Re: mit-krb5/builtin.mk problems on solaris

["OBATA Akio" <obache%netbsd.org@localhost>]

I think that bump of BUILDLINK_API_DEPENDS.mit-krb5 should be reset,
it is the right solution.

Bump of BUILDLINK_ABI_DEPENDS.mit-krb5 is correct, because ABI of
libraries had been changed since 1.8.x.

In pkgsrc, only security/mit-krb5-appl depend on security/mit-krb5 directly.
others are using Kerberos through mk/krb5.buildlink3.mk.
For example, heimdal will be used instead mit-rkb5 for NetBSD by default,
and BUILDLINK_API_DEPENDS.heimda is 0.4e, i.e. too lower.
I don't know about mit-krb5 API changes from 1.4.x to 1.8.x, but I never seen
krb5 API issue on any of packages with builtin heimdal on NetBSD, and
it represent that any of packages require mit-krb6 1.8 API.

If buitin.mk does not exist for a package, it is not so terrible to
maitain BUILDLINK_API_DEPENDS, because multiple version packages will be never 
mixed,
and BUILDLINK_ABI_DEPENDS will support it.
On the other hand, if builtin.mk exists, multiple versions may be mixed 
(builtin and pkgsrc),
and BUILDLINK_ABI_DEPENDS will not help it, because it will be ignored for 
builtin one.
To avoid such situation, BUILDLINK_API_DEPENDS must be managed carefully, for 
example, minimum required
version in whole pkgsrc, or majority.

Bump of BUILDLINK_API_DEPENDS is not wrong, but KRB5_DEFAULT=mit-krb5 is just 
on HP-UX, Darwin and SunOS
by default in pkgsrc, so I feel it is better to allow builtin mit-krb5 on SunOS.
I heard that bulk build of SmartOS is set as PREFER_PKGSRC=yes, so such 
BUILDLINK_API_DEPENDS changes
will not be affected (pkgsrc's one will be always used).

On Mon, 23 Sep 2013 21:35:47 +0900, Tim Zingelman <tez%netbsd.org@localhost> 
wrote:

> I have never understood why in pkgsrc a package defines its own
> minimum version.  Each other package which requires mit-krb5 should be
> expressing the minimum version it finds acceptable.  This is why when
> I updated mit-krb5 to 1.8.x and bumped BUILDLINK_ABI_DEPENDS and then
> someone saw this exact issue and asked that I change it back, I did
> so.  In retrospect I should have added a comment to the buildlink3.mk
> file explaining the situation.  Sorry about that.  I would suggest
> reverting it to mit-krb5>=1.4nb1 as it was, and if any dependent
> package breaks, fix it in there.
>
>  - Tim
>
> On Mon, Sep 23, 2013 at 6:52 AM, Jonathan Perkin <jperkin%joyent.com@localhost> 
> wrote:
> > * On 2013-09-23 at 12:33 BST, OBATA Akio wrote:
> >
> > > See also PR pkg/44597...
> >
> > Thanks.  What I will probably do is lower the requirement in our
> > 2013Q3 release branch and see what happens, then we can decide whether
> > to do the same in pkgsrc later (or perhaps make it OPSYS-specific).
> >
> > Input from Kerberos hackers would be appreciated.
> >
> > --
> > Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com


--
OBATA Akio / obache%NetBSD.org@localhost