New submission from INADA Naoki: add_methods(), add_members(), and add_getset() creates PyUnicode from C string 3 times, and calls PyUnicode_InternInplace 2 times.
1. PyDict_GetItemString() at first. (PyUnicode_FromString() is called). 2. In middle, descr_new() calls PyUnicode_InternFromString(). 3. PyDict_SetItemString() at last. (creates unicode and intern it). Skipping (2) is require adding new private APIs to pass PyUnicodeObject. But I don't think it worth enough. (I'll try it later.) So this patch only remove last temporary unicode. (3 PyUnicode_FromString + 2 PyUnicode_InternInplace) becomes (2 PyUnicode_FromString + 2 PyUnicode_InternInplace). It seems ~1% startup speedup (without site). $ ./python -m performance.benchmarks.bm_python_startup --no-site default: python_startup_no_site: Median +- std dev: 12.7 ms +- 0.1 ms patched: python_startup_no_site: Median +- std dev: 12.6 ms +- 0.1 ms While speedup is small, this patch removes time to think "How large this overhead of GetItemString + SetItemString pair?" while reading code :) Additionally, this patch removes this comment in PyDict_SetItemString: - PyUnicode_InternInPlace(&kv); /* XXX Should we really? */ SetItemString is used to add something to namespace. Changing this behavior affects too widely. So we should do it. ---------- assignee: inada.naoki components: Interpreter Core files: descr-remove-getitemstring.patch keywords: patch messages: 286394 nosy: inada.naoki priority: normal severity: normal status: open title: Reduce temporary unicode object while adding descriptors versions: Python 3.7 Added file: http://bugs.python.org/file46441/descr-remove-getitemstring.patch _______________________________________ Python tracker <rep...@bugs.python.org> <http://bugs.python.org/issue29383> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
