On Thu, Dec 08, 2005 at 11:16:52PM -0500, Rob Kinyon wrote:
: On 12/8/05, Larry Wall <[EMAIL PROTECTED]> wrote:
: > [snip] Certainly, as you speculate, if different authors want
: > to share an API, they can give it an "API" author that knows how to
: > delegate to one of the authors.
:
: Would you mind elaborating on this some more?
I don't really have much elaboration to give. We have policy modules
that can invoke other modules on behalf of the user, and you can view
an API module as a kind of policy module.
We also already know that we need to figure out how to pick between
Funny-1.2-STAN
Funny-1.2-OLLIE
using something like a policy. Certainly one way to handle that would
be to allow whomever is making the decision to create a
Funny-1.2-API
that picks between ABBOT and COSTELLO, as a kind of editorial decision.
There are two elaborations on this theme. One is that
use Funny-1.2
could be considered a request to search for
use Funny-1.2-API
or more likely
use Funny-(1.2.0...)-API
or even
use Funny-(1.2.0...)-(/^.* _API$/)
The second is that "API" if API here is representing some kind of
editorial selection, there are potentially multiple editorial entities,
and they would presumably have different names.
use Funny-(1.2.0...)-USER_API
use Funny-(1.2.0...)-SITE_API
use Funny-(1.2.0...)-CPAN_API
use Funny-(1.2.0...)-RH_API
use Funny-(1.2.0...)-MS_API
use Funny-(1.2.0...)-G_API
use Funny-(1.2.0...)-K_API
use Funny-(1.2.0...)-MOD_API
use Funny-(1.2.0...)-ECMA_API
and then
use Funny-(1.2.0...)-API
might actually be end up looking for
use Funny-(1.2.0...)-(/^ $THEME _API $/)
and then we get to fight over who sets $THEME.
On the other hand, it can reasonably be argued that a "true" API should
be completely agnostic as to its implementation. On the gripping hand,
with roles we are specifically giving our interfaces the right to
specify default implementation, and this is construed as a feature.
So maybe it's okay for an API to specify its default. In fact, arguably
it should be *required* to specify its default so there's clarity about
who has to add more authorial information and who doesn't.
But this is all probably something the CPAN folks would do a better
job designing, not me. I just wanted to point out the underlying
principle that version numbers are like array indices and authors are
like hash keys, and that sometimes you want to have ordered versioning,
and sometimes you don't. But then we have to give users a way to sort
the hash keys.
Hmm, I guess I had more to say about that than I thought I did.
Larry
