René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Use PEP 669 API for cprofile","articleBody":"# Feature or enhancement\r\n\r\nReplace the current `setprofile` mechanism with PEP 669 API for cProfile.\r\n\r\n# Pitch\r\n\r\nIt's faster.\r\n\r\nBefore:\r\n```\r\nHanoi: 0.342869599997357, 1.4579414000036195, 4.252174587699983\r\nfib: 0.15549560000363272, 0.7588815999988583, 4.88040561907301\r\n\r\nHanoi: 0.33156009999947855, 1.394542599999113, 4.2060024713507635\r\nfib: 0.16397210000286577, 0.7469802000050549, 4.555532313070332\r\n\r\nHanoi: 0.3341937000004691, 1.394005099995411, 4.171248889471747\r\nfib: 0.15704140000161715, 0.7399598000047263, 4.711877250184387\r\n\r\nHanoi: 0.33133889999589883, 1.3821628000005148, 4.171447421409387\r\nfib: 0.15387790000386303, 0.754370700000436, 4.902397940064804\r\n\r\nHanoi: 0.3403002000050037, 1.3969628000049852, 4.105089564991276\r\nfib: 0.15468130000226665, 0.761441399998148, 4.922646758121312\r\n```\r\n\r\nAfter:\r\n```\r\nHanoi: 0.3318088000014541, 1.2147841000041808, 3.661096691826309\r\nfib: 0.15522490000148537, 0.6360437999974238, 4.097562955372092\r\n\r\nHanoi: 0.33085879999998724, 1.1502422000048682, 3.476535005279934\r\nfib: 0.15410060000431258, 0.6056611999956658, 3.9302974808580635\r\n\r\nHanoi: 0.35002540000277804, 1.145935599997756, 3.273864125256799\r\nfib: 0.1540030999967712, 0.5974198000039905, 3.8792712615299036\r\n\r\nHanoi: 0.3338971999983187, 1.1459307999975863, 3.431986851052829\r\nfib: 0.16891020000184653, 0.6197690999979386, 3.6692224625343126\r\n\r\nHanoi: 0.3318254999976489, 1.1875411000000895, 3.578812056362467\r\nfib: 0.1544136999946204, 0.5971600999982911, 3.867274082669449\r\n```\r\n\r\n20%+ speed up for overhead.\r\n\r\nI guess the incentive is not in doubt, but I did have some issues when I implemented it.\r\n\r\n1. There is a very slight backward compatibility issue, or it's more like a design decision. The profiler has to disable its own profiling now.\r\n\r\n```python\r\n# This works on main, but I don't think that's intuitive.\r\npr = cProfile.Profile()\r\npr.enable()\r\npr = cProfile.Profile()\r\npr.disable()\r\n```\r\n\r\nWe can make it work as before, I just think this is not the right way to do it with PEP 669. Because of this, I changed an old (15 years) test in `test_cprofile`.\r\n\r\n2. We need to get the actual c function from the descriptor, for which I used the code in the current legacy tracing. However, `_PyInstrumentation_MISSING` is not exposed so I had to store it locally (keep reading it from `sys` is a bit expensive).\r\n\r\n3. On that matter, are we going to expose some APIs on C? It would be nice if I don't have to get `sys.monitoring` and do stuff from there. We have some defined constants but some APIs could be handy. We may be able to reduce the overhead a bit if we have an interface like `PyEval_SetProfile`.\r\n\r\nAddendum:\r\n\u003cdetails\u003e\r\n\u003csummary\u003e Benchmark Code \u003c/summary\u003e\r\n\r\n```python\r\nimport timeit\r\n\r\nhanoi_setup = \"\"\"\r\nimport cProfile\r\ndef test():\r\n    def TowerOfHanoi(n, source, destination, auxiliary):\r\n        if n == 1:\r\n            return\r\n        TowerOfHanoi(n - 1, source, auxiliary, destination)\r\n        TowerOfHanoi(n - 1, auxiliary, destination, source)\r\n    TowerOfHanoi(16, \"A\", \"B\", \"C\")\r\npr = cProfile.Profile()\r\n\"\"\"\r\n\r\nfib_setup = \"\"\"\r\nimport cProfile\r\ndef test():\r\n    def fib(n):\r\n        if n \u003c= 1:\r\n            return 1\r\n        return fib(n - 1) + fib(n - 2)\r\n    fib(21)\r\npr = cProfile.Profile()\r\n\"\"\"\r\n\r\ntest_baseline = \"\"\"\r\ntest()\r\n\"\"\"\r\n\r\ntest_profile = \"\"\"\r\npr.enable()\r\ntest()\r\npr.disable()\r\n\"\"\"\r\n\r\nbaseline = timeit.timeit(test_baseline, setup=hanoi_setup, number=100)\r\nprofile = timeit.timeit(test_profile, setup=hanoi_setup, number=100)\r\nprint(f\"Hanoi: {baseline}, {profile}, {profile / baseline}\")\r\n\r\nbaseline = timeit.timeit(test_baseline, setup=fib_setup, number=100)\r\nprofile = timeit.timeit(test_profile, setup=fib_setup, number=100)\r\nprint(f\"fib: {baseline}, {profile}, {profile / baseline}\")\r\n```\r\n\u003c/details\u003e\r\n\n\n\u003c!-- gh-linked-prs --\u003e\n### Linked PRs\n* gh-103534\n\u003c!-- /gh-linked-prs --\u003e\n","author":{"url":"https://github.com/gaogaotiantian","@type":"Person","name":"gaogaotiantian"},"datePublished":"2023-04-14T06:30:56.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/103533/cpython/issues/103533"}