On Thu, Jan 8, 2015 at 9:39 PM, Ludovic Courtès <ludo@gnu.org> wrote:
> That said, this kind of patch is not very different from the automated
> shebang patching that we do, IMO.

I do not see it this way because the shebang concept is fully
standardized, while here we are patching a "random" piece of code.

> The reason I’m not thrilled by the use of ‘propagated-inputs’ here is
> that that’s not what they were designed for at the beginning.
>
> It may also lead to usability problems.  Remember that when A is a
> propagated input of B, installing B in a profile also installs A in that
> profile.  So, if each GTKish package propagates sound-theme-freedesktop,
> then installing several such packages in the same profile is likely to
> lead to collisions just because the themes differ (those collisions may
> be harmless, but they will at least trigger a bunch of warnings every
> time the user operates on their profile, which is not desirable.)

Thanks for explaining your concern. But doesn't this problem exist
with most uses of propagated-inputs? As one example libssh2 propagates
libgcrypt. If the user installs a different version of libgcrypt (with
a possibly different ABI), doesn't this causes problems (required
version not found, or worse)?

In any case I've updated libcanberra with a patch to find the default
sounds without the need for propagated-inputs. Differently from your
suggested patch I'm proposing to add the default sounds store
directory to the code dealing with XDG_DATA_DIRS. This is because
XDG_DATA_HOME can only be a single directory and is inspected first.
XDG_DATA_DIRS can list an arbitrary number of directories and is only
inspected later. This is designed to allows the user to modify any
theme at his pleasure.

Regards,
Fede