[RFC] base-files: add kernel hook for /boot freespace

[abenson]

<leah2> we should make a kernel install hook check that prints a big
        warning if /boot is full after running it...

Thoughts:

1. Change the threshold size? Make it configurable?
2. Change the threshold to a percentage? I think that's less useful.
3. Change what package this is shipped with? I'm not sure if it applies to all configurations we support/consider.
4. Wording of the message?

[classabbyamp]

+1 for the concept

1. Change the threshold size? Make it configurable?

100MB might be a little bit big? In my /boot each kernel + initramfs + config is ~30MB (zfs and mkinitcpio so that might be different from the usual). maybe 50 or 60 MB?

2. Change the threshold to a percentage? I think that's less useful.

agree that it's less useful because it should be the size of 1-2 kernels, IMO

3. Change what package this is shipped with? I'm not sure if it applies to all configurations we support/consider.

Only the generic kernels keep old ones around, right? so maybe it could go in linux-base, which linux{,-lts,-mainline} all depend on.

4. Wording of the message?

lgtm

[abenson]

100MB might be a little bit big? In my /boot each kernel + initramfs + config is ~30MB (zfs and mkinitcpio so that might be different from the usual). maybe 50 or 60 MB?

I based it on dracut default, which looking at my systems have each kernel+initrd pair at about 90M.

% ls /boot/vmlinuz* | while read ver; do
  du -csh /boot/*$(xbps-uhelper getpkgversion $ver)* | grep total; done
91M	total
91M	total
91M	total
92M	total

maybe it could go in linux-base

That sounds much more appropriate than base-files.

[triallax]

I based it on dracut default, which looking at my systems have each kernel+initrd pair at about 90M.

That number is sure to increase in the future, how can we ensure the limit is updated in a timely manner?

[0x5c]

I based it on dracut default, which looking at my systems have each kernel+initrd pair at about 90M.

That number is sure to increase in the future, how can we ensure the limit is updated in a timely manner?

Considering a move to mkinitcpio seems inevitable, the only update for a long while might be a lowering of the value

[0x5c]

I really like that approach, should help prevent silly systems breakage

[ahesford]

The threshold need not be configurable. Just make it 2x (the size of the largest kernel plus the size of the largest initramfs).