search

regro / cf-scripts

Flagship repo for cf-regro-autotick-bot
Other
49 stars 74 forks source link

openmpi5 migration failed to make PRs for libnetcdf and libpnetcdf #2694

Closed akrherz closed 4 months ago

akrherz commented 4 months ago

My sincere apologies if this is not the appropriate place/repo to report this issue, but @xylar and I noticed that the on-going openmpi5 migration missed libnetcdf and libpnetcdf. I can't see these on the status page and I checked a recent bot-bot run without any logs that I could see.

manual migrations made here: conda-forge/libnetcdf-feedstock#192 conda-forge/libpnetcdf-feedstock#7

PS, the new conda-forge status page is so great, I cried with joy when I first saw it.

beckermr commented 4 months ago

This may in fact be the right issue tracker. We need to check the bot metadata of the feedstocks to see if they have openmpi listed properly. Then we can look for why a PR may or may not have been made. The fact that those feedstocks are missing on the status page is a sign of some deeper issue for sure.

Please direct all kudos on the status page to @jaimergp and the others who made it. I had nothing to do with it and it is indeed very nice.

jaimergp commented 4 months ago

I found these two files:

node_attrs has no matches for openmpi (just a selector). It does have mpich.

pr_info mentiones the openmpi variants in the curl8 unsolvable errors...

Maybe a bug processing variants?

jaimergp commented 4 months ago

the others who made it

Darian and Gabriela from QuantStack! Amazing work really :)

minrk commented 4 months ago

fftw was also skipped, and doesn't show openmpi (or mpich) in its dependencies

beckermr commented 4 months ago

yep indeed looks like a bug when the bot parses the feedstock. :/

beckermr commented 4 months ago

So the bug is due to this change here: https://github.com/regro/cf-scripts/pull/1579

For reasons that are hard to reconstruct, we place the conda_build_config.yml next to the recipe while parsing. This was needed to fix other things, but now the code is picking up more variants than it should and it looses the mpi ones when it collapses them back down. I'm working on a fix.

beckermr commented 4 months ago

I merged a fix that produces the correct metadata. This won't propagate into the bot metadata for several hours at least.

beckermr commented 4 months ago

The PR for fftw was issued so I think we solved this. The bot may not get back to libnetcdf and libpnetcdf or might mark them as done manually. I am going to close for now. Feel free to reopen if you find issues.