[OpenAFS-devel] Making supergroups default to on

[Stephan Wiesand]

On Dec 30, 2012, at 07:41 , Dave B. wrote:

> As its a change to a default, I would object to the change. Especially =
w the potential code issues and potential inability to go back to =
supergroups off. Changing the default is not a bug fix or security fix.
>=20
> Such a change should either be reflected in a major version number =
change such as 1.8  or have a second set of packages, say 1.6.2-sg.

I agree. Reducing the number of optional features is good, but such a =
change shouldn't happen in the stable release series - unless there are =
no doubts that it will just work for everyone, and it doesn't introduce =
any compatibility issues. Turning on supergroups by default is very =
unlikely to happen on 1.6.x from my point of view.

- Stephan

>=20
> IMHO.
>=20
> Sent from my ASUS Pad
>=20
> Jason Edgecombe <jason@rampaginggeek.com> wrote:
>=20
>> On 12/28/2012 03:33 PM, Simon Wilkinson wrote:
>>> On 28 Dec 2012, at 19:20, Andrew Deason wrote:
>>>> I'm not sure if that's really an absolute requirement, though. The =
'code
>>>> quality issues' that I remember were just the opinions of a few =
people
>>>> that would make them uneasy about turning it on. I don't recall =
anyone
>>>> trying very hard to get the default changed, so maybe that's all it
>>>> takes.
>>> There were/are a number of different issues in the supergroups code. =
There are aliasing issues throughout, some of which could be easily =
fixed by inserting memcpy's rather than just using casts/assignments, =
but there are others that require far more in depth analysis. This =
second set of aliasing problems are mainly confined to the caching code, =
rather than the supergroups code itself. It would certainly be possible =
to either disable caching, or rework that code, whilst having the ubik =
database format remain unchanged.
>>>=20
>>> The other major issue is that there is bit twiddling in the caching =
code which makes assumptions about endianness, and about word size. I =
think we have now caught all of these problems - but certainly with the =
1.4 series you couldn't safely run supergroups on a 64 bit machine.
>>>=20
>> That's slightly unsettling. We're running supergroups-enabled cell=20
>> servers on 64bit RHEL5, albeit with YFS modifications.
>>=20
>> My main desire is to reduce the number of conditional features to =
reduce=20
>> code rot and extra ongoing development/testing efforts.
>>=20
>> Jason

--=20
Stephan Wiesand
DESY -DV-
Platanenenallee 6
15738 Zeuthen, Germany