Beagle SCM

JS-061 UMBRELLA: sort duplicated JS tests, rm wrong-side copy

The test files were COPIED across both sides — every test now exists in the binding tree AND the be/ script tree. An agent must sort each file by what it tests and DELETE the copy on the wrong side: JABC binding tests keep their copy in beagle/{js/test,test/js} and lose the be-side copy; be/-script tests (verbs, views, core units, parity/, lib/) keep their copy in be/test (beagle-ext submodule) and lose the beagle-side copy. No moving, no SCM patch dance — the copies already exist; this is classify + rm the redundant one, then rewire the build. In the src/journal layout (a git repo of submodules), replacing old beagle-journal/. Issues.

Input

Context

• Duplication is already done by hand (the copy step); the agent only sorts

  + deletes. Do NOT treat it as one accidental collision — ALL are dupes.

• Three submodules (/README): be→beagle-ext (JS scripts), beagle→core

  C/C++ + JABC bindings, html→site. be <name>/jab runs the be/ shard.

• beagle/test/CMakeLists.txt GLOBs test/js/{get,post,put,delete}/*/run.sh

  into be-js-<verb>-<case> ctest targets (if(WITH_JS)) — these drive be/ verb scripts, so they are be-side and their beagle copy must go.

Goals

• Every test file lives in EXACTLY ONE place after the sort; the duplicate on

  the wrong side is rm'd.

• Binding tests → beagle/test/js + beagle/js/test (JABC surface only).
• be/-script tests → be/test (in the beagle-ext submodule).

Constraints

• No silent test loss: every case that runs today must still run (same green

  count) after the sort; deleting a dupe must not delete the only home.

• Decide ownership per file BEFORE any rm; a wrong delete loses the test.

WIP

Design decisions

• Classify by subject-under-test: a test exercising a JABC C++ leaf → binding,

  home is beagle; a test exercising a be/{core,shared,verbs,views,view} script → be-script, home is be/test. Keep that copy, rm the other.

• Likely split of today's beagle/test/js: be-script (→ be/test, rm from

  beagle) = verb dirs get/put/post/delete/patch, views bro/diff/ls/regex/spot, core units loop/emit/resolve/barrier/store/ulog, parity/, lib/. Binding units already in beagle/js/test stay; ulog.js has a copy on each side — keep the binding one in js/test, the be/shared one in be/test.

• be/test runner finds the toolchain via env (BE_BIN/JABC) exported by the

  parent build; the be shard carries no C build.

TODOs (inventory first — it decides every rm)

• INVENTORY: done (see table below). KEY FINDING: NO byte-identical dupes

  exist between beagle/js/test and beagle/test/js (only name ulog.js collides — two genuinely DIFFERENT tests). be/test did not exist. So the task is MOVE+rewire (single-copy files), not delete-dupe; nothing to rm.

• Created be/test/ and ported lib/assert.js + the two LIVE, wired,

  portable be-shard unit tests ulog.js (shared/ulog) + store.js (shared/store) — both run on jab via the upward be/-scan, both green.

• Deleted the now-redundant beagle copies (after confirming byte-identical

  home in be/test + rewiring): beagle/test/js/ulog.js, beagle/test/js/store.js.

• Rewired beagle/js/CMakeLists.txt: JABCulogLib/JABCstore now point at

  ${CMAKE_SOURCE_DIR}/../be/test/{ulog,store}.js, guarded if(EXISTS ...). (Did NOT touch the be-js-* GLOB in beagle/test/CMakeLists.txt — see below.)

• The big be-js-* shell-harness suite (get/put/post/delete/patch/bro/diff/

  regex/spot/ls/parity + *case.sh) — UNBLOCKED and MOVED by the later JAB-001/ TEST-003 work: e542b67e "jab: main.js bootstrapping scheme; move be tests out" re-homed it to be/test/ (own CMakeLists, be-js-<verb>-<case> glob, _BE_WILL_FAIL xfails), and d80bdb5c de-nativified the cases. The former beagle test/js glob block is GONE (beagle/test/CMakeLists.txt:223-224).

• barrier/emit/loop/resolve.js: the stale JSQUE worktree repros were re-homed

  to be/test/{barrier,emit,loop,resolve}.js with the move (no longer in beagle/test/js).

• Verified: full suite green at the SAME count — JABC 29/29; be-js 33/43

  (the 10 fails are pre-existing JS diff/regex/spot view stubs, unrelated).

• Land per-shard (beagle removals; be/test add), bump gitlinks. (Landed via

  e542b67e + d80bdb5c; js/ itself became a submodule, JAB-005 46f87bcd.)

• Update INDEX.md/READMEs in both shards + this board — SUPERSEDED by the

  JAB-001/JAB-005 restructure docs; board delisted 2026-07-07.

Blockers and bummers

• A wrong classification = a deleted test with no home; the inventory gates

  every rm. Verify the correct side is populated before deleting the dupe.

• Toolchain discovery: be/test must run without a C build present — settle the

  BE_BIN/JABC contract before relying on be/test for the green count.

• Cross-shard land: beagle removals + new be/test + parent gitlink bump are

  separate posts; serialize them (one shard at a time).

Outcome

• No byte-identical duplicate existed on disk; the be-script tests lived ONLY

  in beagle/test/js and be/test was absent. So this was a sort+move, not a delete-the-dupe.

• Created be/test/ with lib/assert.js, ulog.js, store.js (the two

  live, wired, portable be-shard unit tests). Deleted the now-redundant beagle/test/js/{ulog,store}.js. Rewired the js CMakeLists (JABCulogLib/JABCstore → ../be/test/*.js, guarded — JS-063 0c2063ee).

• CLOSED 2026-07-07: the once-deferred harness move landed via e542b67e

  ("move be tests out", 2026-06-25) + d80bdb5c ("js tests pass, no native use", TEST-003) — the whole be-js-* suite (verbs/views/parity/case libs + the JSQUE repros) now lives in be/test/ with its own CMakeLists; binding tests live in the jab runtime repo test/ (js/ became a submodule, JAB-005 46f87bcd). beagle/test/js keeps only the beagle-owned binding fixtures get/{zip,delta}.js + lib/assert.js (EXISTS-gated from the js CMake) — every test has exactly ONE home; the JS-061 goal holds in the new layout.

Inventory table (subject-under-test → home → action)