[Top] [All Lists]

Re: [PATCH 02/25] xfstests: remove bench infrastructure

To: xfs@xxxxxxxxxxx
Subject: Re: [PATCH 02/25] xfstests: remove bench infrastructure
From: Troy McCorkell <tdm@xxxxxxx>
Date: Tue, 26 Mar 2013 15:26:05 -0500
Cc: Dave Chinner <david@xxxxxxxxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <514C853A.20803@xxxxxxx>
References: <1363350489-22257-1-git-send-email-david@xxxxxxxxxxxxx> <1363350489-22257-3-git-send-email-david@xxxxxxxxxxxxx> <20130315193635.GA2025@xxxxxxxxxxxxxxxxxxxx> <20130319223149.GA17758@dastard> <20130321235207.GK2025@xxxxxxxxxxxxxxxxxxxx> <20130322073743.GS17758@dastard> <514C853A.20803@xxxxxxx>
User-agent: Mozilla/5.0 (X11; U; Linux i686 (x86_64); en-US; rv: Gecko/20091204 Thunderbird/3.0
On 03/22/2013 11:22 AM, Troy McCorkell wrote:
On 03/22/2013 02:37 AM, Dave Chinner wrote:
On Thu, Mar 21, 2013 at 04:52:07PM -0700, Phil White wrote:
On Wed, Mar 20, 2013 at 09:31:49AM +1100, Dave Chinner wrote:
I understand why you see value in a benchmark infrastructure, but
that's not the issue here. The issue at hand is whether we should be
trying to maintain arbitrary abstractions that are made redundant by
the cleanup patch in the hope they are useful in the future.

There are solid technical reasons for what I've done, but you
haven't provided any to advocate why we should accept your version
as a better solution. "personal interest" is a good thing to have,
but it's not a convincing technical argument....
Let me be clear:

Personal interest is why I took it upon myself to move this along,
rather than letting it moulder away even further than it has while
we wait for you to respond to our feedback.
/me does a double take

Please don't try to rewrite history.  This patchset has been held up
by SGI steadfastly refusing to remove the bench infrastructure
regardless of the technical merits of doing so, not by me failing to
respond to SGI's feedback.

It makes me wonder what your motives are.  Did you swear some sort of
vendetta against bench and remake?  Is there a blood oath between the
houses of Chinner and common?

It's unnecessary code churn.
You're calling this code churn and implying it's driven by zealotry.
I'd guess this is the first major cleanup you would have seen in the
XFS code base. We've done many and as a matter of principle they do
not leave unnecessary cruft behind.  This is the way we improve the
quality of the code base.

It's not possible to clean up code properly if you don't remove all
the problematic code and interfces when the opportunity arises.  If
it turns out we need it in future, then we pull it back out of the
revision history and use just the bits we need.  We've been through
this cycle several times before as needs have changed. e.g.  have a
look at the history of the XFS_ISIZE macro in the kernel code.

As for my work, there's an advertising slogan which I'm adapting for you
here: "One oughtn't post a whine before its time".  I have not posted
that work yet.
OSS development motto:

"Release early. Release often."

What I do know is that you're making extra work for us.
Other people do development, and they are not going to stop making
changes just because it "makes extra work for you".  I have to keep
tens of thousands of lines of patches current at the moment with all
the CRC changes, but I'm not using that as an excuse to stop other
people making changes...

So my choices are simple here:
1) I could give up -- as everyone else has -- and let this linger forever,
It won't linger forever - I'm really not that far away from sticking
a fork in xfstests...

2) You could accept that this is a change which you have no real reason to
    make, or
3) I can do what I set out to do: Get this patch series into xfstests and
    write extra code to bring my stuff in.

The timeframe in which xfstests can move forward in cases 1 or 2 are
unacceptable to me. So I'm going to do 3, review your March 15th series,
and move this forward.
OK, well lets see how that goes :)



As I understood the history, this reorganization has been talked about for a long time, and agreed upon at the Linux Filesystem meeting.

Dave did the work to reorganized the tests and posted as a RFC. A couple people sent back rough feedback like "the output from this test ends up in the main directory".

SGI wanted to make sure benchmarks are available and easy to use.
Dave said many or all of the benchmarks in xfstests are out of date, he even provided examples of better benchmarks.

We all got sucked into other projects until recently Eric pinged on this series because it
will benefit the whole Linux filesystem community.

Phil took over reporting the series, and reposted it to the list. Dave reposted the series that
he had with updates.

I thought we all reached a common agreement on the reorganization. We all want it. SGI would like to add benchmarking as a follow-up patch. Dave wanted to make sure that if done, benchmarking is done correctly. Dave's latest series is more up to date. I thought
we were going to do a quick re-review, and get it committed ASAP.

We will complete the review of Dave's second patch set and get it committed ASAP.
I expect it will be committed the beginning of next week.

Thank you,
Troy McCorkell
SGI XFS Manager

xfs mailing list

Rich is updating the patch set to TOT. He'll complete it today and do a test run.
The patch series should be committed tomorrow. (Wed March 27)


<Prev in Thread] Current Thread [Next in Thread>