Re: [Gnu-arch-users] crypto signing take 2

[Tom Lord]

    > From: Robert Collins <address@hidden>

    > On Tue, 2003-12-09 at 07:51, Tom Lord wrote:
    >>   1) no special handling of push-mirror

    > I quite liked the special handling - it allowed cute things. Any
    > reason for it's removal?

Really I have no objection to working it back in later.  It just isn't
needed to have useful functionality and I'm not clear about why it
would be important.   What "cute things" do you have in mind?

One problem with it is that it violates a current abstraction barrier:

As proposed (in take 2) signing is essentially hidden within the
transport layer (archive-pfs.c downward).

libarch/archive-mirror.c(mirror_revision) access the transport layer
only indirectly via libarch/archive.c calls such as
arch_archive_put_log.

So when the transport layer gets a file from push-mirror, the
transport layer has no association of that file with the signed copy
in the archive-being-mirrored.

To change that will require adding new parameters to functions like
arch_archive_put_log and then propogating those down to the transport
layer.   But to do so violates the idea that the libarch/archive.c
interface should know nothing about signing.

If we decide to add special push-mirror handling of signatures --
we'll have to think carefully about how to do it cleanly.


    > (Oh, archive-fs.c is -just about- dead here
    > (address@hidden/tla--gpg--1.1)

Wow, that's good news!

-t