Re: [gentoo-alt] unemerging causes a bad symlink removal

[Alan Hourihane]

On Fri, 2009-08-28 at 12:56 +0100, Alan Hourihane wrote:
> On Fri, 2009-08-28 at 11:18 +0200, Fabian Groffen wrote:
> > On 28-08-2009 10:08:11 +0100, Alan Hourihane wrote:
> > > > I guess the problem is more why it wants to remove that /usr.  Maybe
> > > > because of the same problem we once had during install, where paths
> > > > "below" the offset prefix shouldn't be taken into account at all,
> > > > because you should not do anything there, and may not even be able to do
> > > > so.
> > >
> > > All you need to do is look at the CONTENTS file which as something like
> > > this...
> > >
> > > dir /usr
> > > dir /usr/bin
> > > obj /usr/bin/comm 4e28628520359733270907c62c738b3f 1250795126
> > > obj /usr/bin/arch 59d8b1d83cfbc020751c8febc4706901 1250795126
> > > obj /usr/bin/factor a68ac36164e0ac5c1245e358e688942e 1250795126
> > >
> > > Look at the first line.... Enough said...
> >
> > right, but your prefix is '/', so that makes sense.
>
> Right, normally the "rmdir" would fail because there are files still in
> existance in those directories. But with a symlink that doesn't hold
> true.
>
> > the "fix" is most probably that portage shouldn't remove a symlink if it
> > expects a dir.  Put differently, portage should ensure that it only
> > unmerges what it expects.  Would that theory work?
>
> I'm worried about something like this....
>
> dir /usr/share/mypackage
> dir /usr/share/mypackage/test
> dir /usr/share/mypackage/symlink-to-test
>
> Where "symlink-to-test" points to "test" and should be removed. Then
> removal of "mypackage" might fail because the symlink would still exist.

Ah, I see the contents file actually says...

sym /usr/share/mypackage/symlink-to-test

in this case.

So yes, We shouldn't be removing symlinks when it's "dir".

Alan.