Chris Wells wrote:
> Chris Kantarjiev wrote:
> 
>> I'm running 4.0.25 on NetBSD 3 on an amd64. It appears that the 
>> key_buffer_size is being truncated to 4GB, even though I've set
>> it larger.
>>
>> Could someone give me a hint about where to look? I'm a little
>> suspicious of ha_resize_key_cache() which is using longs internally
>> to hold the keybuff_size, but I suspect this is not quite the
>> source of the problem (there's no logic here to truncate the
>> input value).
>>
>> Thanks,
>> chris
>>
> 
> First thought - could 4GB be the file size limit for the filesystem you 
> are using?
> 
> -Chris
> 
The index file is 17GB. File size limit isn't 4GB.

And I don't immediately see how the file size limit would affect the setting of 
an internal variable.

The assorted key_buffer_size code (what a maze of twisty little passages!) has 
been partially arranged to expect ulonglong, but there are some lower layer 
handler routines that cast to long (I'm using MyISAM). That's pretty obviously 
wrong for me - but I haven't even found the place where the input value is being 
truncated so the cast to long does something other than return 0...