[PATCH] store: break up reference cycle introduced in 9cbff8a39a2a

[timeless]

On Thu, May 5, 2011 at 9:28 PM, Sune Foldager <cryo at cyanite.org> wrote:
>> And operating systems will be hostile right back. When you incur a page
>> fault that requires a new page on a memory mapped region under memory
>> pressure, it is _too late_ to nicely ask your process to clean up after
>> itself. The only real option is to kill your process.
>
> And how often does that happen?... again generational collectors. I think
> I had a kernel kill my process once (on linux).

on mobile (esp Maemo), oom-kills happen regularly.

> At least you're not biased ("crazy schemes") ;-). I think, on Windows,
> the CLR is tight with their kernel a bit like that. Maybe crazy, but
> most likely efficient. I don't know the details, but at least CLR
> subscribes to some memory pressure events.

On Maemo the [linux] kernel publishes some memory pressure events too,
the problem is that on the n900 by the time you have pressure, you're
already swapping to death, and it's too late.