| 1 |
On Tue, Jul 26, 2011 at 11:08:15PM +0200, Michał Górny wrote: |
| 2 |
> As Diego pointed out in bug #374807 [1], a simple package (not a kernel |
| 3 |
> module) should not enforce any kernel options during the build process. |
| 4 |
> As of no reply on the bug, I'm posting my question here. |
| 5 |
> |
| 6 |
> Should it just use '~' and warn then? Or should it re-do the checks |
| 7 |
> in pkg_preinst(), this time making the checks fatal? Or maybe it should |
| 8 |
> only perform them there? |
| 9 |
> |
| 10 |
> Also, I think I've seen ebuilds doing kernel checks in pkg_pretend(). I |
| 11 |
> did that once too but now I think that's wrong because kernel sources |
| 12 |
> are not guaranteed to be installed at that point. I am right? |
| 13 |
When packages other than kernel modules are being installed, there is |
| 14 |
absolutely no guarantee that kernel sources will ever be on the machine, |
| 15 |
esp. in larger deployments. There's also related cases (mainly on |
| 16 |
hardened) where it's impossible to check the actual kernel config, the |
| 17 |
data just isn't there. |
| 18 |
|
| 19 |
Do the checks exactly once, probably at pkg_preinst, and issue warnings |
| 20 |
if they fail due to not being set, or cannot be checked. At runtime, |
| 21 |
fail sensibly if the options in question aren't usable (autofs4, |
| 22 |
cgroups, devtmpfs should all be listed in /proc/filesystems). |
| 23 |
|
| 24 |
Failing during preinst isn't an option, for example if I'm installing |
| 25 |
while in a chroot in a different kernel (think of installing a Gentoo |
| 26 |
system, before the first reboot). |
| 27 |
|
| 28 |
-- |
| 29 |
Robin Hugh Johnson |
| 30 |
Gentoo Linux: Developer, Trustee & Infrastructure Lead |
| 31 |
E-Mail : robbat2@g.o |
| 32 |
GnuPG FP : 11AC BA4F 4778 E3F6 E4ED F38E B27B 944E 3488 4E85 |
Archives