The multiarch conundrum
Hi,
I've just been trying to dig through some of the multitude of past discussions
on the planning and implementation of multiarch in Debian (and Ubuntu).
The short version of this for TLDR folks is that nobody seems to have coherently
written down all the problems in a way that they can be addressed, so far as I
can see.
The longer version:
It seems clear that there is no consensus yet on what exactly needs to be done,
let alone how to get there; discussions seem to go round in circles, then
eventually peter out to begin anew at a later date. If there is anywhere where
real progress is being made, then I can't find it, but I think a large part of
the problem is down to poor communication.
My proposal for one way to try moving forward with this is as follows:
Currently the collective discussions about multiarch have taken place in
numerous threads on numerous mailing lists, which makes it hard to find any
specific information.
I believe that a wiki or similar centralised collaborative process is a good way
of tying this together. If there is already some location where this is being
actively worked on in such a fashion then I apologise, and much of this mail
will be rather pointless - in which case that location really ought to be easier
to find :P.
Start with https://wiki.ubuntu.com/MultiarchSpec. The reason for choosing the
Ubuntu wiki page is that there is already some real content there, in an
organised form. If http://wiki.debian.org/multiarch is a better place then fine,
but either way it would be good to have the same content, or at least not
disagreement between the two. Ideally one would act as a pointer to the other -
this is assuming that everyone wants Ubuntu and Debian heading in the same
direction here, which seems to be the consensus.
Within the spec page there are sections describing the design and some notes on
the implementation. One major point which is missing is that each section needs
some description of the objections to it - which surely exist, otherwise we'd
already have multiarch by now, wouldn't we? Currently it looks more like 'well
that's a nice plan; why isn't it happening?'
Further, there has already been some work done which has been rejected or
ignored - this needs to be presented and the reasons documented; it doesn't
really help anyone to forget about past attempts. So far I've found the patches
by Goswin on the Alioth multiarch project page[0]; there must be more scattered
about the place?
Finally, there needs to be some organisation of the many existing threads on the
topic. A good first start would be to search through the list archives and add
links to related discussions on the wiki page - we need to be able to go to one
place and get an overview of the issues, proposals, and objections, without
having to hunt that information down. This way we can hopefully avoid going over
the same issues and making the same mistakes and arguments repeatedly.
If there is no objection, I would like to do the following:
* Merge multiarch information from the Debian wiki into the Ubuntu wiki.
(I'm not too sure about tho organisation of the Ubuntu wiki; would the spec
page be inappropriate for this? Where would be better?)
* Add pointers to any existing patches, and links to any discussions about them
if I can find them in the ML archives.
* Start a list of related discussions and dig out archive links from whatever
mailing lists are appropriate. So far I can think of debian-devel,
debian-dpkg, debian-glibc (?), debian-amd64. Are there others? I don't know
about any Ubuntu lists but I imagine it's been discussed to death there too.
* Start monitoring those lists for multiarch-related discussions and update the
wiki appropriately.
* Possibly stub out some empty 'objections' sub-headings in the spec design, in
the naive hope that somebody reading will see them and think 'no objections?
Lies! I have objections!', and actually fill them in.
Does anybody have opinions on whether any of this would be even remotely useful?
-Nye
[0] https://alioth.debian.org/projects/multiarch/
Reply to: