)]}'
{
  "log": [
    {
      "commit": "592fc79fbfb7cef97b831b96d54a696229a16927",
      "tree": "7c0cb35219c4171554b9d21d0b36774a792e1ac9",
      "parents": [
        "08b22ae2fe1aa2ccb5dad014a72c3aff90650010"
      ],
      "author": {
        "name": "Marc Kupietz",
        "email": "kupietz@ids-mannheim.de",
        "time": "Sun Aug 02 11:45:47 2026 +0200"
      },
      "committer": {
        "name": "Marc Kupietz",
        "email": "kupietz@ids-mannheim.de",
        "time": "Sun Aug 02 11:45:47 2026 +0200"
      },
      "message": "CI: get a backtrace for the crash on MacOS\n\nThe count based half of the test, type 5, which writes the collocation\ndatabase, dies on macos-latest with \"Trace/BPT trap: 5\" after about 15% of the\ntraining, while the word embedding half including the memory mappable\nconversion passes.\n\nIt is not the rocksdb version: the same run passes on Linux against rocksdb\n8.9.1 (ubuntu), 10.2.1 (fedora) and 11.0.4 (alpine, musl), the last one being\nthe version homebrew ships. It also passes with _GLIBCXX_ASSERTIONS, so it is\nnot something the assertions of libstdc++ catch either.\n\nThe job runs the failing command under lldb when it fails, so that the next run\nsays where it traps instead of only that it did.\n\nCo-Authored-By: Claude Opus 5 \u003cnoreply@anthropic.com\u003e\nChange-Id: I109c3928f83240142e668343e843d97f44127c14\n"
    },
    {
      "commit": "08b22ae2fe1aa2ccb5dad014a72c3aff90650010",
      "tree": "8299afd0af6976c4e381a836aaaaad18ec98bfd2",
      "parents": [
        "6d9b00ebc921ccc1df9ddccb3167a4c11e1bcacf"
      ],
      "author": {
        "name": "Marc Kupietz",
        "email": "kupietz@ids-mannheim.de",
        "time": "Sun Aug 02 09:50:17 2026 +0200"
      },
      "committer": {
        "name": "Marc Kupietz",
        "email": "kupietz@ids-mannheim.de",
        "time": "Sun Aug 02 09:50:17 2026 +0200"
      },
      "message": "CI: add a github workflow that also builds on MacOS\n\nBuilds collocatordb and dereko2vec on ubuntu-latest and macos-latest and runs\nthe training test. It also checks that a training writes the memory mappable\nform of the model and that vecs2mmap produces the same thing for a model that\nalready exists.\n\nVerified on Linux by running the steps of the job as they are: both build\nagainst rocksdb 8.9.1 of ubuntu, the test passes, and both files come out with\nthe same size either way. The MacOS job has not run yet.\n\nCo-Authored-By: Claude Opus 5 \u003cnoreply@anthropic.com\u003e\nChange-Id: I21b032b14186cd35057742a8aef98122d0b0aa84\n"
    }
  ]
}