Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 08 Dec 2000 17:57:42 +0900
From:      "Daniel C. Sobral" <dcs@newsguy.com>
To:        Christopher Masto <chris@netmonger.net>, cvs-all@freebsd.org, cvs-committers@freebsd.org
Subject:   Re: cvs commit: src/sys/vm phys_pager.c
Message-ID:  <3A30A286.30E102EC@newsguy.com>
References:  <20001205145908.K8051@fw.wintelcom.net> <Pine.SUN.3.91.1001205180108.24320A-100000@pcnet1.pcnet.com> <20001205152054.M8051@fw.wintelcom.net> <3A2EFBC4.EE8D90B1@newsguy.com> <20001207173453.A18103@netmonger.net>

next in thread | previous in thread | raw e-mail | index | archive | help
Christopher Masto wrote:
> 
>                 object = vm_object_allocate(OBJT_PHYS,
> -                       OFF_TO_IDX(foff + size));
> +                       OFF_TO_IDX(foff + PAGE_MASK + size));
> 
> Honestly, this thread is ridiculous.  If the code didn't work in the
> first place, AND nobody noticed because it's so rarely used, AND the
> change is _obviously_ trivial, then what's the problem here?  "stable"
> is not "immobile".

Typos happen. Go look the cvs repo or cvs-all archives, and you'll see
plenty changes very much like the one above where the committer just
happened to make a typo and not notice.

Also, using cvs also sometimes can result in mistakes, by confusing
versions, mixing local repo with remote repo, etc.

Further, any code with macros and/or #ifs may not compile cleanly under
different options than what the author tested them with.

This is not hypothetical. It *happens*. It has happened in the past, and
will happen again in the future. And risking breaking -stable's world is
just not worth waiting a few hours to see if no one had a problem.

> Even if this fix was wrong (and apparently it was), it doesn't change
> the fact that it was:
> 
>   1. A fix for code that was known to be unusably broken

But, nevertheless, didn't break world.

>   2. Extremely unlikely to hurt anything (esp. given #1)

As far as *functional* changes go, this is right. But the problem is
elsewhere. See above.

>   3. Extremely unlikely to break the world (obvious by inspection)

Experience tells another story. In particular, typos have an annoying
tendency of escaping visual inspection because the human mind tends to
adjust what it is seeing in terms of what it expects to see.

>   4. Trivially reversible

And trivially simple to wait 24 hours.

> You've got the right argument, but you've picked the wrong example
> to apply it to.  There has to be _some_ flexibility in the rules,
> and it would be hard to find a better example of where "shakeout
> in -current" has no value.

It is very, very bad when -stable world breaks. It tarnishes our
reputation, increases support workload, and make people less confident
in the process.

Perhaps I see the rules in a different light than many others. To me,
"common sense" there refers to being able to discern when a commit
_must_ be merged as soon as possible. I have no problems with Alfred
making such a call. But the call he made, and many others seem to
support, is one of "may" be merged instantly. I don't think anything may
be merged instantly unless there are reasons that _requires_ it to be
merged instantly. This, at least, is my view of the rules in the light
of the problems that have happened in the past.

Alfred, I'm not disputing your ability to make judgment calls. I just
think that the risk of silly mistakes (mentioned above), that have
nothing to do with what your _intentions_ were when you wrote the fix,
outweights the benefits to be gained by an instant MFC (as a matter of
fact, I don't see any benefits except your own convenience).

-- 
Daniel C. Sobral			(8-DCS)
dcs@newsguy.com
dcs@freebsd.org
capo@the.great.underground.bsdconpiracy.org

		"The bronze landed last, which canceled that method of impartial
choice."


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3A30A286.30E102EC>