On Sat, 2018-12-22 at 20:11:13 +0100, Adam Borowski wrote: > On Sat, Dec 22, 2018 at 05:54:26PM +0100, Guillem Jover wrote: > > On Fri, 2018-12-21 at 23:57:38 +0100, Thorsten Glaser wrote: > > > No. update-alternatives is too fragile to handle things like > > > /bin/sh and init(8). > > > > While this certainly used to be true, I don't it has been the case for > > a long time now? It seems like oral tradition to me now. :) There is I > > think a single non-wishlist bug report against u-a in the BTS. So if > > there are still problems I'd be more than happy to hear about them. > > I believe this is a misconception based on an actual problem: the > alternatives system cannot be used for /bin/sh symlinks as
> 1. it uses shell scripts itself (making it impossible to recover) and I don't think u-a has used any shell in any of its recent history. The only external thing u-a itself it currently still uses is mv(1) via an execvp(3), *only* in case a rename(3) failed for example due to a cross-device move. > 2. is not atomic, leaving a brief time window when there's no /bin/sh. But it should! It uses a temporary file first and then does the usual atomic rename(3) dance over it. If there is any place where this is not the case, I'm also happy to hear. :) > Nothing of that issue applies to /bin/init which is ever executed only > during the boot process then at controlled moment during package updates. It's still as important, as you might suffer from a sudden crash and be left w/o a /bin/init program. Thanks, Guillem
