Bringing in the debhelper maintainers into the loop again. On Sun, Jan 13, 2019 at 10:31 AM Dmitry Bogatov <kact...@debian.org> wrote:
> > [ Do you want me to re-submit this patch as merge request? ] > Not necessary (yet) > > [2019-01-11 14:58] Felipe Sateler <fsate...@debian.org> > > > [2011-06-14 11:39] Joey Hess <jo...@debian.org> > > > > > > > > part 1 text/plain 618 > > > > Helmut Grohne wrote: > > > > > A failure from the update-rc.d cannot make postinst fail, because > the > > > > > exit code is not checked. > > > > > > > > Yes it is (set -x), but I meant invoke-rc.d anyway, obviously > > > > > > > > So in my view the reason for a postinst failure is unrelated to > > > > > update-rc.d. Can you explain your reasoning? > > > > > > > > Simple separation of concerns, invoke-rc.d is responsible for running > > > > the init script and determining if it fails and propigating a failing > > > > exit status. debhelper allows it to do do. The right way to make 6 be > > > > ignored is to get invoke-rc.d to ignore it, not add cruft to every > > > > postinst that calls it. > > > > > Here I propose patch to implement just that: > > > [...] > > I'm wary of unintended consequences here. Do we have services that return > > exit code 6 but as a failure code? > > Seems there is nobody, who use exit code 6 for another purposes. You can > make sure by checking is output of 'grep -R "exit 6" -C6' in directory > with all init scripts unpacked: > <snip> Rereading invoke-rc.d manpage, it says quite explicitly: > Should an init script be executed, invoke-rc.d always returns the status code returned by the init script. So it looks like this would be quite the change in behavior. I wonder what the best approach it. OTOneH, I sympathize with Joey's argument that these checks should be centralized. OTOtherH, I'm not sure we can consider invoke-rc.d a debhelper implementation detail and change behaviour like this. Maybe a new flag to invoke-rc.d could be used for this? I'm not sure. -- Saludos, Felipe Sateler
