Re: Proposed fix for long-standing +CONTENTS bug

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2020-04-22 at 13:34 BST, Greg Troxel wrote:

> Jonathan Perkin <jperkin%joyent.com@localhost> writes:
> 
> >  * Adding a new pkg_add flag "-i" to ignore dependency errors, the
> >    "fulfilled" warning message will still be printed, but the exit
> >    status will be 0 and no "package addition failed" message will be
> >    printed, so pkgin will consider the command a success even if the
> >    package was not upgraded.
> >
> >  * Switch pkgin to use "pkg_add -iU" instead of "pkg_add -DU"
> 
> I don't follow the need for pkgin to do something different than make
> replace.  I think it's ok to suppress failing exit status on -D and
> thought there was no issue.
> 
> Is there a problem other then leaving stale +REQUIRED_BY entries?
> 
> As a thought experiment, rather than a proposal, would issuing
> "pkg_admin rebuild-tree" after every "pkg_add -DU" avoid problems?

Yeh, perhaps the brute force method might just be the simplest and
least likely to cause problems.

> > I don't like the thought of munging the pkgdb, so this way is cleaner
> > for me, but I realise it's somewhat specific to the pkgin use-case.
> 
> I don't follow "munging".  This seems to be all about ensuring that
> lines in +REQUIRED_BY are installed packges.   Or is this sort of in
> that file, and sort of in a db, and one is ok and one isnt'?

I mean where replace.mk goes and changes the +CONTENTS files and
alters @pkgdep lines to now match installed packages instead of what
the package was actually built with.  I get why it does that, but I
really don't like the thought of an installed package being modified
to be different from its binary package contents just to appease an
inconsistent pkgdb.

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com