conda-forge.yml matrix ignored on rerender?

[minrk]

I'm trying to test rerendering with conda-smithy master and it's failing in a few cases. The first example I tried was https://github.com/conda-forge/mpi-feedstock which is an empty metapackage with two MPI_VARIANT matrix entries in conda-forge.yml. When I rerendered, I got only a single config_.yml without the MPI_VARIANTs.

Is this a bug in rerender or is there an extra step I'm missing to migrate conda-forge.yml matrix entries to conda_build_config.yml?

[isuruf]

Add a recipe/conda_build_config.yaml file with

mpi_impl:
  - mpich
  - openmpi

and you can use mpi_impl as a jinja variable in the meta.yaml.

conda-forge.yml's matrix is removed now. (We should warn that it is removed though)

[jakirkham]

Did this go in the wiki too?

[isuruf]

Yes, see https://github.com/conda-forge/conda-smithy/wiki/Release-Notes-3.0.0.rc1

[jakirkham]

Great, thanks.

[minrk]

Perfect, thanks. Manual step is AOK; a message about what I needed to do on rerender would be great (just a link to that wiki page would have been enough for me).

[jakirkham]

Thanks @minrk. Appreciate your help testing.

Think we will move the wiki content to the docs once we get closer.

We can certainly add a warning (maybe even error?) when the old cba matrix feature appears.

[minrk]

An error like "Can't rerender until you migrate conda-forge.yml matrix to conda_build_config.yml" sounds like a good idea to me, since the rerender result will always be wrong and need to be redone anyway.

[isuruf]

Can you send a PR? You can check for matrix in https://github.com/conda-forge/conda-smithy/blob/master/conda_smithy/configure_feedstock.py#L588

[minrk]

Sure, I'll give it a go.

[minrk]

#702