The TARGET_DIR values in just these few nightly scripts have
trailing slashes. They shouldn't, it'll lead to double slashes
in all the places that use TARGET_DIR as they expect it not to
have one.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
We set $DESTDIR in all these scripts, but then use its value
several times without just...using it. Let's use it!
Signed-off-by: Adam Williamson <awilliam@redhat.com>
For fedmsg_meta we have to provide a 'link'. At present we do
this with a rather large and icky pile of conditionals which
doesn't work properly. It's much simpler to figure out the URL
in the compose script and emit it in the fedmsg, then meta can
just...use it. For older fedmsgs we'll just point to the base
kojipkgs location for the link.
For synced composes, we use the synced URL on dl (this matches
what the meta stuff tried to do before). For non-synced composes
we just use the kojipkgs location.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
The main point here is to include the compose 'shortname' in
compose fedmsgs, as #484 does for post-release compose types.
While doing this I noticed there's rather a lot of mess and
cruft in the various version-related variables here (and a few
others).
`GIT_BRANCH` variable is never used anywhere any more, so it's
removed.
`BRANCH` was never used anywhere, so it is gone. (It appears
to have been added in a25b08c7 but never used). There seemed
to be no useful distinction between `DIST` and `BRANCHED`, and
both actually seem to be used mainly to specify a release
version. The exception is that `BRANCHED` was set to
"Modular-Bikeshed" in nightly-modular.sh , but there is no
justification for this that I can find, and its only practical
use was as the value of the `branch` key in fedmsgs. So far as
I can figure out, the original purpose of the `branch` key was
for when we sent out `compose.branched.(foo)` fedmsgs - it
was e.g. 'f18' when Branched was Fedora 18. As we now don't
send out `compose.branched` messages any more, but include the
release number in the message topic for Branched messages
(e.g. `compose.27`), this purpose no longer applies. So the
key just doesn't seem to have any particular use at all any
more, and I don't think changing its value will have any real
consequences.
Thus, `BRANCH`, `BRANCHED` and `DIST` are replaced by `RELEASE`
and `RELEASE_TITLE`, which specify the release version. Their
values will be identical for numerical release versions, but
for Rawhide and Bikeshed we need to know both the lower-case
and title-case versions of the name; RELEASE_TITLE is the
title-case version.
We use `$SHORT-$RELEASE_TITLE` to replace several instances
of hardcoding in the scripts (e.g. Fedora-Modular-Bikeshed).
The `ATOMIC`, `ATOMICREPO` and `ATOMICDEST` variables that were
blindly copied from nightly.sh to nightly-modular.sh, but never
actually used in it, are also removed.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
When nightly-modular.sh was added, it wasn't set to check the
exit code from pungi-koji, as nightly.sh does. This was I guess
because it didn't do much after the compose finished at that
time, and the author thought it was OK for the few things it
*did* do to happen on both successful and failed composes.
However, it now does more stuff after compose completion, like
nightly.sh, including trying to generate changelog files and
send out a compose report email. But because the pungi-koji
exit code check is missing, it tries to do these things even
for failed composes. Changelog generation will fail when the
compose failed, and the email that gets sent out will have a
missing compose ID (because NEWCOMPOSE_ID and SHORTCOMPOSE_ID
can't be set properly and will be blank), and will be empty
because the changelog file that's supposed to be read in as the
content of the email isn't there. This is why we are getting
an empty mail with subject "Fedora Modular bikeshed compose
report: changes" every time a compose fails.
So, let's put that check in just like nightly.sh does.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
The nightly scripts try to send an announcement about compose starting.
As this happens before Pungi even started, the compose ID is not known
and can not be included in the message. The message about starting rsync
sent later will contain the ID correctly.
Without the json defined there is a traceback in the cron output, but
the errors are not logged anywhere else (but don't actually block
anything in the compose process).
Relates: https://pagure.io/pungi/issue/698
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
