migrator report on mpich-3.3

minrk commented 5 years ago

I opened a migrator for mpich 3.3 (https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/301), and it hasn't gone as expected. I thought I would open this one as a composite issue report for the issues encountered. This might be able to be split into discrete bug reports, but I don't have enough understanding to know which pieces are separate.

ptscotch and mumps-mpi were not included in the graph, and thus didn't receive PRs. Their downstream packages (petsc) also were not treated as having dependencies that needed waiting for. Is this because these are from multiple-output feedstocks that do not match the package name? Manual PRs were issued for these repos, but had to be coordinated manually with other PRs due to propagation of dependencies down the graph
petsc began the process in the 'done' state, despite being built only mpich 3.2 and no PR was issued (manual pr)
hypre was excluded from the list for an unknown reason, and received a manual PR. Is it possible the presence of this PR caused it to be excluded?
vic received a PR, despite already being upgraded to mpich 3.3

CJ-Wright commented 5 years ago

Yes we didn't handle multiple outputs when that migration was issued, there is an open pr for output support #659
I'm not certain why petsc was not issued.
It seems that hypre didn't have a jinja2 variable for mpi until very recently. Prior versions of the graph did not list mpi requirements. https://github.com/regro/cf-graph-countyfair/blob/6ece7aa5419e94424b2e3d7a4b36084c74710bea/node_attrs/hypre.json
How did it get a bot migration for mpich 3.3 in Sept? We only starting issuing them a few days ago. The bot checks only if a rerender changes any of the files. If it does not then the bot assumes that migration has been done for that feedstock, it seems though that a rerender did change things so it issued a migration

minrk commented 5 years ago

Thanks for the knowledge!

How did it get a bot migration for mpich 3.3 in Sept?

There was a previous migrator with the pre-cfep-9 style. I thought that the new approach was what supported multiple outputs, but clearly I misunderstood and that was actually #659

CJ-Wright commented 5 years ago

@minrk is there a reason that scotch doesn't have mpi as a host dep?

minrk commented 5 years ago

is there a reason that scotch doesn't have mpi as a host dep?

scotch doesn't depend on mpi, the ptscotch output does.

CJ-Wright commented 5 years ago

Ah ok, that was running into the "we don't have a copy of the conda_build_config" problem.

regro / cf-scripts

migrator report on mpich-3.3 #661