Repodata patch for 2.6.0?

[h-vetinari]

Just saw that conda-forge/staged-recipes#16888 pulls in

  + libtensorflow                   2.6.0  cpu_hf74009c_2          conda-forge/linux-64       54 MB
  + libtensorflow_cc                2.6.0  cuda111h67a8850_2       conda-forge/linux-64      319 MB

for a recipe that says:

    - tensorflow {{ tf_version_maj_min }}.*
    - libtensorflow_cc {{ tf_version_maj_min }}.*
    - libtensorflow {{ tf_version_maj_min }}.*  # [cuda_compiler_version != "None"]
    - tensorflow-cpu  # [cuda_compiler_version == "None"]
    - tensorflow-gpu  # [cuda_compiler_version != "None"]

Since we added some run-exports in 2.6.0, it could be that the resolver just takes the "lazy" option of taking the unconstrained 2.6.0.

What do you think is the best approach here @xhochy? I can fix the other recipe, but if someone else is pulling in tensorflow and a library anywhere, these might be mismatched for 2.6.0 (AFAIU).

[h-vetinari]

Well, at least one additional reason for not taking 2.6.2 turns out to have been the default libprotobuf version in staged-recipes...

[hmaarrfk]

oh. yeah. you might need to have manual migrations.