Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 07 Nov 2012 01:38:14 +0200
From:      Alexander Motin <mav@FreeBSD.org>
To:        attilio@FreeBSD.org
Cc:        svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org, Konstantin Belousov <kib@freebsd.org>, Ben Kaduk <minimarmot@gmail.com>
Subject:   Re: svn commit: r241896 - in head: . cddl/contrib/opensolaris/lib/libzpool/common/sys share/man/man9 sys/cam/ctl sys/cddl/compat/opensolaris/kern sys/cddl/compat/opensolaris/sys sys/cddl/contrib/openso...
Message-ID:  <50999F66.1020402@FreeBSD.org>
In-Reply-To: <CAJ-FndBRFJFZ_h_fY%2B=Wux_DfMr4V0RrqyiKPh64WmiTDm3o4g@mail.gmail.com>
References:  <201210221750.q9MHot26061585@svn.freebsd.org> <CAK2BMK5c==SJ%2BySe7S70ZJyph_2X%2BdU%2B9zBftdatWqTVsH5rsA@mail.gmail.com> <CAJ-FndCTQjxbhpv-nA_oiVcHbKxwvpG_0qN9Cr4HV7_xfSQbeQ@mail.gmail.com> <CAK2BMK7srogaYt6Y9fp=HYSY64NXwBSFDHTuXiMYhbPmOD2NAg@mail.gmail.com> <CAJ-FndDF%2BM_QALAuL_z9b5X_T4=En7Ek26u0kbqMEANcWLVcLQ@mail.gmail.com> <CAK2BMK7W%2B0EqkRYyvRQoSAChdgcRJXZiSmfoE=Xj3evy4C7THQ@mail.gmail.com> <CAJ-FndBRFJFZ_h_fY%2B=Wux_DfMr4V0RrqyiKPh64WmiTDm3o4g@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On 07.11.2012 01:20, Attilio Rao wrote:
> On Tue, Nov 6, 2012 at 11:17 PM, Ben Kaduk <minimarmot@gmail.com> wrote:
>> I do not wish to belabor the point; we all have better things to do
>> with our time.  Hopefully this is my last message on the topic.
>>
>> On Tue, Nov 6, 2012 at 6:10 PM, Attilio Rao <attilio@freebsd.org> wrote:
>>
>>> The point is that KPI/KBI of -CURRENT can change as long as
>>> __FreeBSD_version is bumped (and if you really want to know my
>>> opinion, I already see this as a forceful thing because it would not
>>> be necessary in my mind, but I second the will of the majority of
>>> developers). So, if the KPI/KBI changes all the thirdy part code,
>>> ports and everything else must adapt.
>>
>> Yes, everything must adapt to changes in -current.  I am arguing that,
>> if it is easy to do so, we should make the user experience for
>> *ordinary users* running -current as nice as possible.  If we do not
>> have ordinary users running current, then our code does not get
>> real-world testing until RC builds, or even the .0 release.  I think
>> it is well-accepted that we want to have the code in -current get
>> real-world testing; making the user experience nicer helps this to
>> happen.  To me, it seems that the user experience is nicer if the KPI
>> change is delayed from the KBI change.  We have mechanisms in place
>> that can enforce __FreeBSD_Version of kernel modules must match the
>> version of the running kernel, so I do not see how this procedure
>> would lead to silent binary incompatibility.
>
> The courtesy you are mentioning here is the __FreeBSD_Version. Having
> stricter rule would just meaning doing under-performing and unclean
> job.
>
>>
>>> MPSAFE flag is not any longer supported and code needs to be ported
>>> appropriately to -CURRENT interface.
>>
>> That is the present state of affairs, yes.  I am asking only, "think
>> of the users; can we make things easier for them?".
>> Maybe not in this case, but as something to keep in mind for the future.
>
> I can understand your concern, but people using -CURRENT must be well
> aware of the fact that this is a development branch and they cannot
> expect too many safety belt mechanisms to be in place.
>
> I think that the current model (break KBI/KPI at will, give
> ports/thirdy part a way to recognize it via __FreeBSD_Version and move
> on) is optimal because it doesn't limit the developer neither leaves
> the user completely without a landmark on how to fix the problem.
>
> It is all balancing in finding compromises :)

I would like to support Ben's position. We may not expect that all the 
world is tracking our development in real time and adapt to our changes 
at the same exact moment. As I understand, before this one change MPSAFE 
flag was mandatory to have. After that one change MPSAFE is missing and 
so must to be removed. Yes, in perspective that part of code in external 
sources will be wrapped with respective ifdef's, but could we give world 
some time window to adapt, leaving constant doing nothing in tree? What 
is the price of having single define, comparing to headache even for one 
user who was brave enough to run HEAD?

-- 
Alexander Motin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?50999F66.1020402>