)]}'
{
  "commit": "3856503e07c533df6d8cbd0eda7c507d007f3e27",
  "tree": "dce02cc6a6f964830909e108816830fd8f117816",
  "parents": [
    "ce145d2f07f4d78a27b6aacc04150164796be8c6"
  ],
  "author": {
    "name": "Evan Brown CA",
    "email": "ezb@google.com",
    "time": "Fri Oct 30 11:14:25 2026 -0700"
  },
  "committer": {
    "name": "Copybara-Service",
    "email": "copybara-worker@google.com",
    "time": "Fri Oct 30 12:18:11 2026 -0700"
  },
  "message": "Deflake raw_hash_set_test under HWASan and hashtablez sampling.\n\nIn Table.InvalidReferenceUseCrashesWithSanitizers, move the pointer capture and rehash-triggering insertion inside the EXPECT_DEATH_IF_SUPPORTED statement block. Previously, t.insert(++i) freed the old backing array before EXPECT_DEATH_IF_SUPPORTED set up its death-test matcher and runner objects; under HWASan (which has no quarantine and 6-bit random pointer tags), those GoogleTest allocations could reuse the just-freed backing array from the LIFO freelist and collide with the old pointer\u0027s 6-bit tag with probability 1/64, causing the dereference not to crash.\n\nAlso call DisableSampling() in ExtendedSooTest.ReserveTwice, SooTest.RehashZeroForSmallTable, SooTest.RangeConstructorReservation, and SooTest.IteratorInvalidAssertsEqualityOperator, which assume 1-element SOO tables remain in SOO mode rather than being sampled onto the heap.\n\nPiperOrigin-RevId: 1010654606\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "4c33b4f8cdb6e162ff00b06704fccd6d14a7b51b",
      "old_mode": 33188,
      "old_path": "absl/container/internal/raw_hash_set_test.cc",
      "new_id": "91fb65fdc26b03a982499e16712966a0bd757b7b",
      "new_mode": 33188,
      "new_path": "absl/container/internal/raw_hash_set_test.cc"
    }
  ]
}