Re: routine-update builds twice, but does not clean after the first
On 07.12.20 16:53, Nilesh Patra wrote:
> On Mon, 7 Dec, 2020, 9:03 pm Steffen Möller, <email@example.com
> <mailto:firstname.lastname@example.org>> wrote:
> That cleaning I had introduced between the builds somewhen was removed
> and to avoid some ping-pong I thought we should talk about it. I
> this is because of some system setting that runs cowbuilder when I run
> it within the source tree. Can routine-update please somehow test for
> that and give a warning to change the local configuration if we decide
> not to clean?
Ok. I now found also this message that Andreas left me in the source tree:
# This should *NOT* be default! The script is for users who are doing
# builds in a chroot environment and do not want to bloat their local
# with unneeded Build-Depends. Steffen, if you really want this please find
# some command line option to trigger it
#if ! dpkg-checkbuilddeps; then
# echo "E: This machine misses above packages as build dependencies. If
the package is already in Debian then run"
# echo " sudo apt-get build-dep"
# echo " to install what is missing and invoke $(basename $0) again,
perferably from within $(pwd) with no arguments to spare redundant
# exit 1
It is a fairly clear indicator that something works different on my
system tham on Andreas'.
"gbp buildpackage -S" indeed needs all dependencies to be installed,
even though the build (for which all dependencies are installed again)
is happening via cowbuilder. A bit annoying, but I can live with it.
"gbp buildpackage --git-pbuilder -S" would also perform .dsc creating
via pbuilder. So, the root of the problem I dare to locate in expecting
a particularly defined gbp.conf somewhere on the systems. Is that
correct? If so, then routine-update should check for that. Preferably,
it should require that external configuration and directly set all flags
that are required.
> Here what it looks like for me, i.e. a perfect build that has the
> routine interrupted:
> So looking at the last commit message "ready to upload to unstable" it
> looks like routine-update did it's work well.
> What is interrupted is fundamentally gbp instead.
> R-U runs a "gbp build --git-builder=pbuilder" at the end and $gbp
> buildpackage cannot proceed without everything being committed.
I just cross-checked (and updated another two packages over it). It is
not two builds that it is doing, it is just a "gbp tag" after the first
(and only) build. And that build leaves newly created files that should
be cleaned by a "./debian/rules clean" on my system.