Joerg Sonnenberger <joerg@britannica.bec.de> writes:

> On Fri, Sep 21, 2007 at 12:50:35PM -0400, Greg Troxel wrote:
>> Joerg Sonnenberger <joerg@britannica.bec.de> writes:
>> 
>> > On Fri, Sep 21, 2007 at 10:43:32AM -0400, Greg Troxel wrote:
>> >> So, I'd like us to design (quickly) a database (file format) to store
>> >> the information usable for pkg_rolling-replace, and maybe automatically
>> >> usable by make replace. I'll assume that something like (1) is done.
>> >
>> > Why can't you just use pkgsrc/doc/CHANGES-* for that?
>> 
>> We would need to change the format; it doesn't have enough information,
>> because it doesn't encode the PGKNAME of the old PKG_PATH.  And it
>> doesn't record anything when PKGNAME changes and PKG_PATH doesn't.
>> 
>> But, there are only 128 rename entries, so we could add a new type for
>> name-without-path, and change Renamed to take a pkgname as well.
>
> I'm not so much opposed to changing the file format, that is reasonable
> easy to do and converting the existing entries is not that hard either.
> My point is that I would prefer it to be the canonical source for this
> kind of information and strongly prefer it over adding some magic IDs
> :-)

So:

        Renamed devel/GConf2 to devel/GConf [wiz 2007-09-21]

would become

        Renamed GConf2 devel/GConf2 to GConf devel/GConf [wiz 2007-09-21]

with all fields being mandatory in all Renamed entries, and we'd fix up
the past.

Is this ok with everyone?