Re: [PATCH RFC] fstests: allow running custom hooks

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2021/7/20 下午11:38, Theodore Y. Ts'o wrote:
> On Tue, Jul 20, 2021 at 04:44:29PM +0800, Qu Wenruo wrote:
>> Anyway, if building a stable and complex API just for hooks, then it's
>> completely against my initial purpose.
>
> So you said you are only using this for debugging purposes; at least
> in my workflow, when I'm trying to debug a particular test case, I'm
> *only* run a single test.

Under most case, yes, and that's also what I usually do.

But I have seen some strange case which doesn't trigger error by itself,
but needs some other tests to be run before the triggering one.

>  So today, the way I handle this is to run
> hook scripts in the gce-xfstests framework before and after running
> the check script.  It does mean that there is excess work which gets
> traced from the check script setting up and cleaning up, but it's
> mainly been good enough for me.
>
> It would be slightly nicer if there was a way to start and stop
> various debugging hooks (e.g., starting and stopping traces, clearing
> and grabing the lock_Stat info), etc.,) jut before and starting the
> test.  But it was never important to me to propose changes to the
> upstream xfstests.

 From the feedback, I guess that's the case.
I would no longer consider to upstream any simple debug purposed code.

Thanks,
Qu
>
> In answer to the question which Darrick and Dave asked, yes, I could
> do this via the exclude list.  It just seemed that if we're going to
> add a programmtic hook, maybe it would make sense to do it via the
> hook script.  It is true that there are ways to do this, though,
> although there is a difference to a static excldue list that has to
> get created for all tests that might be run, versus hook script that
> is run for each test.  <Shrug>  There are other ways of doing things;
> the question is whether the hook script approach is sufficiently
> better that it's worth getting something like this upstream.  It may
> be that keeping things like this in each developer's own private test
> runner is the right thing.   I have "gce-xfstests --hooks <hookdir>" for
> my own use, and have had it for years.
>
> Cheers,
>
> 					- Ted
>