test(#795): enforce PR diff regressions in CI

This commit is contained in:
Hermes Agent
2026-07-17 12:21:35 -05:00
parent 92d6f263b9
commit 462ae8c7e2
2 changed files with 90 additions and 77 deletions

View File

@@ -16,43 +16,43 @@ assert_contains() {
[[ "$haystack" == *"$needle"* ]] || fail "expected output to contain: $needle" [[ "$haystack" == *"$needle"* ]] || fail "expected output to contain: $needle"
} }
# A fresh clone is on main, while PR #795's changes exist only at the provider PR ref. # A fresh clone's local HEAD is unrelated to PR #795, whose changes target the
# The git stub returns the PR patch only when build_diff_context fetches and diffs that # synthetic non-main base release/next and exist only at the provider PR ref.
# explicit ref; diffing main...HEAD reproduces the empty/wrong-branch bug. # The git stub returns the PR patch only for those two explicit fetched refs.
if [[ "${1:-all}" != "fail-closed" ]]; then if [[ "${1:-all}" != "fail-closed" ]]; then
FETCH_LOG="$TMP_DIR/fetch.log" FETCH_LOG="$TMP_DIR/fetch.log"
DIFF_LOG="$TMP_DIR/diff.log" DIFF_LOG="$TMP_DIR/diff.log"
export FETCH_LOG DIFF_LOG export FETCH_LOG DIFF_LOG
( (
# shellcheck source=common.sh # shellcheck source=common.sh
source "$SCRIPT_DIR/common.sh" source "$SCRIPT_DIR/common.sh"
detect_platform() { detect_platform() {
PLATFORM=gitea PLATFORM=gitea
export PLATFORM export PLATFORM
} }
tea() { tea() {
if [[ "$*" == "pr list --fields index,base --output simple" ]]; then if [[ "$*" == "pr list --fields index,base --output simple" ]]; then
printf '795 main\n' printf '795 release/next\n'
return 0 return 0
fi fi
return 1 return 1
} }
git() { git() {
case "${1:-}" in case "${1:-}" in
remote) remote)
printf 'https://git.example.test/mosaicstack/stack.git\n' printf 'https://git.example.test/mosaicstack/stack.git\n'
;; ;;
fetch) fetch)
printf '%s\n' "$*" >> "$FETCH_LOG" printf '%s\n' "$*" >> "$FETCH_LOG"
;; ;;
diff) diff)
printf '%s\n' "$*" >> "$DIFF_LOG" printf '%s\n' "$*" >> "$DIFF_LOG"
if [[ "$*" == *"refs/remotes/origin/pr/795/head"* && "$*" != *"HEAD"* ]]; then if [[ "$*" == *"refs/remotes/origin/release/next...refs/remotes/origin/pr/795/head"* && "$*" != *"HEAD"* ]]; then
cat <<'DIFF' cat <<'DIFF'
diff --git a/pr-change.ts b/pr-change.ts diff --git a/pr-change.ts b/pr-change.ts
--- a/pr-change.ts --- a/pr-change.ts
+++ b/pr-change.ts +++ b/pr-change.ts
@@ -60,45 +60,51 @@ diff --git a/pr-change.ts b/pr-change.ts
-old -old
+actual-pr-change +actual-pr-change
DIFF DIFF
fi fi
;; ;;
*) *)
return 1 return 1
;; ;;
esac esac
} }
diff_context=$(build_diff_context pr 795) diff_context=$(build_diff_context pr 795)
assert_contains "$diff_context" "actual-pr-change" assert_contains "$diff_context" "actual-pr-change"
) )
assert_contains "$(cat "$FETCH_LOG" 2>/dev/null || true)" "refs/pull/795/head" assert_contains "$(cat "$FETCH_LOG" 2>/dev/null || true)" "refs/heads/release/next:refs/remotes/origin/release/next"
assert_contains "$(cat "$DIFF_LOG" 2>/dev/null || true)" "refs/remotes/origin/pr/795/head" assert_contains "$(cat "$FETCH_LOG" 2>/dev/null || true)" "refs/pull/795/head"
if grep -q 'HEAD' "$DIFF_LOG" 2>/dev/null; then assert_contains "$(cat "$DIFF_LOG" 2>/dev/null || true)" "refs/remotes/origin/release/next...refs/remotes/origin/pr/795/head"
fail "Gitea PR diff used local HEAD" if grep -q 'HEAD' "$DIFF_LOG" 2>/dev/null; then
fi fail "Gitea PR diff used local HEAD"
echo "ok - Gitea PR mode fetches and diffs the explicit PR head" fi
echo "ok - Gitea PR mode fetches and diffs explicit non-main base and PR head refs"
fi fi
# Exercise the CLI path with all provider/network mutations stubbed. An empty PR diff # Exercise both CLI paths with all provider/network mutations stubbed. An empty
# must be an actionable nonzero failure before Codex runs and before auto-post can run. # PR diff must fail before Codex runs and before either auto-post path can run.
if [[ "${1:-all}" != "pr-head" ]]; then if [[ "${1:-all}" != "pr-head" ]]; then
SANDBOX="$TMP_DIR/sandbox" SANDBOX="$TMP_DIR/sandbox"
mkdir -p "$SANDBOX/tools/codex/schemas" "$SANDBOX/tools/git" "$SANDBOX/bin" mkdir -p "$SANDBOX/tools/codex/schemas" "$SANDBOX/tools/git" "$SANDBOX/bin"
cp "$SCRIPT_DIR/common.sh" "$SCRIPT_DIR/codex-code-review.sh" "$SANDBOX/tools/codex/" cp "$SCRIPT_DIR/common.sh" \
cp "$SCRIPT_DIR/schemas/code-review-schema.json" "$SANDBOX/tools/codex/schemas/" "$SCRIPT_DIR/codex-code-review.sh" \
"$SCRIPT_DIR/codex-security-review.sh" \
"$SANDBOX/tools/codex/"
cp "$SCRIPT_DIR/schemas/code-review-schema.json" \
"$SCRIPT_DIR/schemas/security-review-schema.json" \
"$SANDBOX/tools/codex/schemas/"
cat > "$SANDBOX/tools/git/detect-platform.sh" <<'STUB' cat > "$SANDBOX/tools/git/detect-platform.sh" <<'STUB'
detect_platform() { PLATFORM=gitea; export PLATFORM; } detect_platform() { PLATFORM=gitea; export PLATFORM; }
get_repo_info() { printf 'mosaicstack/stack\n'; } get_repo_info() { printf 'mosaicstack/stack\n'; }
STUB STUB
cat > "$SANDBOX/tools/git/pr-review.sh" <<'STUB' cat > "$SANDBOX/tools/git/pr-review.sh" <<'STUB'
#!/bin/bash #!/bin/bash
printf 'POST %s\n' "$*" >> "$POST_LOG" printf 'POST %s\n' "$*" >> "$POST_LOG"
STUB STUB
chmod +x "$SANDBOX/tools/git/pr-review.sh" chmod +x "$SANDBOX/tools/git/pr-review.sh"
cat > "$SANDBOX/bin/git" <<'STUB' cat > "$SANDBOX/bin/git" <<'STUB'
#!/bin/bash #!/bin/bash
case "${1:-}" in case "${1:-}" in
rev-parse) exit 0 ;; rev-parse) exit 0 ;;
@@ -108,34 +114,40 @@ case "${1:-}" in
*) exit 1 ;; *) exit 1 ;;
esac esac
STUB STUB
cat > "$SANDBOX/bin/tea" <<'STUB' cat > "$SANDBOX/bin/tea" <<'STUB'
#!/bin/bash #!/bin/bash
if [[ "$*" == "pr list --fields index,base --output simple" ]]; then if [[ "$*" == "pr list --fields index,base --output simple" ]]; then
printf '795 main\n' printf '795 release/next\n'
exit 0 exit 0
fi fi
exit 1 exit 1
STUB STUB
cat > "$SANDBOX/bin/codex" <<'STUB' cat > "$SANDBOX/bin/codex" <<'STUB'
#!/bin/bash #!/bin/bash
printf 'CODEX %s\n' "$*" >> "$CODEX_LOG" printf 'CODEX %s\n' "$*" >> "$CODEX_LOG"
exit 99 exit 99
STUB STUB
chmod +x "$SANDBOX/bin/git" "$SANDBOX/bin/tea" "$SANDBOX/bin/codex" chmod +x "$SANDBOX/bin/git" "$SANDBOX/bin/tea" "$SANDBOX/bin/codex"
POST_LOG="$TMP_DIR/post.log" POST_LOG="$TMP_DIR/post.log"
CODEX_LOG="$TMP_DIR/codex.log" CODEX_LOG="$TMP_DIR/codex.log"
export POST_LOG CODEX_LOG export POST_LOG CODEX_LOG
set +e
PATH="$SANDBOX/bin:$PATH" "$SANDBOX/tools/codex/codex-code-review.sh" -n 795 \
>"$TMP_DIR/review.stdout" 2>"$TMP_DIR/review.stderr"
review_status=$?
set -e
[[ "$review_status" -ne 0 ]] || fail "empty PR diff returned success instead of failing closed" for review_kind in code security; do
[[ ! -s "$CODEX_LOG" ]] || fail "Codex ran for an empty PR diff" : > "$POST_LOG"
[[ ! -s "$POST_LOG" ]] || fail "review auto-post ran for an empty PR diff" : > "$CODEX_LOG"
assert_contains "$(cat "$TMP_DIR/review.stderr")" "Error:" review_script="$SANDBOX/tools/codex/codex-${review_kind}-review.sh"
assert_contains "$(cat "$TMP_DIR/review.stderr")" "PR #795" set +e
echo "ok - empty PR diff fails closed before Codex and auto-post" PATH="$SANDBOX/bin:$PATH" "$review_script" -n 795 \
>"$TMP_DIR/${review_kind}.stdout" 2>"$TMP_DIR/${review_kind}.stderr"
review_status=$?
set -e
[[ "$review_status" -ne 0 ]] || fail "${review_kind} review returned success for an empty PR diff"
[[ ! -s "$CODEX_LOG" ]] || fail "Codex ran for an empty ${review_kind} PR diff"
[[ ! -s "$POST_LOG" ]] || fail "${review_kind} review auto-post ran for an empty PR diff"
assert_contains "$(cat "$TMP_DIR/${review_kind}.stderr")" "Error:"
assert_contains "$(cat "$TMP_DIR/${review_kind}.stderr")" "PR #795"
echo "ok - empty ${review_kind} PR diff fails closed before Codex and auto-post"
done
fi fi

View File

@@ -24,7 +24,8 @@
"build": "tsc", "build": "tsc",
"lint": "eslint src", "lint": "eslint src",
"typecheck": "tsc --noEmit", "typecheck": "tsc --noEmit",
"test": "vitest run --passWithNoTests" "test": "vitest run --passWithNoTests && pnpm run test:framework-shell",
"test:framework-shell": "bash framework/tools/codex/test-pr-diff-context.sh"
}, },
"dependencies": { "dependencies": {
"@mosaicstack/brain": "workspace:*", "@mosaicstack/brain": "workspace:*",