Numpy run environment constraint

thomashampson · August 8, 2025, 12:36pm

I’ve pinned numpy to 2.x in the host environment of my package, and am attempting to allow >=2, <=2.x in the run environment by adding the following to the list of run dependencies:

- {{ pin_compatible("numpy", min_pin="x", max_pin="x.x") }}

Once my package is built, I am unable to install anything except numpy 2.x into the same conda environment as my package. I spotted in the build logs that this is coming in and overriding my pin_compatible run contraint:
pin_run_as_build='OrderedDict([('\''python'\'', {'\''min_pin'\'': '\''x.x'\'', '\''max_pin'\'': '\''x.x'\''}), ('\''r-base'\'', {'\''min_pin'\'': '\''x.x'\'', '\''max_pin'\'': '\''x.x'\''}), ('\''boost'\'', {'\''max_pin'\'': '\''x.x'\''}), ('\''numpy'\'', {'\''min_pin'\'': '\''x.x'\'', '\''max_pin'\'': '\''x.x'\''})])'

which appears to come from here:

github.com/conda/conda-build

conda_build/render.py

118f4f820


      
          from .build import get_all_replacements
          
          get_all_replacements(variant)
          mv = metadata.copy()
          mv.config.variant = variant
          # start with shared list:
          mv.config.variants = all_variants
          
          pin_run_as_build = variant.get("pin_run_as_build", {})
          if mv.numpy_xx and "numpy" not in pin_run_as_build:
              pin_run_as_build["numpy"] = {"min_pin": "x.x", "max_pin": "x.x"}
          
          conform_dict = {}
          for key in used_variables:
              # We use this variant in the top-level recipe.
              # constrain the stored variants to only this version in the output
              #     variant mapping
              conform_dict[key] = variant[key]
          
          for key, values in conform_dict.items():
              mv.config.variants = (

I’m wondering if this is correct, given the run_export defined in the numpy recipe is so generous:

github.com/conda-forge/numpy-feedstock

recipe/recipe.yaml

5ff53c6b9


      
              - python-build
              - cython
              - libblas
              - libcblas
              - liblapack
            run:
              - python
              - if: rc_num > -1
                then: _numpy_rc
            run_exports:
              - numpy >=${{ default_abi_level }},<3
            run_constraints:
              # enforce eviction of package from anaconda defaults
              - numpy-base <0a0
          
          tests:
            - python:
                imports:
                  - numpy
                  # reference for public API is effectively PUBLIC_MODULES under
                  # https://github.com/numpy/numpy/blame/main/numpy/tests/test_public_api.py

I know that I can define my own pin_run_as_build entry for numpy to avoid this, but should I have to do that?