Hey Steve,
On Mon, Mar 12, 2007 at 04:38:30PM -0700, Steve Langasek wrote:
> | --- net6-1.3.1/debian/rules
> | +++ net6-1.3.1/debian/rules
> | @@ -1,4 +1,5 @@
> | #!/usr/bin/make -f
> | +#!/usr/bin/make -f
> | # -*- makefile -*-
> Um, doubled interpreter line?
Uh... Gotcha. Sorry, I don't know where that comes from. I will fix that
in a future upload, as luckily enough this doesn't do any harm. *cough*
> > obby changes:
> > * Adopt dpatch:
> > - Add a patch to retry IPv6 if it's unavailable.
> Obby also includes changes to the dependencies of libobby-0.4-dev. Why does
> the libnet6-1.3-dev dependency need to change so frequently? If it actually
> needs to be changed so often, doesn't that mean the dependency will be wrong
> on a simple rebuild of the package?
Well, obby is a library relying heavily on templates. As the newer obby
retries on IPv6, which is specified in the header, it depends on the new
parameter introduced with the net6 upload. Thus the new tighter dependency.
I don't see why the dependency should be wrong on a simple rebuild,
though.
> Anyway, I'm not sure that 'retry' is the correct solution here, but I don't
> have any evidence to the contrary, so I've unblocked both of these.
Binding to the socket fails when the IPv6 socket is tried (which is used
for both IPv6 and IPv4) and kernel support is not present. Thus it's
then safe to retry with IPv4-only. Currently binding just fails with
"Address family not supported" (from mind only), because IPv6 support is
explicitly requested by the API which is not available on the target
platform at times. (Which is more critical with Sobby on a server than
with Gobby on a client...)
> > gobby changes:
> > * Fix mime type mapping.
> > * Lack of avahi-daemon now fails silently instead of a verbose message
> > confusing the user. Zeroconf is a `nice to have' but in no means
> > required.
> I don't see that a freeze exception is justified here.
Well, it's also a rebuild against current net6 and obby, and as the
former two are mostly template libraries the changes only get
instanciated in the final application which provides the selector
template argument, which is Gobby.
Thus the changes in net6 and obby are not visible until Gobby is
rebuilt, and as the two changes are annoying I fixed them on the way.[1]
> > The IPv6 patch added a new default parameter. Tests of mine shows that
> > it was ABI- and API-compatible, but I guess you judge that better than I
> > could.
> Ok, then why have you bumped the shlibs in this version? I guess the 'fatal'
> class is a new exported class?
I guess technically a shlibs bump is not necessary as we did not modify
symbols in the real library, that's right. In fact I only wanted obby to
depend on the newer net6 library for cleanliness, because it needs to
depend on the newer development package.
I already questioned the sense of packaging template libraries because
they export only very few symbols and have the remaining code in headers
which are only instantiated in the final application. This is just as
bad as statical linking. I guess it's too late to merge all the packages
into one big package building the whole package set... But I will
certainly consider that for the time after Etch...[2]
Kind regards,
Philipp Kern
Debian Developer
[1] At "work" I cannot install avahi-daemon because we use local as a
domain suffix. Thus all users get this annoying warning at startup.
And the mime types bug prevented Gobby to provide proper
highlighting for C++ out of the box, although it is supported. This
part of the application depends on the proper specification of the
type.
[2] Except of course that you state that you cannot risk this situation
in Etch and thus decide to remove the packages altogether.
Attachment:
signature.asc
Description: Digital signature