Date: Thu, 8 Jan 98 11:00:51 EST
From: bhyde@gensym.com (Ben Hyde)
Message-Id: <9801081600.AA23844@gensym1.gensym.com>
To: new-httpd@apache.org
Subject: Re: voting
In-Reply-To: <19980108005704.26364@texas.net>
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

Observations (old man blathering):

In my shop we use a "voting" procedure near releases.
We call it witnessing, not voting.  I've used this
device since the late 1970s so I've got some experience
under my belt (or is that just beer?).

As the release approachs the quality must get very
tight and the problems tend to be the most subtle and
architectual.  Sleep depravation causes team IQ to
plumet.  Customers have no sympathy.

Things are different in my world, somebody is usually
in charge.  I've usually had the burden of being the
that person, or his lawyer, so I could decide when to
move in and out of the witness mode.

Witnessing is the consequence of a "successful" code
freeze.  It's something project managers want, badly.
That's a slippery beast and I've had to back up when I
mandated it too soon.

There are always named parts of the project that need
to have witnessing postponed, they just aren't ready
yet.  I like to say: "you nail the lid on the coffin on
nail at a time."

New complex ports are almost always late to the party.

Performance tuning is usually done late in projects.
Managing it's risk against the stablization is always
difficult.  Particularly difficult if the guy doing it
is too clever and hence bits off big archetectual
changes.  It's great if you can get one really
competent person to focus on it and then give him a
single witness to guard against sleep and enthusiasm
problems.

I find that there are always individuals who's stature
fogs the judgement of the witness, it's occationally a
serious problem.

Suggestions:

I think the group could adopt some refinements to 
the voting rules.
  1. Named projects can be voted to have temporary 
     lower voting requirements.
  2. Commits that trigger regressions maybe fixed
     with NO voting - the coder should be prepared
     to ask forgiveness when he makes things worse.
  3. The release manager - in exchange for doing
     the damn job - is allowed to grant temporary
     wavers.

I'd propose that both the performance work and the
window's porting work be reduced to requiring only
a single witness for the next few weeks.

 - ben hyde