Re: CVS commit: pkgsrc/net/xymon

[David Holland <dholland-pkgchanges%netbsd.org@localhost>]

On Fri, May 11, 2012 at 01:07:37PM +0200, S.P.Zeidler wrote:
 > >  > > The problem really does appear to be that it's installing files in an
 > >  > > OWN_DIRS dir, though.
 > >  > 
 > >  > By no means. That's what *_DIRS exist for.
 > > 
 > > It was complaining that a file is both in the PLIST and in
 > > CHECK_FILES_SKIP. OWN_DIRS puts things in CHECK_FILES_SKIP. It's not
 > > *that* broad a jump to deduce that these shouldn't be combined.
 > 
 > Ah, I see. Yes, that is a very nice red herring. 

Well, red is the best colour for herring :-)

 > >  > > (It usually doesn't in destdir mode.)
 > >  > 
 > >  > I'm not sure how you get to that conclusion.
 > > 
 > > As far as I know, in destdir mode, you need to create ${PREFIX}/bin
 > > explicitly, usually with INSTALLATION_DIRS.
 > 
 > Not only in destdir mode IME, e.g. $PREFIX/libexec may not yet exist for
 > an early package. MAKE_DIRS also creates the directory in the right phase.
 > The effective difference for a directory under PREFIX seems to be that
 > it gets put into CHECK_FILES_SKIP if listed in MAKE_DIRS but not if
 > in INSTALLATION_DIRS or REQD_DIRS.

Right.

 > If you have dir a in INSTALLATION_DIRS and subdir a/b in OWN_DIRS (because
 > it is expected to contain files that should not be deleted when
 > deinstalling the package for updates) deinstalling the package will
 > nuke a and thus also a/b. That's unfortunate.

That seems badly broken. What do Python and such do with this? They
have lots of dirs used by subpackages that don't get nuked like this.

Then again I suppose because there *are* packages involved the dir
refcounts for those will be >1 when python is uninstalled.

Also, does this happen if the package doesn't have one of its own
files in directory b?

-- 
David A. Holland
dholland%netbsd.org@localhost