Re: libcroco and xz

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Fri, Jun 15, 2012 at 07:02:14PM +0000, David Holland wrote:
> On Fri, Jun 15, 2012 at 02:56:20PM +0200, Joerg Sonnenberger wrote:
>  > > >>I see...
>  > > >>
>  > > >>using *.xz archive, so marked as build depend on archive/xz,
>  > > >>but check-shlib claims using shlibs from build depends.
>  > > >>It's false alarm.
>  > > >
>  > > >I can reproduce it now. It is *not* a false alarm. libxml2 should be
>  > > >pulling in a full xz dependency, since libxml2.so depends on it now, but
>  > > >it only gets counted as build dependency.
>  > > 
>  > > No, not only directly build dependency, but also indirectly full
>  > > dependency.
>  > > In other words, directly build dependency on `xzcat' command from
>  > > archive/xz,
>  > > and indirect full dependency on `liblzma' from archive/xz.
>  > 
>  > Right, there is no such thing as an indirect full dependency...
> 
> Sure. Call it whatever you want. But libcroco does not directly depend
> on xz or liblzma. Grepping the entire libcroco distribution for 'lzma'
> matches only some makefile goop for creating .tar.lzma files.
> 
> We should not be adding a bogus xz runtime dependency to libcroco (and
> everything else that uses libxml2) and the checks done by pkgsrc
> should not demand one or encourage people to add one incorrectly.

The problem I have with the commit is that it also hides the reverse
case. A dependency is declared as build time only, used by a dependency
as full dependency, but explicitly linked against. I'm more inclined to
care about that case than the reverse.

Joerg