On Tue, May 12, 2009 at 2:42 PM, M W487 <[email protected]> wrote:
> My first and most important thought is: to please provide a tool that
> will take all my related, scattered, config-like files (that you plan
> to replace), and one by one, parse them and output them into your new
> format on sysout, and comment the new format output- with where the
> file should live in the default organization.  Then I will look over
> all these outputs, and knit together "one ring to rule them all".

While I understand the anticipated frustration, that's probably not
going to happen since I'm focusing on designing a new configuration
system for CPAN Testers 2.0, which itself is a re-write.  So in this
instance, I'm asking about how to learn from the mistakes of the past,
not literally about what should be migrated from 1.0 to 2.0.

To the extent that the smoking tools migrate, it will be up to them to
port their old configuration to the new system if they so desire.

I get the point of the teaching story, but this is a case where I'm
following the principle of "planning to throw one away" -- where what
I'm throwing away is many of the crufty design choices that plague CT
1.0.

What CT 2.0 will be doing, however, is trying to require as little
configuration as possible and have sane defaults.  Adam Kennedy wants
to release Strawberry perl with a "zero-config" setup pre-installed
for CPAN Testers.

For example, at the Birmingham hackathon, a big constraint to how
Ricardo and I designed user identification and authentication for the
Metabase back-end was the need to allow users to join CPAN Testers
without having to register on a site, get an API key and password and
so on.

So the way it will work is that the CT 2.0 will generate and store a
"profile" (with name, email, API key and password) and submits it to a
Metabase.  If the profile doesn't exist, it is added to the Metabase
and used to associate with submitted reports.  (There may be
subsequent confirmation that the email address is valid, etc.)

Profile generation should be the only thing necessary to configure CT
2.0 and most of it is automated.  Testers with multiple machines just
need to copy the profile file to each machine to ensure that all tests
are recorded under the same profile.

-- David

Reply via email to