Hello Martin,

Martin Steigerwald [2016-11-30  9:20 +0100]:
> Also agreed to that… libsystemd is almost one third of the size of libc6.so 
> here… and it seems upstream basically stuffes *everything* into it, including 
> reading process attributes that IMHO would be a task for a *different* shared 
> object like the much lighter libprocps.so.6.0.0.
> But the discussion would need to be brought upstream

Please let me clarify that you are talking about two different issues:

  1. libsystemd being too big, which is the part which can and should
  be discussed on the upstream systemd list indeed, as that's not
  something which is appropriate to change downstream. But this is
  entirely unrelated to this bug report.

  I sympathize with this, and maybe the earlier split into three
  smaller libraries was the better choice. And if someone refrains
  from starting the discussion with a tone like "you guys suck, break
  everything, and want to dominate the world", it might even be
  successful :-) (please forgive me the exaggeration)

  2. Debian supporting separate /usr without an initrd, and by
  extension, if boot-critical bits can link to stuff in /usr, which
  this bug is about. This is entirely a downstream Debian packaging
  issue (we could move liblz4 to /lib, like we did with other
  libraries), and does not belong on an upstream ML.

  Personally I think this part is a lost cause, both for technical
  reasons that have existed for a long time (which I will not repeat
  here), and even more so because the ongoing move to the /usr merge
  will make this completely obsolete -- if the *entire* OS is in /usr,
  then there is no alternative to an initrd anyway. This will finally
  make the whole design simpler, robust, maintainable, and reduce
  combinatorial explosion.


Martin Pitt                        | http://www.piware.de
Ubuntu Developer (www.ubuntu.com)  | Debian Developer  (www.debian.org)

Reply via email to