Standalone: set PYTHONPATH=AIRFLOW_HOME to match runtime image

[jeremybeard]

Summary

Aligns standalone-mode PYTHONPATH with the production Astro runtime Docker image.

Standalone mode previously set PYTHONPATH=AIRFLOW_HOME/include only. That diverged from the runtime image, which sets PYTHONPATH=AIRFLOW_HOME (astro-runtime/image/Dockerfile#L19). This change brings standalone in line with the image.

Why

The previous behavior was introduced by #2060 with the stated intent of "mirroring the docker image" — but the docker image actually bakes AIRFLOW_HOME into PYTHONPATH, not AIRFLOW_HOME/include. The comment in the original code was also factually inaccurate. So in practice, imports that work in production (and pass astro deploy --parse) could fail when the same project was tested locally with astro dev start --standalone or astro dev pytest --standalone.

Concrete example that breaks under the current code but works in production:

# dags/hybrid_dag.py
from dags.my_blueprints import Extract  # works in prod, fails under --standalone

The prefixed from include.<module> import … pattern that the official docs recommend (Custom hooks and operators) continues to work both before and after this change.

Behavior change

• ✅ from include.foo import bar — continues to work (include/ is findable as a namespace package under AIRFLOW_HOME).
• ✅ from dags.foo import bar — now works (matches production).
• ⚠️ import foo where foo.py lives flat in include/ — no longer resolves under standalone. This only ever worked under standalone (never in docker mode), so the breakage surface is small. The fix for users is to switch to from include.foo import …, which the docs already teach as the canonical pattern.

Test plan

• go test ./airflow/ -run 'TestAirflow/TestStandaloneBuildEnv' (12 subtests, all pass)
• Verified test is robust to inherited PYTHONPATH: PYTHONPATH=/some/inherited/path go test ./airflow/ -run 'TestAirflow/TestStandaloneBuildEnv$' passes (the test now calls t.Setenv("PYTHONPATH", "") so the strict equality assertion doesn't flake on developer/CI envs)
• End-to-end: built a local binary, ran an integration test suite against it that uses from dags.<module> import … in a Python DAG under astro dev pytest --standalone. Live scheduler loads the DAG; astro dev pytest --standalone (DagBag scan) passes

[coveralls-official]

Coverage Report for CI Build 8

Coverage increased (+0.002%) to 39.653%

Details

• Coverage increased (+0.002%) from the base build.
• Patch coverage: 1 uncovered change across 1 file (6 of 7 lines covered, 85.71%).
• No coverage regressions found.

Uncovered Changes

File	Changed	Covered	%
airflow/standalone.go	7	6	85.71%

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	66078
Covered Lines:	26202
Line Coverage:	39.65%
Coverage Strength:	9.38 hits per line

---

💛 - Coveralls