feat(flow-010): affected_set.sh diff to crates, reverse-dep closure, nextest filter (E5)

ci/explicit-test-commands.d/815-affected-set-rows.cmd

@@ -0,0 +1,2 @@
+# FLOW-010 E5: affected-set tool rows (leaf+dependents vs cargo tree oracle, core/Cargo.lock/unknown -> ALL).
+bash scripts/check_affected_set.sh

ci/sections.yml

@@ -1951,6 +1951,11 @@ jobs:
         run: setsid --wait bash scripts/check_lockfile_current.sh
       - name: "lockfile guard must still turn RED (case table)"
         run: setsid --wait bash scripts/check_lockfile_current.sh --self-test
+      # FLOW-010 E5: runs cargo metadata/tree, so it lives in the cargo job, not guard-tree-steps.
+      - name: "Affected-set tool case table (FLOW-010 E5, RED rows and mutants)"
+        run: setsid --wait bash scripts/check_affected_set.sh
+      - name: "Affected-set readers guard (every crate that reads other crates' files is listed; RED on a new one)"
+        run: setsid --wait bash scripts/check_affected_readers.sh && setsid --wait bash scripts/check_affected_readers.sh --self-test
       # #2640: the SAME defect class as the two steps above, one level up. The
       # dogfood release runner existed as two 1170-line copies whose divergence
       # was, in every hunk, a pin for a release verifier that one copy had and

scripts/affected_set.sh

@@ -0,0 +1,77 @@
+#!/usr/bin/env bash
+# FLOW-010 E5: the diff -> the set of workspace packages whose tests it can change.
+#
+#   scripts/affected_set.sh [--filter] [--diff-from FILE] [BASE [HEAD]]
+#
+# Prints one package per line, or the single word ALL. With --filter it prints the
+# nextest filterset instead (`package(a) | package(b)`, or `all()` for ALL).
+# Fail-safe: anything this tool cannot place in a package is ALL.
+#   crates/<pkg>/...        the package + its reverse-dependency closure (cargo metadata)
+#   a CORE crate            ALL (its closure is the workspace; naming it would only hide that)
+#   Cargo.toml of a crate   ALL (a manifest edit changes features/deps, which no path can bound)
+#   docs/, root *.md, ci/ (sections.yml, explicit-test-commands.d/, ...), Cargo.lock, root Cargo.toml,
+#   rust-toolchain.toml, .cargo/, .github/, Makefile, any other path outside a crate
+#                           ALL (tests read README.md, CLAUDE.md, ci/sections.yml and docs/ at run time)
+# Any non-.rs file under crates/<pkg>/ (README.md, *.json, *.toml, ...) is ALL: other crates' tests read them
+# (aprender-core readme_contract.rs reads every crates/*/README.md; aprender-contracts reads aprender-serve/kernel-registry.json).
+# An empty diff, or a diff git cannot compute (bad base: exit 128), is ALL: nothing was
+# measured, so nothing may be skipped.
+# git diff runs with --no-renames: a moved file is its old path AND its new path (a rename
+# out of a crate would otherwise show only the destination and the source crate would be skipped).
+# --diff-from reads the changed paths from FILE (tests). AFFECTED_META / AFFECTED_ROOT
+# replace `cargo metadata` and the checkout root (tests; synthetic graphs).
+set -euo pipefail
+# CORE = ALL on touch. apr-cli is here because other crates' tests read its sources (aprender-core readme_contract.rs reads
+# apr-cli/src/commands_enum.rs; spec_checklist_*, toyota_principles_tests). Reader edges are not in cargo metadata, so a
+# crate whose source another crate reads at run time must be listed (aprender-quant: aprender-serve/capability.rs reads
+# aprender-quant/src/ggml_type.rs). Any crates/*/build.rs is ALL: contracts/build_helper.rs reads crates/*/build.rs.
+CORE="${AFFECTED_CORE:-aprender-core aprender-compute aprender-contracts apr-cli aprender-quant}"
+# READERS = crates whose tests read OTHER crates' files at run time (CARGO_MANIFEST_DIR/.., crates/, ../<crate>). CORE ones are ALL on touch, but that does not put them in a non-ALL set; every non-ALL set includes them, because the touched crate may be what they read. scripts/check_affected_readers.sh
+# goes RED on a reader this list (or CORE) lacks.
+READERS="${AFFECTED_READERS:-apr-cli apr-format aprender-build-sha aprender-cbtop aprender-cgp aprender-ci-tools aprender-compute aprender-compute-xtask aprender-contracts aprender-contracts-cli aprender-contrastive-data aprender-core aprender-data aprender-db aprender-decide aprender-distribute aprender-explain aprender-gpu aprender-mcp aprender-orchestrate aprender-present-cli aprender-present-core aprender-present-lib aprender-present-terminal aprender-present-yaml aprender-profile aprender-qa-certify aprender-qa-cli aprender-qa-gen aprender-qa-report aprender-qa-runner aprender-quant aprender-rag aprender-registry aprender-review-experiment aprender-rollup aprender-serve aprender-shell aprender-simulate aprender-test-cli aprender-test-lib aprender-train aprender-train-distill aprender-train-inspect aprender-train-shell aprender-tsp aprender-update aprender-verify-ml aprender-viz aprender-viz-ttop}"
+filter=0 from=""
+while [ $# -gt 0 ]; do
+  case "$1" in
+    --filter) filter=1; shift ;;
+    --diff-from) from="$2"; shift 2 ;;
+    *) break ;;
+  esac
+done
+case "$from" in *..*) echo "affected_set: refusing a --diff-from with \"..\": $from" >&2; exit 2 ;; esac   # bashrs SEC010
+if [ -n "$from" ]; then files="$(cat "$from")"
+else files="$(git diff --name-only --no-renames "${1:-origin/main}...${2:-HEAD}")" || files=""; fi
+emit() { # $@ = package names, or ALL
+  if [ "${1:-}" = ALL ]; then [ "$filter" = 1 ] && echo 'all()' || echo ALL; return; fi
+  if [ "$filter" = 1 ]; then printf '%s\n' "$@" | sed 's/.*/package(&)/' | paste -sd'|' | sed 's/|/ | /g'; else printf '%s\n' "$@"; fi
+}
+[ -n "$files" ] || { emit ALL; exit 0; }
+if [ -n "${AFFECTED_META:-}" ]; then meta="$(cat "$AFFECTED_META")"
+else meta="$(cargo metadata --no-deps --format-version 1 --offline 2>/dev/null || cargo metadata --no-deps --format-version 1)"; fi
+root="${AFFECTED_ROOT:-$(git rev-parse --show-toplevel)}"
+dirs="$(jq -r --arg root "$root/" '.packages[] | [.name, (.manifest_path | sub($root;"") | sub("/?Cargo.toml$";""))] | @tsv' <<<"$meta")" || { emit ALL; exit 0; }
+[ -n "$dirs" ] || { emit ALL; exit 0; }
+touched=""
+while IFS= read -r f; do
+  case "$f" in
+    '') continue ;;
+    crates/*/build.rs) emit ALL; exit 0 ;;
+    crates/*.rs) ;;
+    crates/*) emit ALL; exit 0 ;;
+  esac
+  # longest manifest dir that prefixes the path
+  pkg="$(awk -F'\t' -v f="$f" 'index(f, $2 "/") == 1 && length($2) > best { best = length($2); n = $1 } END { print n }' <<<"$dirs")"
+  [ -n "$pkg" ] || { emit ALL; exit 0; }
+  case " $CORE " in *" $pkg "*) emit ALL; exit 0 ;; esac
+  touched="$touched"$'\n'"$pkg"
+done <<<"$files"
+touched="$(printf '%s\n' "$touched" | grep -v '^$' | sort -u || true)"
+[ -n "$touched" ] || { emit ALL; exit 0; }   # nothing placed in a package: nothing may be skipped
+# The closure, then aprender-core: its falsification_spec_v10 test scans every crate's src, so ANY non-ALL set includes it (and the READERS).
+res="$(jq -r --argjson t "$(printf '%s\n' "$touched" | jq -R . | jq -s .)" --arg rd "$READERS" '
+  (.packages | map({name, deps: [.dependencies[].name]})) as $p
+  | def grow($s): ($s + [$p|.[] | select(any(.deps[]; . as $d | $s | index($d))) | .name] | unique);
+    (.packages | map(.name)) as $m
+    | (($t | until(grow(.) == .; grow(.))) + ($rd | split(" ")) | unique) | .[] | select(. as $n | $m | index($n))' <<<"$meta")" || { emit ALL; exit 0; }
+[ -n "$res" ] || { emit ALL; exit 0; }
+mapfile -t out <<<"$res"
+emit "${out[@]}"

scripts/check_affected_readers.sh

@@ -0,0 +1,88 @@
+#!/usr/bin/env bash
+# FLOW-010 E5: a crate whose Rust can reach OTHER crates' files at run time is a reader edge that cargo metadata
+# cannot see. scripts/affected_set.sh must list it in READERS (the only list that puts a crate in a non-ALL set),
+# or a change to the crate it reads would skip its tests. RED on a crate with any path-escape form and no entry.
+# Over-approximates on purpose: a false positive only adds a crate to READERS (more tests run); a miss skips tests.
+# Flagged in any .rs: a string holding `..`, parent(), ancestors(), current_dir, a "crates/" string or "crates",
+# CARGO_WORKSPACE_DIR, workspace_path_or_skip, read_dir(...) naming MANIFEST/crates/root/workspace, format!(...)
+# naming MANIFEST_DIR or crates/. Comment-only lines and files under a target/ path component are skipped.
+# Usage: check_affected_readers.sh [crates-dir] | --self-test | --list
+set -euo pipefail
+SET="${AFFECTED_SET_SH:-scripts/affected_set.sh}"
+# the exact file:line exemptions (a crate-wide exemption hides a later reader in the same crate)
+ALLOW="${AFFECTED_READERS_ALLOW:-crates/aprender-bench-tokenizer/src/lib.rs:50}"   # joins ../.. only for <root>/tokenizer.json and <root>/models/
+# crates/ dirs outside the workspace (excluded): a change there is unplaceable -> ALL, so they need no READERS entry
+NONMEMBER="${AFFECTED_NONMEMBER:-facades}"
+PAT='"\.\."|"[^"]*(\.\./|/\.\.)[^"]*"|\bparent[[:space:]]*\(|Path::parent\(|Component::ParentDir|show-toplevel|\.pop\(\);[^;]*\.push\(|ancestors\(\)|current_dir|"[^"]*crates/|"crates"|CARGO_WORKSPACE_DIR|workspace_path_or_skip|read_dir\([^)]*(MANIFEST|crates|root|workspace)|format!\([^)]*(MANIFEST_DIR|crates/)'
+listed() { sed -n 's/^READERS="${AFFECTED_READERS:-\(.*\)}"$/\1/p' "$SET" | tr ' ' '\n' | grep -v '^$' | sort -u; }
+hits() { # $1 = crates dir; prints crate names
+  { grep -rEn --include='*.rs' "$PAT" "$1" 2>/dev/null || true
+    # file-level: CARGO_MANIFEST_DIR and .pop( anywhere in one file (rustfmt may wrap the let or split pop/push across lines)
+    for f in $(grep -rlE --include='*.rs' 'CARGO_MANIFEST_DIR' "$1" 2>/dev/null || true); do
+      grep -qE '\.pop[[:space:]]*\(' "$f" && echo "$f:0:pop-with-MANIFEST_DIR"
+    done; } \
+    | grep -vE '^[^:]*/target/' | grep -vE '^[^:]+:[0-9]+:[[:space:]]*//' \
+    | awk -F: -v allow=" $ALLOW " -v nm=" $NONMEMBER " '{ k = $1 ":" $2; if (index(allow, " " k " ")) next; n = split($1, a, "/"); if (index(nm, " " a[2] " ")) next; print a[2] }' | sort -u
+}
+nonmember_in_workspace() { # a NONMEMBER dir that cargo metadata lists is a member: its exemption would hide a reader
+  local m d
+  m="$({ if [ -n "${AFFECTED_META_JSON:-}" ]; then cat "$AFFECTED_META_JSON"; else cargo metadata --no-deps --format-version 1 --offline 2>/dev/null; fi; } || true)"
+  [ -n "$m" ] || return 0
+  for d in $NONMEMBER; do
+    if printf '%s' "$m" | grep -qF "/crates/$d/"; then echo "RED: NONMEMBER dir crates/$d is a workspace member (cargo metadata lists it): drop it from NONMEMBER"; return 1; fi
+  done
+}
+check() {
+  nonmember_in_workspace || return 1
+  local bad="" l c; l="$(listed)"
+  for c in $(hits "$1"); do printf '%s\n' "$l" | grep -qx "$c" || bad="$bad $c"; done
+  if [ -n "$bad" ]; then echo "RED: crates can reach other crates' files but are not in READERS (scripts/affected_set.sh):$bad"; return 1; fi
+  echo "ok: every crate with a path-escape form is in READERS"
+}
+if [ "${1:-}" = "--list" ]; then hits crates; exit 0; fi
+if [ "${1:-}" = "--self-test" ]; then
+  t=$(mktemp -d); trap 'rm -rf -- "${t:?}"' EXIT
+  SET="$(pwd)/$SET"; cd "$t"
+  fail=0
+  fresh() { rm -rf -- "${t:?}/crates"; mkdir -p "$t/crates/$1/src"; }   # one unlisted crate holding $2
+  expect_red() { # name crate body
+    fresh "$2"; printf '%s\n' "fn f() { $3 }" >"$t/crates/$2/src/x.rs"
+    if (check crates >/dev/null 2>&1); then echo "FAIL self-test: not caught: $1"; fail=1; fi
+  }
+  expect_red 'join("crates")' zz-a 'let _ = root.join("crates");'
+  expect_red 'MANIFEST_DIR.parent().parent()' zz-b 'let _ = Path::new(env!("CARGO_MANIFEST_DIR")).parent().unwrap().parent();'
+  expect_red 'root.join("crates/aprender-core")' zz-c 'let _ = root.join("crates/aprender-core");'
+  expect_red 'format! with ../../crates' zz-d 'let _ = format!("{}/../../crates/x", env!("CARGO_MANIFEST_DIR"));'
+  expect_red 'ancestors().nth(2)' zz-e 'let _ = here.ancestors().nth(2);'
+  expect_red 'read_dir("../")' zz-f 'let _ = std::fs::read_dir("../");'
+  expect_red 'current_dir().join("..")' zz-g 'let _ = std::env::current_dir().unwrap().join("..");'
+  expect_red 'relative read_to_string' zz-h 'let _ = std::fs::read_to_string("../../crates/x/src/lib.rs");'
+  expect_red 'MANIFEST_DIR.parent().join(apr-cli)' zz-i 'let _ = Path::new(env!("CARGO_MANIFEST_DIR")).parent().unwrap().join("apr-cli");'
+  expect_red 'workspace_path_or_skip!' zz-j 'let _ = workspace_path_or_skip!("crates/x/..");'
+  expect_red 'read_dir(workspace_root)' zz-k 'let _ = std::fs::read_dir(workspace_root);'
+  expect_red 'CARGO_WORKSPACE_DIR' zz-l 'let _ = env!("CARGO_WORKSPACE_DIR");'
+  expect_red 'format! with MANIFEST_DIR' zz-m 'let _ = format!("{}/x", env!("CARGO_MANIFEST_DIR"));'
+  expect_red 'a *contracts*/*target* crate name' zz-contracts-target-new 'let _ = root.join("crates");'
+  expect_red 'a line that mentions target/contracts' zz-n 'let _ = root.join("crates"); /* target contracts docs/ */'
+  expect_red 'PathBuf pop then push' zz-p 'let mut p = PathBuf::from(env!("CARGO_MANIFEST_DIR")); p.pop(); p.push("aprender-core");'
+  expect_red 'Component::ParentDir' zz-q 'let _ = Component::ParentDir;'
+  expect_red 'git rev-parse --show-toplevel' zz-r 'let _ = Command::new("git").args(["rev-parse", "--show-toplevel"]);'
+  expect_red 'Path::parent(x)' zz-s 'let _ = Path::parent(&here);'
+  expect_red '.parent ()' zz-t 'let _ = here.parent ();'
+  expect_red 'a non-member dir is only exempt when named' facades-x 'let _ = here.ancestors();'
+  expect_red 'rustfmt-wrapped let + split pop/push' zz-u $'let mut m =\n        PathBuf::from(env!("CARGO_MANIFEST_DIR"));\n    m.pop();\n    m.push("aprender-core");'
+  expect_red 'spaced .pop (' zz-w $'let mut m = PathBuf::from(env!("CARGO_MANIFEST_DIR"));\n    m.pop ();'
+  printf '{"packages":[{"manifest_path":"/x/crates/facades/zz-evil/Cargo.toml"}]}' >"$t/meta.json"; fresh zz-v
+  if (AFFECTED_META_JSON="$t/meta.json" check crates >/dev/null 2>&1); then echo "FAIL self-test: a NONMEMBER dir listed by cargo metadata passed"; fail=1; fi
+  # the bench-tokenizer exemption is the one line, not the crate
+  fresh aprender-bench-tokenizer; printf 'fn f() {}\nfn g() { let _ = root.join("crates"); }\n' >"$t/crates/aprender-bench-tokenizer/src/lib.rs"
+  if (AFFECTED_READERS_ALLOW="crates/aprender-bench-tokenizer/src/lib.rs:1" check crates >/dev/null 2>&1); then echo "FAIL self-test: an allow entry exempted a different line"; fail=1; fi
+  # a listed reader, a comment-only mention and a target/ file are green
+  fresh aprender-serve; echo 'fn f() { let _ = root.join("crates"); }' >"$t/crates/aprender-serve/src/y.rs"
+  (check crates >/dev/null 2>&1) || { echo "FAIL self-test: a listed reader was flagged"; fail=1; }
+  fresh zz-o; echo '// root.join("crates")' >"$t/crates/zz-o/src/y.rs"
+  (check crates >/dev/null 2>&1) || { echo "FAIL self-test: a comment-only line was flagged"; fail=1; }
+  [ "$fail" -eq 0 ] && echo "ok: self-test"
+  exit "$fail"
+fi
+check "${1:-crates}"