)]}'
{
  "commit": "c4333a12708a917d1cfb6418c04be45793ecc392",
  "tree": "27202db3ca633167b3d6a7656e29520651dd999e",
  "parents": [
    "50f82d062b3b59a09d6b63648df22d46f7f2b010"
  ],
  "author": {
    "name": "Sergey B Kirpichev",
    "email": "skirpichev@gmail.com",
    "time": "Tue Mar 10 16:40:51 2026"
  },
  "committer": {
    "name": "GitHub",
    "email": "noreply@github.com",
    "time": "Tue Mar 10 16:40:51 2026"
  },
  "message": "gh-144173: fix flaky test_complex.test_truediv() (#144355)\n\nPreviously, component-wise relative error bound was tested.  However,\nsuch bound can\u0027t exist already for complex multiplication as one can be\nused to perform subtraction of floating-point numbers, e.g. x and y for\nz0\u003d1+1j and z1\u003dx+yj.\n\n```pycon\n\u003e\u003e\u003e x, y \u003d 1e-9+1j, 1+1j\n\u003e\u003e\u003e a \u003d x*y*y.conjugate()/2;a\n(1.0000000272292198e-09+1j)\n\u003e\u003e\u003e b \u003d x*(y*y.conjugate()/2);b\n(1e-09+1j)\n\u003e\u003e\u003e b \u003d\u003d x\nTrue\n\u003e\u003e\u003e (a.real-b.real)/math.ulp(b.real)\n131672427.0\n```",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "0c7e7341f13d4ea12e0f3507fe329ddc8540e666",
      "old_mode": 33188,
      "old_path": "Lib/test/test_complex.py",
      "new_id": "bee2aceb1870276f6660e1080c3eabab26917f2f",
      "new_mode": 33188,
      "new_path": "Lib/test/test_complex.py"
    }
  ]
}