Hey,

You were right and wrong. Using min() wouldn't help since we pass the
overflowed result of the addition, which could be say 1, even when there is
_much_ more memory available on the computer. It could happen on 64-bit systems
too when the user has a lot of memory (say 64 GiB, we don't see these today a
lot, if at all).

I've prepared a new version of the patch, hopefully it is good enough.
It seems that on Mac and Unix systems, the function mch_avail_mem() isn't used
or even defined at all, so this patch is only needed for Windows at the time.

--
Jelle Geerts

On Sun, Aug 24, 2008 at 5:00 AM, George V. Reilly <[EMAIL PROTECTED]> wrote:
>
> 2008/8/23 Jelle Geerts <[EMAIL PROTECTED]>:
>> Hello,
>>
>> When allocating memory, lalloc() is called, which uses mch_avail_mem(). But
>> mch_avail_mem() sometimes fails. It returns the available physical memory +
>> available page file memory, which can sometimes wrap around 32-bits.
>>
>> Also, mch_avail_mem() uses GlobalMemoryStatus() which isn't working properly 
>> on
>> computers with more than 4 GiB of memory installed. There is an extension
>> available on NT platforms 0x0500, called GlobalMemoryStatusEx(). As opposed 
>> to
>> GlobalMemoryStatus(), the extension works reliably with systems that have 
>> more
>> than 4 GiB of memory installed.
>>
>> Please see the attachment for the diff file for my patch. If the comments I
>> have added are too wordy or someone does not like them, just remove them ;)
>>
>> Thanks,
>> Jelle Geerts
>
> It looks to me like your patch will fail if ms.dwAvailPhys is more
> than 4GiB, due to underflow of (0xffffff00 - ms.dwAvailPhys). This is
> a DWORDLONG, which is always an unsigned 64-bit int.
>
> Also, shouldn't that be msex.ullAvailPhys, not ms.dwAvailPhys in the
> (_WIN32_WINNT >= 0x0500) && !defined(_WIN64) case?
>
> Better, I think, to return (long_u) min(0xFFFFFF00, msex.ullAvailPhys
> + msex.ullAvailPageFile). Or for Bram to widen this signature to a
> 64-bit int for all platforms. Linux and Mac boxes with >4GiB are going
> to be common soon.
> --
> /George V. Reilly [EMAIL PROTECTED]
> http://www.georgevreilly.com/blog http://blogs.cozi.com/tech
>
> >
>

--~--~---------~--~----~------------~-------~--~----~
You received this message from the "vim_dev" maillist.
For more information, visit http://www.vim.org/maillist.php
-~----------~----~----~----~------~----~------~--~---

Raspunde prin e-mail lui