> And if the application specifies both "X" and "not X"? 

That would be an error.

> > An alternate API would be to have one flag per property and always require
> a > get before a set.
> 
> I think I would prefer this approach. This is clean, no special cases,
> and allows simple checking of system defaults. The application can
> always store the default bit vector if it creates multiple sockets.

A psychological aspect of such an API is that we need to nail down
what the defaults are for all implementations, since application
writers might skip invoking the API unless they are going to set
some flag. But there is indication that some implementations
want to prefer temporary addresses over public addresses, whereas
others will follow the default address selection document and do the
reverse.
So do we make the one bit PREFER_SRC_TMP or PREFER_SRC_PUBLIC?
Will applications that have a preference for either always explicitly state it
by invoking the API?

The psychological benefit of the two flags is that we don't have to choose
the default; we can say that any application which has an explicit preference
should always invoke the API to state that preference.

  Erik

--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to