kernel-package postinst hook.

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

kernel-package postinst hook.

Sven Luther
00:01 < joeyh> I noticed that postinst_hook = /sbin/update-grub leaks stdout
into debconf and causes it to exit nonzero in at least some circumstances
00:02 < joeyh> d-i sets that by default in kernel-image.conf, is this
something that is being dealt with?
00:03 < waldi> joeyh: kill ola or nmu it
00:04 < joeyh> ola?
00:04 < waldi> maintainer of grub
00:07 < waldi> he promissed an upload some days ago
00:08 < fjp> joeyh: #344767 marked as pending
00:08 < joeyh> the right fix is really to fix grub?
00:08 < waldi> yes
00:08 < joeyh> hm
00:11 < waldi> is there another fix excluding fix debconf to not longer use 0
and 1?
00:11 < joeyh> you could just run postinst_hoot with >&2
00:12 < joeyh> the chances someone will want to use debconf there seem to be
zero to me
00:12 < waldi> no
00:13 < joeyh> given that standlone programs that udeb debconf are rather
heavily frowned on
00:19 < joeyh> but hey, "<waldi> no", that resolves any questions I may have
had

Joeyh, mkmvmlinuz, which is depended upon by all powerpc kernels, does use
debconf to chose if it should produce the zImage-like kernel or not, in case
there are multiple boot-loaders. MAnoj tried to do it like you around
sylvester, and it resulted in 2.6.14-6 (iirw) being not only uninstallable on
powerpc, but freezing the upgrade process and generally leaving the dpkg
database in a hosed state when aborting, that required manual intervention in
the postinst scripts to fix, the worst possible solution. I believe this was
RC severity, and Manoj backed out this change again.

Furthermore, the policy says that all packages doing install time interaction
should use debconf, and now that kernel package is properly debconfified,
mandating the hooks to do interaction the non-debconf way is not something you
want to do, especially if you consider d-i, on top of it being
policy-breaking.

And i believe, but i have only Manoj's word for it, that Manoj asked you
about this bug last week while we where trying to solve this same problem,
altough it may well be that he only tried contacting you and that his emails
got lost or you are still wading through your email backlog :)

In any case the k-p changelog should be explicit enough about the issue.

Friendly,

Sven Luther


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]