Merge pull request #91 from crcresearch/feat/manifest-sync-trees

feat(manifest): sync trees — ship the test harness to derived and adopted repos (closes #90)

Author: psaboia

scripts/adopt.sh

@@ -289,6 +289,12 @@ ADD_PATHS=()
 while IFS= read -r _path; do
     ADD_PATHS+=("$_path")
 done < <(lw_manifest_assemble_active_files "" "$AGENT")
+# Sync-tree members (TEMPLATE_SYNC_TREES, e.g. the test harness) resolved
+# from the template checkout on disk — dir mode, since adopt copies from
+# $TEMPLATE_ROOT rather than a git ref (#90).
+while IFS= read -r _path; do
+    ADD_PATHS+=("$_path")
+done < <(lw_manifest_tree_files dir "$TEMPLATE_ROOT")
 for path in "${ADD_PATHS[@]}"; do
     src="$TEMPLATE_ROOT/$path"
     dst="$TARGET/$path"

scripts/check-template-version.sh

@@ -90,6 +90,10 @@ FILES=()
 while IFS= read -r _path; do
     FILES+=("$_path")
 done < <(lw_manifest_assemble_active_files "$REPO_ROOT" "")
+# Sync-tree members (TEMPLATE_SYNC_TREES) resolved from the template ref (#90).
+while IFS= read -r _path; do
+    FILES+=("$_path")
+done < <(lw_manifest_tree_files ref "$TEMPLATE_REF")
 
 IN_SYNC=()
 OUT_OF_DATE=()

scripts/lib/template-manifest.sh

@@ -65,6 +65,24 @@ TEMPLATE_SHARED_INFRA=(
     "scripts/wiki-write-protocol/scenarios/07-livelock-retry/run.sh"
     "scripts/wiki-write-protocol/scenarios/08-session-start-auto-pull/run.sh"
     "scripts/wiki-write-protocol/scenarios/09-session-start-divergent/run.sh"
+    # CI workflow for the test harness below. Ships at instantiation; synced
+    # here so derived repos' CI follows the harness instead of freezing (#90).
+    ".github/workflows/test-harness.yml"
+)
+
+# Trees the template owns end-to-end, synced as WHOLE TREES. Membership is
+# resolved from the sync source at run time via lw_manifest_tree_files —
+# the template ref for update/check, the template checkout for adopt —
+# never enumerated here: a static list of the harness's ~110 files would
+# rot on every added fixture (#90; three field cases of frozen inherited
+# harnesses, one with red CI for ten days).
+#
+# Contract: copy-no-delete. Files the template removed linger on hosts
+# until cleaned manually; files the HOST added under a tree are never
+# touched — deliberate, because features install their tests INTO this
+# tree (feature.json tests.destination = scripts/test/tests/unit/<name>/).
+TEMPLATE_SYNC_TREES=(
+    "scripts/test"
 )
 
 # Claude Code overlay files. Active when adopt is invoked with
@@ -191,6 +209,24 @@ lw_manifest_assemble_active_files() {
     fi
 }
 
+# Echo the current member files of every declared sync tree, one per line.
+# Two modes, matching the two authoritative sources:
+#   ref: $2 is a git ref readable from CWD (e.g. template/main); membership
+#        from git ls-tree — what update/check sync against.
+#   dir: $2 is a directory root (the template checkout, for adopt);
+#        membership from find, paths relative to that root, sorted.
+# Silent-empty when the tree is absent from the source: the consumer's
+# normal missing-file reporting handles it (no special case here).
+lw_manifest_tree_files() {
+    local mode="$1" src="$2" tree
+    for tree in ${TEMPLATE_SYNC_TREES[@]+"${TEMPLATE_SYNC_TREES[@]}"}; do
+        case "$mode" in
+            ref) git ls-tree -r --name-only "$src" -- "$tree" 2>/dev/null ;;
+            dir) ( cd "$src" 2>/dev/null && find "$tree" -type f 2>/dev/null | sort ) ;;
+        esac
+    done
+}
+
 # Echo the HOST_OWNED operation type for $1, or empty string if not host-
 # owned. Used by adopt to look up which grant type a host-owned path expects.
 lw_manifest_known_grant_type() {

scripts/test/tests/integration/harness-tree-sync/assertions.sh

@@ -0,0 +1,68 @@
+#!/usr/bin/env bash
+# Assertions: sync trees declared in the manifest (issue #90).
+#
+# update/check must resolve scripts/test membership from the TEMPLATE REF
+# (delivering new files, replacing stale ones, never touching host-added
+# files under the tree); adopt must resolve it from the template checkout
+# on disk and ADD the harness to adopted hosts.
+
+STAGE="$SANDBOX/harness-tree-sync"
+H="$STAGE/host"
+REPO_ROOT_HTS="$(cd "$HERE/../.." && pwd)"
+
+# --- update leg -------------------------------------------------------------
+UP_LOG="$STAGE/update.log"
+( cd "$H" && bash scripts/update-from-template.sh --template-url="$STAGE/template-src" ) \
+    > "$UP_LOG" 2>&1
+RC=$?
+
+assert "update exits 0" "[ $RC -eq 0 ]"
+assert "stale harness runner REPLACED with the template's" \
+    "grep -qF 'NEW-RUNNER-MARKER' '$H/scripts/test/run.sh'"
+assert "old runner content gone" \
+    "! grep -qF 'OLD-RUNNER-MARKER' '$H/scripts/test/run.sh'"
+assert "new harness lib delivered" \
+    "[ -f '$H/scripts/test/lib/guard.sh' ]"
+assert "new fixture delivered" \
+    "[ -f '$H/scripts/test/tests/unit/sample/assertions.sh' ]"
+assert "CI workflow delivered" \
+    "[ -f '$H/.github/workflows/test-harness.yml' ]"
+assert "host-owned feature test untouched (copy-no-delete)" \
+    "grep -qF 'HOST-FEATURE-TEST-MARKER' '$H/scripts/test/tests/unit/myfeat/assertions.sh'"
+
+# --- convergence + check leg ------------------------------------------------
+UP2_LOG="$STAGE/update2.log"
+( cd "$H" && bash scripts/update-from-template.sh --template-url="$STAGE/template-src" ) \
+    > "$UP2_LOG" 2>&1
+assert "second update reports zero changed files" \
+    "grep -qF 'Changed (0):' '$UP2_LOG'"
+
+CK_LOG="$STAGE/check.log"
+( cd "$H" && bash scripts/check-template-version.sh --template-url="$STAGE/template-src" ) \
+    > "$CK_LOG" 2>&1
+CK_RC=$?
+assert "check-template-version exits 0 (no drift incl. harness)" \
+    "[ $CK_RC -eq 0 ]"
+
+# --- adopt leg (dir-mode enumeration from the real template checkout) -------
+# Template-checkout only: this leg runs the CHECKOUT's own scripts/adopt.sh
+# against its own tree. In a derived project that binary is the frozen
+# instantiation-vintage adopt (not manifest-synced, by design), so its
+# behavior is unspecifiable here — observed in the field: a stub-era copy
+# dry-runs and ADDs nothing. Same discriminator as manifest-shape's guard.
+if [ ! -f "$REPO_ROOT_HTS/CLAUDE.md.template" ]; then
+    skip "harness-tree-sync adopt leg" "not a template checkout (derived project; local adopt.sh vintage is unspecified)"
+    return 0 2>/dev/null || true
+fi
+AD_LOG="$STAGE/adopt.log"
+( cd "$STAGE/adopt-host" && \
+  bash "$REPO_ROOT_HTS/scripts/adopt.sh" --target=. --apply --agent=none ) \
+    > "$AD_LOG" 2>&1
+AD_RC=$?
+assert "adopt --apply exits 0" "[ $AD_RC -eq 0 ]"
+assert "adopt ADDed the real harness runner" \
+    "[ -f '$STAGE/adopt-host/scripts/test/run.sh' ]"
+assert "adopt ADDed the harness lib" \
+    "[ -f '$STAGE/adopt-host/scripts/test/lib/template.sh' ]"
+assert "adopt ADDed the CI workflow" \
+    "[ -f '$STAGE/adopt-host/.github/workflows/test-harness.yml' ]"

scripts/test/tests/integration/harness-tree-sync/patch.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# Patch: fixture for manifest sync-trees (issue #90).
+#
+# Field case: scripts/test/ ships at instantiation but was absent from the
+# manifest, so derived repos ran a harness frozen at creation vintage —
+# red CI for days on naval-sensor-fusion (pre-#73 fixtures against a
+# derived tree). The fix declares scripts/test as a SYNC TREE whose
+# membership is resolved from the sync source at run time.
+#
+# Effects: creates $SANDBOX/harness-tree-sync/ with:
+#   template-src/  stand-in template repo (branch main): a minimal harness
+#                  tree (run.sh, lib/, one fixture), the CI workflow file,
+#                  and one SHARED_INFRA control file
+#   host/          derived project with CURRENT sync tooling (copied from
+#                  the working tree under test), a STALE harness runner,
+#                  and a host-owned feature test that the sync must never
+#                  touch (copy-no-delete contract)
+#   adopt-host/    virgin repo for the adopt leg (dir-mode enumeration)
+#
+# Hermetic: remotes are local paths.
+
+set -uo pipefail
+
+HERE_PATCH="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+TEMPLATE_ROOT="$(cd "$HERE_PATCH/../../../../.." && pwd)"
+
+STAGE="$SANDBOX/harness-tree-sync"
+mkdir -p "$STAGE"
+g() { git "$@"; }
+
+# --- template-src ---
+T="$STAGE/template-src"
+g init -q "$T"
+g -C "$T" symbolic-ref HEAD refs/heads/main
+mkdir -p "$T/scripts/test/lib" "$T/scripts/test/tests/unit/sample" "$T/.github/workflows"
+printf '#!/usr/bin/env bash\n# NEW-RUNNER-MARKER (post-guard harness)\n' > "$T/scripts/test/run.sh"
+printf '# guard lib shipped by template\n'                               > "$T/scripts/test/lib/guard.sh"
+printf '# sample fixture assertions\n'       > "$T/scripts/test/tests/unit/sample/assertions.sh"
+printf 'name: test-harness\non: [push]\n'    > "$T/.github/workflows/test-harness.yml"
+printf 'control shared file\n'               > "$T/llm-wiki.md"
+g -C "$T" add -A
+g -C "$T" commit -q -m "template content"
+
+# --- host: current tooling + stale harness + host-owned feature test ---
+H="$STAGE/host"
+g init -q "$H"
+mkdir -p "$H/wiki/hosty.wiki" "$H/scripts/lib" \
+         "$H/scripts/test/tests/unit/myfeat"
+: > "$H/wiki/hosty.wiki/SCHEMA_hosty.md"
+cp "$TEMPLATE_ROOT/scripts/update-from-template.sh"   "$H/scripts/"
+cp "$TEMPLATE_ROOT/scripts/check-template-version.sh" "$H/scripts/"
+cp "$TEMPLATE_ROOT"/scripts/lib/*.sh                  "$H/scripts/lib/"
+printf '#!/usr/bin/env bash\n# OLD-RUNNER-MARKER (pre-guard, fails in derived checkouts)\n' \
+    > "$H/scripts/test/run.sh"
+printf '# HOST-FEATURE-TEST-MARKER: installed by a feature; sync must never touch this\n' \
+    > "$H/scripts/test/tests/unit/myfeat/assertions.sh"
+
+# --- adopt leg: virgin host, adopted from the REAL template working tree ---
+A="$STAGE/adopt-host"
+g init -q "$A"
+g -C "$A" remote add origin "https://github.com/example-org/adopt-host.git"
+printf '# Adopt Host\n' > "$A/README.md"
+g -C "$A" add -A
+g -C "$A" commit -q -m "host baseline"
+
+echo "  harness-tree-sync patch applied: fixtures at $STAGE"

scripts/test/tests/smoke/manifest-convergence/patch.sh

@@ -110,10 +110,13 @@ fi
     source "$TEMPLATE_ROOT/scripts/lib/template-manifest.sh"
     # adopt mode: empty repo_root, AGENT drives overlay inclusion.
     lw_manifest_assemble_active_files "" "claude-code"
+    # Sync-tree members (dir mode, same source adopt itself uses, #90).
+    lw_manifest_tree_files dir "$TEMPLATE_ROOT"
 } > "$STAGE/expected-claude.txt" 2>/dev/null
 {
     source "$TEMPLATE_ROOT/scripts/lib/template-manifest.sh"
     lw_manifest_assemble_active_files "" "none"
+    lw_manifest_tree_files dir "$TEMPLATE_ROOT"
 } > "$STAGE/expected-none.txt" 2>/dev/null
 
 echo "  manifest-convergence patch applied: A=$A B=$B"

scripts/test/tests/unit/manifest-shape/assertions.sh

@@ -169,3 +169,32 @@ assert_eq "known_grant_sentinel(CLAUDE.md) is empty (managed-block delegates to
     "$(lw_mcall "lw_manifest_known_grant_sentinel 'CLAUDE.md'")"
 assert_eq "known_grant_sentinel(unknown path) is empty" "" \
     "$(lw_mcall "lw_manifest_known_grant_sentinel 'random/path.txt'")"
+
+# --- Sync trees (TEMPLATE_SYNC_TREES + lw_manifest_tree_files, #90) --------
+# Membership is resolved at run time, so the invariants here are about the
+# declarations and the accessor, not a file list.
+assert_eq "TEMPLATE_SYNC_TREES is non-empty" "1" \
+    "$(lw_mcall "[ \${#TEMPLATE_SYNC_TREES[@]} -gt 0 ] && echo 1 || echo 0")"
+
+# Every declared tree exists as a directory in the template checkout.
+_missing_trees="$(lw_mcall "
+    for t in \"\${TEMPLATE_SYNC_TREES[@]}\"; do
+        [[ -d '$REPO_ROOT_LIB'/\$t ]] || echo \"\$t\"
+    done")"
+assert "every sync tree exists in the template checkout" "[ -z \"$_missing_trees\" ]"
+
+# No static entry may live under a declared tree: that would be double
+# ownership (the tree resolver and the static list would both sync it).
+_tree_overlap="$(lw_mcall "
+    for t in \"\${TEMPLATE_SYNC_TREES[@]}\"; do
+        for f in \"\${TEMPLATE_SHARED_INFRA[@]}\" \"\${TEMPLATE_OVERLAY_CLAUDE[@]}\" \"\${TEMPLATE_OVERLAY_CURSOR[@]}\"; do
+            case \"\$f\" in \"\$t\"/*) echo \"\$f\";; esac
+        done
+    done")"
+assert "no static manifest entry lives under a sync tree" "[ -z \"$_tree_overlap\" ]"
+
+# The dir-mode accessor enumerates real members from a checkout.
+assert "tree_files dir-mode lists the harness runner" \
+    "lw_mcall \"lw_manifest_tree_files dir '$REPO_ROOT_LIB'\" | grep -qx 'scripts/test/run.sh'"
+assert "tree_files dir-mode returns a substantial member set" \
+    "[ \"$(lw_mcall "lw_manifest_tree_files dir '$REPO_ROOT_LIB'" | wc -l | tr -d ' ')\" -gt 50 ]"

scripts/update-from-template.sh

@@ -144,6 +144,11 @@ FILES=()
 while IFS= read -r _path; do
     FILES+=("$_path")
 done < <(lw_manifest_assemble_active_files "$REPO_ROOT" "")
+# Sync-tree members (TEMPLATE_SYNC_TREES, e.g. the test harness) resolved
+# from the template ref — membership is never a static list (#90).
+while IFS= read -r _path; do
+    FILES+=("$_path")
+done < <(lw_manifest_tree_files ref "$TEMPLATE_REF")
 
 # --- Diff and apply ---
 CHANGED=()