On Sep 27, 2017, at 10:54 AM, Julian Kunkel <juliankunkel(a)googlemail.com> wrote:
Dear all,
I just now submitted a pull request:
https://github.com/IOR-LANL/ior/pull/9
The semantics is pretty similar, if you think about it:
when you do "-r -R" it means read data into a buffer, then read data
into another buffer and compare both buffers.
That is simply an unnecessary feature as it does not allow to detect,
e.g., data corruption except if it happens between those reads.
Now -R reads the data and compares it with the already *known* buffer,
so it will detect data corruption that occurred after writing the
data. That can even been a year after the data has been written...
Since that also removed quite some complexity that was previously
needed to exchange the buffer data, I believe that is the semantics
that we want to have.
I would agree with Julian here. Reading the data twice and comparing them
is only going to detect if the storage is horribly broken. I'd always
(incorrectly) thought that this was comparing the data to known-good values
as the write verify code is doing, so IMHO this is a definite improvement.
Cheers, Andreas
2017-09-27 18:04 GMT+02:00 John Bent <johnbent(a)gmail.com>:
>
>
>> On Sep 27, 2017, at 8:51 AM, Julian Kunkel <juliankunkel(a)googlemail.com>
wrote:
>>
>> Indeed I believe the current -R option should be replaced by comparing
>> the value with the expected (written) value.
>
> I suggest that we check with the IOR maintainers before changing the semantics of an
argument. I know them and I suspect they won't like this. Instead, perhaps just leave
-R alone and add a new flag?
>
> John
--
http://wr.informatik.uni-hamburg.de/people/julian_kunkel
Cheers, Andreas