Open CJ-Wright opened 5 years ago
On a related note ABI tracker puts up it's reports as json
. See: https://raw.githubusercontent.com/lvc/abi-reports/master/report/libtool.json
We may be able to use this in the linter/cron job to let people know if new migrations may be needed.
IMO, there's little value checking against ABI laboratory packages. If you want to pursue this, you should look at packages that has a shared library in PREFIX/lib/
, but doesn't have a run_exports
.
What do you mean "little value checking against ABI laboratory"? Are the packages there out of date? Are the reports incorrect?
If you want to list packages that might need pinning, there's no point in looking at ABI laboratory. They have far fewer packages.
This was not a comprehensive approach for sure. I don't know if everything that ships a shared library needs pinning in practice, especially if they are extremely stable. I mostly looked at ABI lab because it is the only tooling I know of which will let us know if a new version breaks backwards compat. Are there other tools which we should use?
As a side note the ABI lab people seem to be willing to add packages to their tracker as needed, arrow just went up the other day.
Would change protobuf
to libprotobuf
. The former is the Python library; so, doesn't need an ABI pin. However the latter is the C++ library, which does.
I think that any application of this would need a LUT in the middle to translate between our names and ABI lab's.
Here are all the packages ABI lab tracks and we ship, but are not in the pinning repo. (sorted by number of descendants)
I'm not saying we should pin all of these, but I did want to raise awareness. There are 184 feedstocks here.
@conda-forge/core thoughts?