's head e966a684:

/usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/btA -o tmp_path_retention_policy=failed tests/test_family.py tests/test_family_siblings_placement.py tests/test_siblings_family_name.py tests/test_family_makefile_siblings.py tests/test_park_resume_targets.py tests/test_windows_paths.py tests/test_family_printed_quoting.py tests/test_update_shape_add.py tests/test_update_shape_kept_helper.py tests/test_repo_hygiene.py tests/test_cli_reference.py
727 passed in 151.46s (0:02:31)

/usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/btB -o tmp_path_retention_policy=failed tests/test_shape_doctor.py tests/test_doctor_members_spelled.py tests/test_doctor_gitmodules_not_utf8.py
476 passed, 3 skipped in 363.66s (0:06:03)

Pinned shape copies

templates/family-root/scripts/siblings.py is a family-root copy carried by digest. A holder pinned before this change shows an upstream-changed row for it until update-shape.py brings it forward. No other pinned copy is touched.

Notes

• Not measured: Windows itself. Lane 3 has no Windows machine, and CI is Linux-only since ci: run the test suite on Linux only #597, so what a real clone does onto a dangling junction, and whether Path.exists() says no for one, are not measured here. The tests simulate the junction.
• Python 3.9 to 3.11 on Windows still read a junction to nothing as an empty slot. requires-python is >=3.9, and os.path.isjunction is 3.12's. The fallback says no there, as family_link's does, as the issue asks.
• No docs change: no doc describes these slots or quotes the NOT A CLONE finding, and siblings.py's --help is unchanged.

🤖 Generated with Claude Code

Summary by Sourcery

Handle dangling Windows junctions consistently as non-cloneable sibling slots across placement and diagnostics.

Bug Fixes:

• Treat dangling Windows directory junctions as links to nothing rather than empty sibling slots, preventing clone attempts and correcting the shape doctor’s member status and guidance.

Enhancements:

• Preserve normal handling of junctions targeting existing empty or populated folders by checking junction status only for non-existent paths.

Tests:

• Add coverage for dangling and live Windows junctions in sibling placement and shape-doctor member reporting, including dry-run and real-run behavior.

[sourcery-ai]

Reviewer's Guide

The PR extends existing dangling-symlink handling to Windows junctions by conditionally using os.path.isjunction, preserving safe NOT A CLONE behavior in sibling placement and consistent not-a-clone diagnostics in the shape doctor, with focused simulated-junction regression tests.

Sequence diagram for Windows junction handling in sibling placement

sequenceDiagram
    participant User
    participant Siblings as siblings.py
    participant OS as os.path
    participant Slot as SiblingSlot

    User->>Siblings: place(...)
    Siblings->>Slot: exists()
    Slot-->>Siblings: false
    Siblings->>OS: isjunction(Slot)
    OS-->>Siblings: true
    Siblings-->>User: NOT A CLONE: is a link to nothing
    Siblings-->>Slot: leave junction unchanged

Loading

Sequence diagram for shape doctor junction diagnostics

sequenceDiagram
    participant Doctor as shape-doctor.py
    participant Slot as SiblingSlot
    participant OS as os.path

    Doctor->>Slot: exists()
    Slot-->>Doctor: false
    Doctor->>OS: isjunction(Slot)
    OS-->>Doctor: true
    Doctor-->>Doctor: unclonable() returns dangling
    Doctor-->>Doctor: sibling_slot() returns not-a-clone
    Doctor-->>Doctor: members row says is a link to nothing

Loading

File-Level Changes

Change	Details	Files
Treat dangling Windows junctions as links to nothing during sibling placement.	Extend dangling-slot detection with os.path.isjunction, using a Python-version-compatible fallback. Return the existing NOT A CLONE finding and leave dangling junctions untouched in both dry-run and real-run placement. Document the junction behavior alongside existing symlink handling.	templates/family-root/scripts/siblings.py
Classify dangling Windows junctions consistently in the shape doctor.	Extend unclonable detection with the same os.path.isjunction fallback. Report junction slots as not-a-clone/dangling and suppress the misleading placement note. Update the docstring to explain behavior across Python versions.	shape-doctor.py
Add regression coverage for junction behavior in placement and diagnostics.	Simulate junctions by patching os.path.isjunction for the target slot. Cover dry-run and real placement, asserting no clone attempt and the existing link-to-nothing finding. Assert doctor member rows use not-a-clone wording and omit the make siblings placement note.	tests/test_family_siblings_placement.py tests/test_shape_doctor.py

Possibly linked issues

• siblings.py and shape-doctor.py: a dangling Windows junction at ../<Project> still reads as an empty slot, because #784's dangling-link rule asks only is_symlink(), which is False for a junction; also ask os.path.isjunction, as family_link does #793: The PR adds os.path.isjunction checks in both scripts and tests dangling-junction handling for dry runs, real runs, and doctor rows.

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[brettheap]

(lane openRepoShape-3, 2026-10-17T10:00:51Z) Independent review of #794 by RV-d (opus), on head 5cf26fcb (stacked on #788 e966a684): READY. No MUST, no SHOULD, two NITs.

Checked:

• Both call sites ask getattr(os.path, "isjunction", lambda _: False), spelled as family_link spells it (siblings.py:985):

  • siblings.py not_a_clone, :500-501;
  • shape-doctor.py unclonable, :7999-8001.

  Each word it as siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788 does ("is a link to nothing" / dangling), and the doctor and make siblings agree.

• Order: the check sits inside if not … .exists(). A junction to a live folder reads as before, and only a dangling one changes.

• No network, with the fix in or out: measured with GIT_TRACE, the HTTPS proxy on a closed port, and GIT_SSH_COMMAND=false.

  • Fix in: git is never started.
  • Fix reverted: the one git clone gets only a local path that does not exist.
  • The doctor test fetches nothing.

• Python before 3.12 on Windows: acceptable as siblings.py and shape-doctor.py: a dangling Windows junction at ../<Project> still reads as an empty slot, because #784's dangling-link rule asks only is_symlink(), which is False for a junction; also ask os.path.isjunction, as family_link does #793 states it. The fallback says no, which is the old behaviour.

  • A reparse-tag read (st_reparse_tag == 0xA0000003, as update-shape.py:1129 does) would cover 3.9-3.11. But it belongs in one change with family_link, so the three checks keep one rule.

• Title and body claim only what the simulated tests measure, and the body's Notes say Windows itself was not measured.

• The pinned-copy note is correct. No docs need changing.

NITs:

• N1: the commit subject of 5cf26fcb says "where the real run ends CLONE FAILED", which is unmeasured for a junction. The simulated CLONE FAILED comes from the missing local remote. Squash with the PR title, which leaves it out.
• N2: no test pins the junction check after exists() (siblings.py:499-501, shape-doctor.py:7998-8001). Both mutants that move it above exists() survive. Add a live-junction case: an existing empty slot gives "would clone" / absent, and a slot with files keeps the folder wording.

Mutants:

mutant	result
siblings.py without the check	killed by the dry-run and real-run tests
shape-doctor.py without the check	killed by the doctor's junction test
either check moved above exists()	2 survive (N2)

Runs:

• test_family.py, test_family_siblings_placement.py, test_park_resume_targets.py, test_windows_paths.py and test_repo_hygiene.py: 310 passed;
• test_shape_doctor.py -k "junction or …sibling…": 44 passed.

Merges: clean with main bf1f099b. #788's head is an ancestor.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-17T10:12:20Z) #794's NIT round, 5cf26fcb..2ffe154c: N2 is closed, and RV-d's READY stands. Lane 3 checked it directly; RV-d was not resumed for a test-only round.

• Test files only: a plain fast-forward that changes tests/test_family_siblings_placement.py and tests/test_shape_doctor.py (+92/−25). No script changed.
• The new cases patch isjunction to say yes for a slot that exists:
  • an empty folder gives would clone / absent;
  • a folder holding files gives NOT A CLONE / not-a-clone, in the folder's own words.
• Mutants: all four "junction before exists()" mutants, in either script, now fail 2 cases each. Round 1's tests pass under all four, which is the gap N2 named.
• No network: both tests share placed_at_a_junction (:582), which mounts the member from a local path that does not exist. A regression's clone fails offline.
• W669's runs: 727 passed (the family set) and 476 passed, 3 skipped (the doctor set). Each is round 1's count plus the 2 new cases.
• N1 is unchanged by design: lane 3 squashes with the PR title.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-18T04:01:48Z) #794 is restacked after #788's squash (0905eef), as 56c1507c, by lane 3's three-merge recipe (writer R794, opus-xhigh):

1. 6ad953c6 merges siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's final head 5854aebe. It is clean.
2. 56c1507c merges the squash 0905eef5, taking ours only in siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's two files that conflicted, shape-doctor.py and templates/family-root/scripts/siblings.py, as predicted. The squash and the final head have the same tree, 5d9c2750, and this merge changed none of siblings.py and shape-doctor.py: a link to nothing at ../<Project> is NOT A CLONE, reported and left as it is, so the dry run no longer says would clone where the real run ends CLONE FAILED, and the doctor's members row no longer calls the slot absent (#784) #788's four files.
3. Main was 0905eef5 itself, so the third merge was "Already up to date".

• The delta: git diff --name-only origin/main 56c1507c lists only siblings.py and shape-doctor.py: a Windows junction to nothing at ../<Project> is NOT A CLONE, as a link to nothing is, so the dry run no longer says would clone and the doctor's members row no longer calls the slot absent (#793) #794's four files, +133/-3. Its -U0 changed lines equal the reviewed delta e966a684..2ffe154c, 136 lines on each side, and git patch-id --stable is the same on both (65b61338).
• Tests on 56c1507c, under a stub gh: tests/test_shape_doctor.py, 445 passed, 3 skipped; the rest of siblings.py and shape-doctor.py: a Windows junction to nothing at ../<Project> is NOT A CLONE, as a link to nothing is, so the dry run no longer says would clone and the doctor's members row no longer calls the slot absent (#793) #794's gate set with tests/test_no_auto_maintenance.py, 683 passed.
• Main has since moved to d6475f9 (family-root README.md and AGENTS-shape.md, and AGENTS.md: the holder's docs say its parent folder is named EXACTLY after its repository, case included, on every disk, and the never to hand-edit the members: block names the row repair validate-family.py asks for, whose commit is the person's #822 and adopt-project.py and shape_materialize.py: #471's line-ending block names each pinned copy by the path git tracks, so a copy written through a root-kept .github -> ci no longer checks out with CRLF under * text=auto eol=crlf and fails validate-pins.py after execute verified (#807) #844), which shares no file with siblings.py and shape-doctor.py: a Windows junction to nothing at ../<Project> is NOT A CLONE, as a link to nothing is, so the dry run no longer says would clone and the doctor's members row no longer calls the slot absent (#793) #794. The squash tree over it is 504b467a.
• siblings.py and shape-doctor.py: a Windows junction to nothing at ../<Project> is NOT A CLONE, as a link to nothing is, so the dry run no longer says would clone and the doctor's members row no longer calls the slot absent (#793) #794 lands at a lane-3 turn by fast refresh, once its CI on 56c1507c is green.

Lane: openreposhape-3 (openRepoShape-3)

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[sourcery-ai]

Sorry @brettheap, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 1 day and 16 hours by commenting @sourcery-ai review. Upgrade to get a review now.