RFR(S): 8025942: Implement os::Bsd:available_memory

[Gerard Ziemski]

A single run of the code will verify that it's coded correctly. An 
assert would be needed if the conditions were likely to change during 
runtime.

If we were, for example, dynamically calculating the attribute size, 
then things might go wrong, but that's not the case here - we use a 
constant.


cheers

On 10/11/2013 10:52 AM, Daniel D. Daugherty wrote:
> Actually an 'assert' is exactly what you want in this case.
> The call shouldn't fail unless we have a coding error so in the
> event that we introduce a code error in the future, the assert
> will fire.
>
> Dan
>
>
> On 10/11/13 9:47 AM, Gerard Ziemski wrote:
>> hi Harold,
>>
>> I don't think that's necessary - the call will fail in 2 cases:
>>
>> 1. null host
>> 2. wrong attribute size
>>
>> Re 1. we use mach_host_self() to get the host - if that were to fail, 
>> we'd have failed long before here.
>> Re 2. the XNU implementation of the call checks that attribute size 
>> is set to HOST_VM_INFO64_COUNT, which is what we use
>>
>> so in my opinion no assert is needed, and adding it might in fact 
>> muddle the waters as any future reader of the code might very well 
>> think that it's likely to fail.
>>
>>
>> cheers
>>
>> On 10/11/2013 10:33 AM, harold seigel wrote:
>>> Hi Gerard,
>>>
>>> Since host_statistics64 is unlikely to fail, can you add an assert 
>>> to ensure that we catch the rare case when it fails?
>>>
>>> Thanks, Harold
>>>
>>> On 10/11/2013 11:26 AM, Gerard Ziemski wrote:
>>>> hi David,
>>>>
>>>> On 10/10/2013 11:32 PM, David Holmes wrote:
>>>>> Hi Gerard,
>>>>>
>>>>> On 11/10/2013 5:14 AM, Gerard Ziemski wrote:
>>>>>> Please review this fix that implements os::Bsd:available_memory()
>>>>>>
>>>>>> Description:
>>>>>>
>>>>>> This is a simple change - we use a similar implementation that Apple
>>>>>> itself uses (see
>>>>>> http://opensource.apple.com/source/system_cmds/system_cmds-498.2/vm_stat.tproj/vm_stat.c) 
>>>>>>
>>>>>> to implement BSD (Apple platform only) implementation for finding 
>>>>>> out
>>>>>> available (free) memory.
>>>>>
>>>>> Is host_statistics64 likely to fail? If not then I think we should 
>>>>> trap any failures, at least with an assert, so that they are 
>>>>> noticed. Otherwise this might fail all the time and we would be 
>>>>> none the wiser.
>>>>
>>>> I looked at the implementation of host_statistics64 in XNU src and 
>>>> the only reason it will fail is if we pass wrong parameters (such 
>>>> as NULL host). Otherwise, it is guaranteed to return a valid value, 
>>>> so there does not seem to be a need for an assert here.
>>>>
>>>>
>>>> cheers
>>>
>>>
>>>
>>
>
>