cvs commit: ports UPDATING ports/databases/postgresql-devel Makefile distinfo pkg-plist ports/databases/postgresql80-server Makefile

[Jacques A. Vidrine]

On Mon, Jan 24, 2005 at 03:43:40PM +0100, Palle Girgensohn wrote:
> 
> 
> --On söndag, januari 23, 2005 11.22.50 +0100 Mathieu Arnold 
> <mat at FreeBSD.org> wrote:
> 
> >+-le 23/01/2005 10:06 +0000, Sean Chittenden écrivait :
> >| seanc       2005-01-23 10:06:21 UTC
> >| Port epoch bumped   because 8.0.rc* is greater than 8.0.0.
> >
> >Note quite sure about that :
> >$ pkg_version -t 8.0.rc5.2005.01.16 8.0.0
> ><
> 
> Uh, 4.10 differs from 5.3 here:
> 
> FreeBSD 4.10$ pkg_version -t 8.0.0rc5 8.0.0
> >
> FreeBSD 4.10$ pkg_version -t 8.0rc5.2005.01.16 8.0.0
> >
> 
> FreeBSD 5.3$ pkg_version -t 8.0.0rc5 8.0.0
> <
> FreeBSD 5.3$ pkg_version -t 8.0rc5.2005.01.16 8.0.0
> <

This is partially pilot error.  Note that "8.0.rc5" and "8.0rc5" are
really completely different versions.

Unfortunately, this was changed by revision 1.5 of
src/usr.sbin/pkg_install/lib/version.c: several strings including "rc"
are now handled specially in direct contradiction of our documented
Package Naming Conventions.  I think the change was wrong and
insufficiently reviewed, but it is too late now.

Of course, ports that actually correctly follow the Porter's Handbook
never be named such that this makes a difference.  In other words, the
Porter's Handbook recommends against names such as "8.0rc" or
"8.0beta" in favor of "8.0.r" and "8.0.b".

> portupgrade has it own algorithm. In conflict with pkg_version-5.3,
> it considers 8.0.0rc or even 8.0.0beta to be greater than 8.0.0.

That's because portupgrade does the right thing and follows the
Porter's Handbook when it comes to version numbers.

> pkg_version-5.3 does the right thing, and to me it looks like the
> portupgrade's algorithm for versioning is overly simple (I've never
> used ruby though, perhaps it has some fancy operator overload
> stuff?):

No, see above.

[...]
> I haven't checked if this is fixed in 4.11, but since it is not even out,
> bumping port epoch is unfortunately necessary. :(

It was not necessary--- you just made typos in your tests ("8.0.rc5",
not "8.0rc5").  However, it is now necessary that it stays.

Cheers,
-- 
Jacques A Vidrine / NTT/Verio
nectar at celabo.org / jvidrine at verio.net / nectar at FreeBSD.org