statvfs() weirdness on alpha (RH-7.2 and CentOS-4.2)

[Sergey Tikhonov]

Oliver Falk wrote:

> Sergey Tikhonov wrote:
>
>> Michal Jaegermann wrote:
>>
>>> On Mon, Nov 14, 2005 at 10:37:46AM +0100, Oliver Falk wrote:
>>>  
>>>
>>>> On 11/14/2005 07:48 AM, Sergey Tikhonov wrote:
>>>> [ ... ]
>>>>  
>>>>
>>>>>> For now I changed 64-bits fields in /usr/inclide/bits/typesizes.h 
>>>>>>       
>>>>>
>>>> OK, this sounds like a fix, that should hit mainstream as well
>>>>   
>>>
>>>
>>> Hm, this sounds more like a workaround for a bug in glibc applied in
>>> order to avoid recompilation of everything which depends on glibc.  It
>>> appears that you will get screwed for a change if you will need from
>>> stat values which do not fit into 32 bits.  Not that this worked too
>>> well so far.  That probably violates some definitions/expectations 
>>> too.  No idea what long range consequences are one way or another.
>>>  
>>>
>> Yes, you are correct - it is work around. :) It is too late to 
>> rebuild everything.
>> The correct fix will be in FC5 (devel).
>
>
> Oh yes, my mistake. I wanted Sergey to post the _real fix_ to 
> glibc-people; Of course not the workaround... :-)

I submitted a bug report to glibc bugzilla. The problem is (with recent 
findings from Bert de Bruijn) I am not sure what correct
fix should be. Back in 1999 the size of fields were changed from 64 bits 
to 32.

Regards,

-- 
Sergey Tikhonov
Solvo Ltd.
tsv at solvo.ru