Fixes#181.
When initializing the buildroot for a module build, we used to set up
some build 'groups' for the tag: `build` and `srpm-build`. These are
the lists of RPMs that koji is supposed to install into the buildroot
before anything else is done. Crucial stuff goes here, like `git` in
the `srpm-build` group so that koji can clone the repo in the first
place.
We had those lists hardcoded before. This list changes that to use the
`buildroot` and `srpm-buildroot` profiles of the modules which are our
dependencies (recursively).
This will allow people like @psabata and the base-runtime to make
changes to the build groups for the generational core and work around
their own problems, instead of having to ask us to expand that list.
There were a couple ways to do this:
- I could've cloned the SCM repos for all dependencies and gotten their
profiles from the modulemd source there. This seemed flimsy because
we only want to depend on the profiles of modules that were *really*
built.
- We could modify PDC to stuff the modulemd contents in there. We
already get some dep and tag info from PDC. My thought here was that
it would be too heavyweight to store every copy of the modulemd file
in PDC for every build ever. We already have it in MBS.
- Lastly, and this is what I did here, I just referred to MBS' own
database to get the profiles. This seems to work just fine.
One side-effect is that we need the build profiles from the manually
bootstrapped modules that were put together by hand, and were never
built in the MBS. In order to work around that, I added an alembic
upgrade script which pre-populates the database with one fake
bootstrapped base-runtime module. We can expand on that in the future
if we need.
Fixes#203.
The modulemd files we had used to all pin the commit hashes of the
components they were pulling in. We changed this with modulemd-1.0 and
now modules can define their components in terms of their *stream*
(a.k.a. their dist-git *branch*). This is awesome, and is a big
stepping stone towards greater things.
The problem is that we used to take that git ref literally from the
modulemd and pass it on to koji. Now that instead we are passing on a
branch name to koji, it falls down. Koji requires us to send it an
explicit commit hash.
This change takes that branch name and converts it to an explicit commit
hash when the module is first submitted.
Verified that things are working again:
- http://koji.stg.fedoraproject.org/koji/taskinfo?taskID=90084980
This is only one half of the modulemd-1.0 changes needed around PDC.
This was done to get module handling their dependencies when they are
defined in terms of streams instead of NVRs.
In PDC we used to store the dep as one whole string, i.e.::
base-runtime-0.25-11
Now we're going to store them with some more structure, since we can't
assume monotonically increasing version values::
dependency: base-runtime
stream: master
Pull-requests to PDC and PDC-updater are forthcoming.
