[PATCH blktests v2 03/11] common/xfs: Create common helper to verify block device with xfs

Logan Gunthorpe logang at deltatee.com
Wed Oct 7 11:53:44 EDT 2020



On 2020-10-06 6:20 p.m., Chaitanya Kulkarni wrote:
> On 10/6/20 16:59, Logan Gunthorpe wrote:
>>> The mount dir should be a parameter and not the hardcode value
>>> to make it reusable.
>> I also disagree here. It is already reusable and is used in a number of
>> places; none of those places require changing the mount directory. If
>> and when a use comes up that requires a different directory (not sure
>> what that would be), a parameter can be added. It is typically standard
>> practice in the Linux community to not add features that have no users
>> as it's confusing to people reading the code.
>>
>> Logan
>>
> Well if you are making a helper it should be generic if you have a usecase,

"Generic" isn't a binary yes/no quality. Why add the mount option (that nobody is using) 
and not a size option as well that nobody uses? For that matter, fio has a ton of options
we could expose. (think io-method, read/write pattern, etc, etc). The criteria we
decide upon which options get exposed as arguments is what the code that's actually
using it needs -- not what's available or what you imagine future use cases might be.
If there are no users in the code it should not be exposed. If a use case comes along,
an argument can easily be added when the new test is added to the code base.

> mounted on different mount points not just one which is important testcase,
> 
> that will require a prep patch.

So? That's normal.
 
> Why can't we do that right now when we have a clear usecase ?

We don't have a clear use case that's being added to the code though... We 
have an imagined use case that hasn't been written. Add the feature when you
add this use case.

Logan



More information about the Linux-nvme mailing list