-
-
Couldn't load subscription status.
- Fork 560
Description
Some migrators might need to set a higher c_stdlib_version, c.f. conda-forge/conda-forge.github.io#2467
However, we really need to avoid the bot breaking our stdlib-pinning by naïvely closing such migrations (c.f. #5559 #7304). I think we should add some tests that ensure that the content of
conda-forge-pinning-feedstock/recipe/conda_build_config.yaml
Lines 21 to 24 in 3a22222
| c_stdlib_version: # [unix] | |
| - 2.17 # [linux] | |
| - 10.13 # [osx and x86_64] | |
| - 11.0 # [osx and arm64] |
contains what's expected. This is because changes like
-c_stdlib_version: # [unix]
- - 2.17 # [linux]
- - 10.13 # [osx and x86_64]
- - 11.0 # [osx and arm64]
+c_stdlib_version: # [osx and x86_64]
+ - 10.14 # [osx and x86_64]in e04df46 actually cause {{ stdlib("c") }} to pull in to pull in the highest available version, which on linux (as of today) leads to packages requiring __glibc >=2.34, which is waaaaaay newer than what we currently have or want (see conda-forge/conda-forge.github.io#2383). Same for osx-arm64.
Duplicating these pins in some pre-commit tests is annoying duplication, but given the blast radius of the bot messing something up in that configuration, we really should add a safety net IMO.