On Sun, 2016-06-05 at 12:22 -0700, Josh Triplett wrote: > [Please CC me on replies. Aside: in my previous mail, I set both > Mail-Followup-To and Reply-To to include myself. What mailer and mechanism did > you use to reply that didn't look at either of those headers?] Evolution, Reply to List (Ctrl-L). > Ben Hutchings wrote: > > On Sun, 2016-06-05 at 07:39 -0700, Josh Triplett wrote: > > > Ben Hutchings wrote: > > > > On Sun, 2016-06-05 at 11:29 +0200, Marco d'Itri wrote: > > > > > On Jun 05, Ben Hutchings <ben@decadent.org.uk> wrote: > > > > > > The postinst script for linux-image-* behaves differently on fresh > > > > > > installation vs upgrade. For a fresh installation, it updates the > > > > > > default symlinks /vmlinuz and /initrd.img to point to the new kernel > > > > > > and initramfs versions. On upgrade it generally doesn't. > > > > > BTW, can we remove these? At least on x86 they should not be useful > > > > > anymore since lilo has bit rotten. > > > > > > > > Unfortunately there are many boot loaders (and custom configurations) > > > > that rely on them, not just lilo. But they can be disabled by adding > > > > 'do_symlinks = no' to /etc/kernel-img.conf. > > > > > > Which bootloaders still rely on them? Is there a list somewhere? It > > > might be feasible to go through and fix those bootloaders. (Not just to > > > avoid needing a hardcoded kernel path, but more generally to support > > > booting more than just the current kernel and possibly one previous > > > kernel.) > > > > > > For that matter, could do_symlinks default to no on i386 and amd64? > > > > Perhaps. I'd rather not make the hardcoded default architecture- > > dependent, but the default set by the installer could be changed. > > Sounds plausible. Which package should I file that wishlist bug on? Possibly base-installer. You could ask on debian-boot. > > > In the meantime, how feasible would it be to extract the logic from the > > > linux-image postinst and put it into a helper program? > > [...] > > > > That would be silly, because the old logic was awful. But I did write > > a helper program and added it to linux-base: > > > > commit af7c52a66d513d7d67abc803d1b36c53e4463bc4 > > Author: Ben Hutchings <ben@decadent.org.uk> > > Date: Mon May 30 18:41:42 2016 +0100 > > > > Add linux-update-symlinks command for use by package maintainer scripts > > > > This is simpler than the current logic in the linux maintainer > > scripts: > > > > - Now that I know that hard-linking and copying to the default locations > > were broken, I only need to worry about making symlinks > > - The logic for updating on installation/upgrade and removal is kept > > together rather than divided across two scripts > > - It uses existing functions in the DebianLinux module > > > > But it's also smarter: it builds a list of versions and the files that > > belong to them, prioritised based on the current type of change, the > > existing symlinks and version sorting. So it can always point the > > symlinks somewhere sensible, unless the last image is being removed. > > > > and the linux maintainer scripts will use that in future. > > Sounds great! > > Would that potentially make it easier to run via dpkg trigger, rather than > postinst, so that it does less duplicate work during an apt run that involves > multiple kernel-related packages? No, I think that would result in worse decisions about what the default kernel version should be. Ben. -- Ben Hutchings The obvious mathematical breakthrough [to break modern encryption] would be development of an easy way to factor large prime numbers. - Bill Gates
Attachment:
signature.asc
Description: This is a digitally signed message part