Beagle SCM

SUBS-036: be post recursion into an ALREADY-ATTACHED sub tries to CREATE its synth dot-branch (PUTDUP) instead of FF — aborts the parent commit

A be post recursing into a submodule that is ALREADY on its synthetic dot-branch ?/<sub>/.<parent>/<pbranch> issues a CREATE (be put ?<synth>) for that branch — which already exists — and fails PUTDUP → BEDOGEXIT, aborting the whole parent commit. Per Submodules §"Committing detached subs" the create must be idempotent: "later posts fast-forward it" (exists → FF/attach, never re-create). SUBS-035 fixed the POSTDET / detached-commit half; this is the DISTINCT create-vs-FF half its hermetic test never exercised (there the sub was freshly created, not already attached). This is the ORIGINAL-session PUTDUP, still unfixed. Hit live 2026-06-26 committing COMMIT-004; worked around with be post --nosub. Will also block the journal-level gitlink bump (journal → be → be/test). Method: Issues.

Input

Context

• Live: from ~/journal/be, be post recursed post test into be/test (clean,

  attached at c47f767f on ?/beagle-ext-tests/.beagle-ext/.journal) → sniff: put: ?/beagle-ext-tests/.beagle-ext/.journal already exists / PUTDUP / BEDOGEXIT / "aborting parent commit — sub recursion failed (worst=BEDOGEXIT)".

• The sub was clean and ALREADY on its dot-branch; the recursion still issued a

  CREATE rather than an FF/no-op. So be post on a sub already at its synth tip cannot be driven by the parent.

• Distinct from SUBS-035 (which made a DETACHED sub commit): here the sub is

  ATTACHED and clean; the bug is the non-idempotent synth-branch CREATE.

• Workaround that landed COMMIT-004: be post --nosub (skips recursion; the

  gitlink bump was already correctly staged as put test#c47f767f).

Goals

• A parent be post recursing into a sub already on ?/<sub>/.<parent>/<pbranch>

  FF/attaches (or no-ops if already at tip), NEVER PUTDUP. Per Submodules: first post creates the branch, later posts fast-forward it.

• A clean sub at its dot-branch tip is a no-op (no spurious create/commit) and the

  parent still bumps the gitlink.

Constraints

• Make the synth-branch step idempotent: exists → FF (skip if already == tip);

  absent → create. Reuse SUBSDirty / the SUBS-027b R2 emit; ragel/REFS only.

• Do NOT regress SUBS-035 (first-post create on a detached sub), SUBS-006

  (clean sub no-op), or SUBS-027b (R2). Repro-first (CLAUDE §17).

• Localize: the recursion's synth-branch create that errs PUTDUP — likely

  sniff/PUT.c (the put ?<synth> arm) and/or beagle/SUBS.c bepost_bump_sub / the recurse cb / bepost_synth_child_uri.

WIP

Design decisions

• The create-vs-FF decision lives in bepost_anchor_sub (the R2 anchor): it issued

  bare be put ?<synth> → PUTCreateBranch (create-ONLY, errs PUTDUP if present). FIX: read the sub tip (R4) via SNIFFSubReadTip and issue be put ?<synth>#<tip>, routing through PUTSetBranch (create-or-repoint) — idempotent: absent → create, present → re-point/FF to the tip. The forward-only bump gate (bepost_bump_sub returns OK only on a real advance) already guarantees the tip is forward.

• User-facing be put ?<branch> keeps its strict create-only PUTDUP (DIS-020);

  only the internal anchor switched to the tip-form. No PUTCreateBranch change.

TODOs

• Repro-first hermetic: parent + sub ALREADY on its dot-branch; be post →

  assert FF, NO PUTDUP, the parent commit completes and the gitlink bumps. test/post/55-sub-attached-reanchor-idempotent (RED bare-synth, GREEN tip-form).

• Make the synth-branch anchor idempotent (exists → FF; absent → create).

  beagle/BE.cli.c bepost_anchor_sub → ?<synth>#<tip> (PUTSetBranch).

• Regression: SUBS-035 (post/54), SUBS-006 (post/29), SUBS-027b (post/49-52)

  still pass; ctest 439/439 (WITH_SSH=ON, WITH_JS=OFF), 0 fail.

Blockers and bummers

• None. Unblocks clean recursive be post (the journal-level gitlink bump uses it);

  --nosub is no longer needed once this lands.

Outcome

• LANDED ae64dc7f ("SUBS-036 idempotent sub synth-branch anchor", beagle trunk 26Jun); test/post/55-sub-attached-reanchor-idempotent registered in test/CMakeLists.txt. CLOSED (audit 2026-07-07).
• Root cause: bepost_anchor_sub

  (beagle/BE.cli.c:3025) anchored the SUBS-027b R2 keep-alive with bare be put ?<synth>, which dispatches (sniff/SNIFF.exe.c:1156) to PUTCreateBranch (sniff/PUT.c:1220) — a CREATE-ONLY writer that returns PUTDUP when the dot-branch already exists. On a LATER advance of an already-attached sub the branch IS present → PUTDUP → BEDOGEXIT, aborting the parent commit. SUBS-035/post-54 only exercised the FIRST anchor (branch absent), never the re-anchor; this is that create-vs-FF half.

• Fix (file:line): beagle/BE.cli.c bepost_anchor_sub (~3025) — read the sub

  tip (R4) via SNIFFSubReadTip and issue be put ?<synth>#<tip> instead of bare ?<synth>. The #<tip> form routes through PUTSetBranch (create-or-repoint), idempotent: absent → create, present → FF to the tip. SUBSNOSEC (no tip yet) → no-op. ~16 LoC, single function; no other file changed in beagle/sniff.

• Test: test/post/55-sub-attached-reanchor-idempotent (SSH-gated, registered in

  test/CMakeLists.txt). RED pre-fix: sniff: put: ?/sub/.parent already exists / PUTDUP / BEDOGEXIT / "aborting parent commit". GREEN post-fix: round-1 CREATEs the synth branch, round-2 FFs it to the new tip, parent commit completes, gitlink bumps.

• Tests: build-debug 439/439 (WITH_SSH=ON, WITH_JS=OFF), 0 fail; SUBS-035 (post/54),

  SUBS-006 (post/29), SUBS-027b (post/49/50/51/52) all green.

• Files changed (relative to beagle root): beagle/BE.cli.c,

  test/post/55-sub-attached-reanchor-idempotent/run.sh (new), test/CMakeLists.txt (register the case). STOPPED before committing.