)]}'
{
  "commit": "f267e897930e2f50f06c1b6c60d7027972b381c7",
  "tree": "aaba01ff5c8543d8cecc09401e750eca6c7b52d6",
  "parents": [
    "b17357363c1c6c28ec22b1107d42052312222fad"
  ],
  "author": {
    "name": "Matt Arsenault",
    "email": "Matthew.Arsenault@amd.com",
    "time": "Tue Nov 03 19:49:41 2026 -0800"
  },
  "committer": {
    "name": "Copybara-Service",
    "email": "copybara-worker@google.com",
    "time": "Tue Nov 03 19:52:10 2026 -0800"
  },
  "message": "libunwind: Use libgcc\u0027s function context layout for sjlj (#238128)\n\nMake call_site an i32 and the __data words pointer\nsized. compiler-rt\u0027s SjLj_Function_Context was not defined in the same\nway as libgcc.\n\nThis was broken for all 64-bit targets. Every field was i32, so the\nexception pointer in __data[0] was truncated. It just happens that no\n64-bit target defaults to SJLJ exceptions except VE. Instead of fixing\nthat, a VE special case was introduced in the libunwind sources and\nSjLjEHPrepare, which had a special TargetMachine hook for this data\nsize. In reality it should just be a uintptr_t which we can already\ninfer from the DataLayout.\n\nThis is an ABI break for 64-bit targets other than VE when using\nlibunwind built for SjLj: the __data, personality, lsda and jbuf\noffsets move, so compiled code and libunwind must be updated\ntogether. That configuration was already broken by the pointer\ntruncation and is not the default anywhere. 32-bit targets are\nunchanged, and on VE only call_site narrows; all offsets are\nunchanged.\n\nCo-authored-by: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: 46eda9fcb393f6663ba7acac6a301ef4cc7331cf\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "a3551b328950d13022fa3bbd31b9361ce9acd681",
      "old_mode": 33188,
      "old_path": "src/Unwind-sjlj.c",
      "new_id": "fe39eca39458641d61340fdce597c8761e0c2db7",
      "new_mode": 33188,
      "new_path": "src/Unwind-sjlj.c"
    }
  ]
}