Re: Building against testing is good (was: Propossed Project: Odyssey)
On Wed, Oct 24, 2001 at 10:43:03PM +0200, Bernd Eckenfels wrote:
> On Wed, Oct 24, 2001 at 06:04:47PM +0100, Jules Bean wrote:
> > If you don't build against the lib versions in unstable, then the libs
> > will never move to testing. They won't move to testing, because doing
> > so would break packages (the packages which depend on the old
> > versions).
>
> Notmally the automatically generated dependecies are a "equal or newer"
> relationship. This means newer libs from unstable can move to testing
> anytime.
>
> This is only a problem with dev packages or if the major number of the lib
> ix chnaged. But in that case the new lib should not conflict with the old,
> anyway.
I'm talking about 'major version' changes, yes. To be more accurate,
I'm talking about soname changes (when we, in debian, change the
packagename).
When the soname changes, the new lib generally does not conflict with
the old. However, they cannot both be in testing, because the older
one will no longer be buildable from source; testing doesn't permit
there to be more than one different version of the source package.
So, the testing scripts when they install the new version of the
libgal (source) package, and along with it all the new binaries: for
the sake of the argument, including libgal11. The old version is
removed, along with all its binaries: including libgal10.
This would make any package which Depends: libgal10 uninstallable, so
testing won't do it. Or it won't unless it makes things better 'on
average', I'm not sure exactly. Certainly it won't do it if nobody
compiles against libgal11 and everyone is compiled against libgal10.
This would not be a problem if the libgal package also uploaded, along
with the new libgal source package, legacy libgalNN source packages
which build legacy (section: oldlibs) libgalNN binary packages. But
this approach is not normally used for fast-moving unstable libraries.
Jules
Reply to: