Conda build environment variables not set during build

Question

Conda build fails to set some build environment variables during the build process.

During build I get the following error message

%SRC_DIR%>et "CONDA_BUILD=1"
'et' is not recognized as an internal or external command,
operable program or batch file.

Inspecting the build_env_setup.bat (temporary file in build work folder) it is suppose to set "CONDA_BUILD=1" .

That is, it has somehow lost the s from set . This happens for roughly half of the set calls in the build_env_setup.bat .

I have tried building on another Windows machine and (using the equivalent build.sh ) on Debian (wsl) and it works. So it seems to be an error on my machine. Uninstall + install Miniconda did nothing.

What can cause this error and how may I resolve it?

Update / Extra Info

• This behavior is observed using PowerShell 7.1.3. When using CMD all the variables are set.
• The generated build_env_setup.bat has Unix style line-endings (LF) changing to CRLF make the build_env_setup.bat script work.
• There is a Ø in one of the paths set (on my system) removing that line also makes the (remaining) build_env_setup.bat script work.

Conclusion seems to be that there is some string encoding / reading mismatch when running the various build scrips using PowerShell in the (edge?) cases with special characters in paths.

---

Old formulation

I am attempting to build a conda package from a standard src-layout python library but my builds keeps failing when using a Windows build script bld.bat with the error message:

'(_h_env)%SRC_DIR%>"" setup.py install
'""' is not recognized as an internal or external command,
operable program or batch file.

This seems to indicate that the %PYTHON% environment variable is not defined. (Which I was assuming to exist based on Environment variables and Writing the build script files )

If I add a build section to the meta.yaml :

build:
  noarch: python
  script: "{{ PYTHON }} setup.py install"

Then the build works.

What is the reason for this behavior and where is the error when using the build script?

Setup

• OS: Windows 10
• conda: 4.13.0
• conda-build: 3.21.9
• python: 3.10.4

Folders and files

│   setup.py
│
├───conda_recipe
│       bld.bat
│       meta.yaml
│
└───src
    └───conda_build_test
            module.py
            __init__.py

bld.bat

"%PYTHON%" setup.py install
if errorlevel 1 exit 1

meta.yaml

package:
  name: conda_build_test
  version: {{ data.get('version') }}

source:
  path: ..

requirements:
  build:
    - python
    - setuptools
  run:
    - python

Answer 1

Notice that the documentation on environment variables states that %PYTHON% is only defined when python is specified as a host requirement (eg, in a cross-compiling situation). In a native build situation, where python is directly in the build requirements, the build environment's bin/ will be on PATH and one can use python directly. That is,

bld.bat

python %SRC_DIR%/setup.py install

For a comparable example, the Conda Forge multineat feedstock recipe uses a setup.py -based build.

Alternatively, you can place the python requirement under host instead of build and that should enable use of %PYTHON% variable.

I'm not entirely clear on why there isn't a failure after the meta.yaml edit. Somehow the {{ PYTHON }} jinja template must not be the same as the environment variable. However, the examples on Conda Forge I find that use a similar {{ PYTHON }} setup.py install seem to also list python as a host requirement (eg, wfdb feedstock ).