[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#400952: rc order of portmap,nis,autofs



Hi Luk,

> So, that's something we don't want to do, certainly not at this stage of
> the release cycle.
> 
> What's wrong with Steinar's suggestion to change the name of the autofs
> script to be something between 19nis and 20apache?

conclusion first: If that's the ultimate response of the release team, I'll
accept it.

But IMO this solution is an ugly hack and highly counterintuitive - init
scripts are config files after all, and if I wanted to adapt a package's
initscript to my needs, I'd expect to find it at /etc/init.d/${package},
not /etc/init.d/zz${package} or the like.

Here's a small analysis to support my cause, done on yesterday's i386 Sid.
This is only the result, the log can be found here[1].

* There are currently 16 packages that relate in any way to the portmap
  package (including even Recommends and Suggests, and passing over main,
  contrib and non-free (the latter two containing none)).
* Of these, 12 provide an initscript.
* 11 of these 12 install theirs with a priority at or above 20, the only
  exception being nis at 19; none of these will be affected by the proposed
  change.

Add to that the fact that we are moving the portmap script to a lower value,
so the only type of problem could be services that conflict with the portmap
daemon starting earlier... OTOH the portmapper is usually already started in
rcS, so any conflicts of this kind should have surfaced years ago.

I can understand your reluctance to accept such a change at this time, but I
think it is unjustified in this case. Hacking around the alphabetical order
of init scripts by arbitrary renaming seems like the Wrong Thing[tm] to
me... and I can see no danger at all of causing RC (or any other) bugs. I'll
prepare an upload nonetheless, in case you insist.


Regards,

Jan

[1] http://www-pool.math.tu-berlin.de/~hesso/portmap_relations.html

Attachment: signature.asc
Description: Digital signature


Reply to: