On Mon, Feb 3, 2014 at 12:24 PM, Devin Jeanpierre <jeanpierr...@gmail.com> wrote: > On Sun, Feb 2, 2014 at 4:07 PM, Chris Angelico <ros...@gmail.com> wrote: >> On Mon, Feb 3, 2014 at 10:40 AM, Roy Smith <r...@panix.com> wrote: >>> I'm reasonably sure you posted this as humor, but there is some truth in >>> what you said. In the crypto/security domain, you often want to keep a >>> key or cleartext around only for the time it's needed, and scrub the >>> memory it was occupying as soon as it is no longer in use. >>> >>> I don't know how you would do that in Python. >> >> I did, but you're right. >> >> It's fundamentally not possible in pure Python, because there's no way >> to flag a block of memory as "do not page this to disk". For what >> you're talking about to be at all possible, you would need support >> from the language, from the OS, > > mlock on linux, VirtualAlloc (?) on windows. This can be done in > CPython after the fact, but you'd want the memory to be unpageable > before it has contents put in it, not after.
Yeah, like I said it needs language and OS support. There's no way to call those APIs from pure Python code. (And I'm not sure whether those functions work without appropriate CPU-level support; for instance, what happens if your OS is running inside a VM?) > Destroying memory is comparatively easy, as you say -- just make the > object's internal state "invalid", rather than adding anything to the > language. Yeah. Works fine if you have a cooperative system that checks that state every time; otherwise, it'd need at least some language support (the goal is to have its destructor called and all its references wiped, so resources get released). ChrisA -- https://mail.python.org/mailman/listinfo/python-list
