Filebench [PSARC/2007/448 FastTrack timeout 08/16/2007]

[Joseph Kowalski]

eric kustarz wrote:
>
> On Aug 2, 2007, at 1:44 PM, Joseph Kowalski wrote:
>>
>> Oh yea, a real observation on this proposal:
>>
>>    Why not Volatile?  Benchmarks are often enhanced on a whim and 
>> sometime
>>    those whims can be incompatible.  What is the value (in this rare 
>> case) of more
>>    stability.
>>
>> - jek3
>>
>
> This is in regards to the Location in the "Interfaces" sections, correct?
No.  Actually for the "Invokation: Uncommitted" line seems more like it.
> 3. Interfaces
>
>         /usr/benchmarks/filebench/filebench
>         Location: Committed
>         Invocation: Uncommitted
>         Human-readable output: Not an Interface
The fact you used "Not an Interface" shows that you read the taxonomy.  :-)

>         /usr/share/man/man1/filebench.1
>         Location Uncommitted
>
> So if we classify this as "Volatile" then we can change whatever 
> /usr/benchmarks/filebench/filebench does in an incompatible manner 
> with previous versions?
>
> But if we go with "Committed", then we cannot?
>
> Yes, i've read:
> http://www.opensolaris.org/os/community/arc/policies/interface-taxonomy/
> but i'd like clarification :)
>
> eric
I can't get too wrapped up between Committed/Uncommitted for 
"Location".  Uncommitted makes it a little easier to remove it 
completely.  We would probably be smoking something if were to just move 
its location (which is why we gotta get it right).

What I'm not seeing why we should do anything to make this harder to 
just suck in the latest version.  Who cares if a benchmark is 
incompatable with its previous version?  Anybody who understands 
benchmarks even a little bit, they have to always cite the benchmark 
version.

I know this kinda goes against our DNA, but I think being newer is 
better than stable for benchmarks.

We usually don't cite a stability level for a man page.  Its kinda 
recursive anyway, after all, the man page is the thing which tells us 
what the stabilities are.  :-)

So, my *suggestion* is:

3. Interfaces

        /usr/benchmarks/filebench/filebench
        Location: Committed                                  (but 
Uncommitted also seems OK.)
        Invocation: Volatile
        Human-readable output: Not an Interface   (but Volatile can also 
be used)

Others may feel differently.

- jek3