On Thu, Jan 14, 2016 at 3:03 AM, Albert-Jan Roskam <sjeik_ap...@hotmail.com> wrote: > > These two pages are quite nice. The author says the memory used by small > objects is > never returned to the OS, which may be problematic for long running processes.
The article by Evan Jones discusses a patch to enable releasing unused arenas (i.e. "how the problem was fixed"). Starting with 2.5, unused arenas do get released back to the heap. Here's the diff in which Tim Peters merged in a "heavily altered derivative" of Evan's patch [1]. Also, 2.7 and 3.3 bypass C malloc/free and the process heap to instead use mmap/munmap on POSIX when available. This avoids the heap high-water mark problem. Similarly, 3.4 switched to using VirtualAlloc/VirtualFree on Windows. 3.4 also introduced the PyObjectArenaAllocator and associated C API functions [2] to allow modifying the default allocators. [1]: https://hg.python.org/cpython/diff/685849bd905c/Objects/obmalloc.c [2]: https://docs.python.org/3/c-api/memory.html#customize-pyobject-arena-allocator _______________________________________________ Tutor maillist - Tutor@python.org To unsubscribe or change subscription options: https://mail.python.org/mailman/listinfo/tutor
