ensure undefined script_env variables are undefined in multi-output build environment

[jameslamb]

Description

Fixes #5323

"Use Environment Variables" in the conda-build docs says the following about variables in build.script_env in recipes (my emphasis)

If a listed environment variable is missing from the environment seen by the conda-build process itself, a UserWarning is emitted during the build process and the variable remains undefined.

(link)

As of #5105, conda-build does not behave that way for multi-output recipes. As of that PR, any variable mentioned in the recipe's build.script_env but not found in the environment where conda-build is called from will be set to the empty string '' in the build script environment.

This PR proposes ensuring that such variables remain unset in the build script environment.

Thanks for your time and consideration.

Benefits of these changes

Removes a source of behavioral difference between top-level and multi-output builds.

Reduces a source of friction when using build scripts that are sensitive to the difference between environment variables being unset and being set to an empty string. (for example, with optional AWS credentials for sccache like this).

Checklist - did you ...

• Add a file to the news directory (using the template) for the next release's release notes?
• Add / update necessary tests?
• Add / update outdated documentation?

How I tested this

The unit test I've added here fails on main.

git checkout main -- conda_build/build.py
pytest tests/test_subpackages.py::test_build_script_does_not_set_env_from_script_env_if_missing
# Failed: DID NOT RAISE <class 'subprocess.CalledProcessError'>

And prints this message from the build script in stdout of the build process, showing that the unset environment variable is set to ''.

val... 

But passes on this branch.

[codspeed-hq]

CodSpeed Performance Report

Merging #5322 will not alter performance

_{Comparing jameslamb:fix/undefined-env-vars (380c796) with main (6dbdec8)}

Summary

✅ 3 untouched benchmarks

[jameslamb]

I've added an issue write-up (#5323), so this can be documented even if this specific PR isn't accepted.

Thanks for your time and consideration.

[jaimergp]

Thanks for the write-up. I think this looks good to go, but I wasn't involved in the mentioned PR. Tagging @msarahan and @kenodegard to see what they think.

[mbargull]

LGTM!

[jameslamb]

As of #5264, a codepath that was raising subprocess.CalledProcessError started raising conda_build.exceptions.BuildScriptException, breaking the unit test I'd added here.

I've merged latest main into this PR and updated the test. See f7ac317

[jameslamb]

I've updated this to latest main.

Is there anything else you'd like to see on this PR? More tests, docs, etc.?

[jakirkham]

Most of the CI jobs are seeing a known failure, which happens outside this PR and is being debugged: #5426

The macOS Python 3.12 canary jobs see 2 failures. They appear unrelated to this PR (which has been passing in the past). Have raised them as issues:

• Failing test_rpath_symlink: <Signals.SIGABRT: 6> #5429
• Failing test_inspect_hash_input: InvalidTypeError #5430

Given there is support for this change and these issues appear unrelated, will go ahead and merge to fix this bug