vvzvlad/gitmost
1
0
Fork 0
Code Issues 37 Pull Requests 20 Packages Projects Releases Activity

fix(ci): дедуп ночного фаззера по хэшу контрпримера, а не по префиксу заголовка

Browse Source 
Ночной property-фаззер при находке контрпримера заводил/обновлял issue,
дедуплицируя по ПРЕФИКСУ ЗАГОЛОВКА. Из-за этого при уже открытом issue по
багу A другой баг B с тем же префиксом заголовка считался дубликатом и
молча терялся до закрытия первого issue — реальные вторые баги глотались.

Теперь дедуп идёт по стабильному короткому хэшу самого контрпримера:
- из вывода fast-check извлекается блок «Counterexample:» (минимальный
  падающий вход) до строки «Shrunk N time(s)»/«Got error»; сид, path и
  счётчик усадки в хэш НЕ входят, поэтому один и тот же баг с разными
  сидами даёт один хэш;
- sha256, первые 12 hex-символов, кладутся в заголовок и в
  машиночитаемый маркер тела `<!-- counterexample-hash: ... -->`;
- поиск открытого issue матчит этот хэш в заголовке или маркер в теле.

Итог: два РАЗНЫХ контрпримера дают два РАЗНЫХ issue, а повторная находка
ТОГО ЖЕ контрпримера по-прежнему схлопывается в существующий.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
agent_coder
2026-07-11 18:53:33 +03:00
parent 185f184989
commit bf087c0a10
1 changed files with 41 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/nightly-property.yml
+41 -10
View File
@@ -124,9 +124,17 @@ jobs:
exit "$FAILED"
# A GENUINE counterexample: fast-check printed a shrunk minimal case and its
# reproducing seed into property-output.txt. File a dedup-guarded issue whose
# title prefix is UNIQUE to counterexamples, so an infra failure (handled by
# the next step under a different title) can never poison this dedup.
# reproducing seed into property-output.txt. File a dedup-guarded issue.
#
# Dedup is keyed on a HASH of the SHRUNK COUNTEREXAMPLE (the minimal failing
# input), NOT on the issue title prefix. Keying on the prefix would let a
# single open issue swallow every OTHER counterexample (a different bug B whose
# title shares the prefix would be treated as a duplicate and stay silent until
# the first issue is closed). Hashing the shrunk example instead means two
# DIFFERENT counterexamples get two DIFFERENT issues, while a re-find of the
# SAME counterexample still dedupes onto the existing one. The infra-failure
# step (below) still keys on its own distinct title, so it can never poison
# this dedup either.
- name: File counterexample issue
# always() is REQUIRED: the fuzz step exits nonzero on a failing shard,
# so a bare `if:` (implicitly success() && ...) would skip this step
@@ -146,25 +154,48 @@ jobs:
echo "No fast-check counterexample signature — infra failure, handled by the next step."
exit 0
fi
TITLE="${TITLE_PREFIX} (seed=${FAIL_SEED})"
# Extract the SHRUNK counterexample block: the "Counterexample:" line(s)
# up to (but excluding) the "Shrunk N time(s)" / "Got error" line. This is
# the minimal failing INPUT and is STABLE across the different seeds/paths
# that reach the same bug — unlike the seed, path, or shrink count (which
# precede/follow this block and vary run-to-run) and unlike the whole
# output (which embeds those varying parts). Hashing THIS is what makes the
# dedup identity the bug itself rather than an incidental run detail.
CE_TEXT=$(awk '/Counterexample:/{c=1} /Shrunk [0-9]+ time|Got error/{c=0} c{print}' property-output.txt)
if [ -z "$CE_TEXT" ]; then
# No parseable shrunk block (unexpected — the signature check above
# already confirmed fast-check output). Fall back to the reproducing
# seed so we still emit a stable identity instead of silently deduping.
CE_TEXT="seed:${FAIL_SEED}"
fi
# Stable short id: first 12 hex chars of sha256 over the counterexample.
CE_HASH=$(printf '%s' "$CE_TEXT" | sha256sum | cut -c1-12)
# Machine-readable marker embedded in the issue body; the open-issue search
# below matches on it (and on the hash in the title) so identity travels
# with the issue regardless of any human title edits.
CE_MARKER="<!-- counterexample-hash: ${CE_HASH} -->"
export CE_HASH CE_MARKER
TITLE="${TITLE_PREFIX} [${CE_HASH}] (seed=${FAIL_SEED})"
# Best-effort dedup: skip if an open issue with the counterexample title
# prefix already exists. A failure of this check must NOT block creation.
# Dedup on the counterexample hash: skip only if an OPEN issue already
# carries this exact hash (in its title or its body marker). A different
# counterexample has a different hash and is NOT deduped. A failure of this
# check must NOT block creation.
EXISTING=""
if EXISTING=$(curl -sS \
-H "Authorization: token ${GITHUB_TOKEN}" \
"${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/issues?state=open&limit=100"); then
if printf '%s' "$EXISTING" \
| node -e 'let s="";process.stdin.on("data",d=>s+=d).on("end",()=>{let a;try{a=JSON.parse(s)}catch{process.exit(1)}if(!Array.isArray(a))process.exit(1);const p=process.env.TITLE_PREFIX;process.exit(a.some(i=>typeof i.title==="string"&&i.title.startsWith(p))?0:1)})'; then
echo "An open '${TITLE_PREFIX}' issue already exists — skipping creation."
| node -e 'let s="";process.stdin.on("data",d=>s+=d).on("end",()=>{let a;try{a=JSON.parse(s)}catch{process.exit(1)}if(!Array.isArray(a))process.exit(1);const h=process.env.CE_HASH,m=process.env.CE_MARKER;process.exit(a.some(i=>(typeof i.title==="string"&&i.title.includes(h))||(typeof i.body==="string"&&i.body.includes(m)))?0:1)})'; then
echo "An open issue for counterexample ${CE_HASH} already exists — skipping creation."
exit 0
fi
fi
# Build the JSON body with the test output SAFELY escaped (never hand-
# interpolate the counterexample into JSON).
BODY_TEXT=$(printf 'A nightly property fuzz SHARD failed with a fast-check counterexample.\n\n- failing shard seed: `%s`\n- NUM_RUNS (per shard): `%s`\n- run: %s\n\nReproduce locally:\n\n```\nPROPERTY_SEED=%s PROPERTY_NUM_RUNS=%s pnpm --filter @docmost/prosemirror-markdown exec vitest run test/generative/\n```\n\nfast-check shrinks the failure to a minimal counterexample. Commit it as a permanent fixture under `packages/prosemirror-markdown/test/fixtures/counterexamples/` + a case in `counterexamples.test.ts`, then FIX the converter (do not weaken a property). See `packages/prosemirror-markdown/README.md`.\n\nTail of the test output (contains the shrunk counterexample):\n\n```\n%s\n```\n' \
"$FAIL_SEED" "$NUM_RUNS" "$RUN_URL" "$FAIL_SEED" "$NUM_RUNS" "$(tail -n 120 property-output.txt)")
BODY_TEXT=$(printf 'A nightly property fuzz SHARD failed with a fast-check counterexample.\n\n- counterexample hash: `%s`\n- failing shard seed: `%s`\n- NUM_RUNS (per shard): `%s`\n- run: %s\n\nReproduce locally:\n\n```\nPROPERTY_SEED=%s PROPERTY_NUM_RUNS=%s pnpm --filter @docmost/prosemirror-markdown exec vitest run test/generative/\n```\n\nfast-check shrinks the failure to a minimal counterexample. Commit it as a permanent fixture under `packages/prosemirror-markdown/test/fixtures/counterexamples/` + a case in `counterexamples.test.ts`, then FIX the converter (do not weaken a property). See `packages/prosemirror-markdown/README.md`.\n\nTail of the test output (contains the shrunk counterexample):\n\n```\n%s\n```\n\n%s\n' \
"$CE_HASH" "$FAIL_SEED" "$NUM_RUNS" "$RUN_URL" "$FAIL_SEED" "$NUM_RUNS" "$(tail -n 120 property-output.txt)" "$CE_MARKER")
jq -n --arg title "$TITLE" --arg body "$BODY_TEXT" \
'{title: $title, body: $body}' > payload.json