Download backend/tests/test_codegen_preflight.py from firebird-technologies/blog2video-cpu: direct link, hf CLI and curl.
- Browser
- Download file 2.82 kB
-
https://huggingface.co/spaces/firebird-technologies/blog2video-cpu/resolve/main/backend/tests/test_codegen_preflight.py
- Command line
-
hf download hf://spaces/firebird-technologies/blog2video-cpu/backend/tests/test_codegen_preflight.py
-
curl -L -o test_codegen_preflight.py https://huggingface.co/spaces/firebird-technologies/blog2video-cpu/resolve/main/backend/tests/test_codegen_preflight.py
2.82 kB
| """The preflight report: can it tell an armed pipeline from a defanged one? | |
| Every codegen quality gate fails OPEN, so a missing toolchain degrades output | |
| silently. The deployed image shipped without @babel/standalone and Level-2 | |
| validation stopped running for every generation, with nothing in the logs and | |
| no symptom except worse templates. These tests pin the report that makes that | |
| difference visible. | |
| """ | |
| from __future__ import annotations | |
| from app.services import codegen_preflight | |
| def test_report_returns_a_line_per_check() -> None: | |
| ok, lines = codegen_preflight.preflight_report() | |
| assert len(lines) == len(codegen_preflight.CHECKS) | |
| assert isinstance(ok, bool) | |
| def test_every_line_is_labelled_ok_or_down() -> None: | |
| _, lines = codegen_preflight.preflight_report() | |
| for line in lines: | |
| assert "[OK ]" in line or "[DOWN]" in line | |
| def test_a_probe_that_raises_reports_down_rather_than_exploding() -> None: | |
| """The report runs at startup; it must never be able to take boot down.""" | |
| def _boom() -> tuple[bool, str]: | |
| raise RuntimeError("toolchain on fire") | |
| original = codegen_preflight.CHECKS | |
| codegen_preflight.CHECKS = [*original, ("exploding probe", _boom)] | |
| try: | |
| ok, lines = codegen_preflight.preflight_report() | |
| assert ok is False | |
| assert any("toolchain on fire" in line and "[DOWN]" in line for line in lines) | |
| finally: | |
| codegen_preflight.CHECKS = original | |
| def test_the_report_agrees_with_the_toolchain_it_found() -> None: | |
| """The report must be CONSISTENT, not green. | |
| Asserting `ok is True` here was wrong: it encoded "this machine has npm | |
| dependencies installed", which is true of a dev checkout and false of the | |
| backend CI job (Python only — no esbuild, no babel) and of any environment | |
| that legitimately runs without them. Failing open there is the designed | |
| behaviour, so a green report is not what this test is for. | |
| What must hold everywhere is that the summary matches the lines: the report | |
| claims all-armed only when every individual probe is armed. That catches a | |
| real regression in the aggregation while staying true in every environment. | |
| """ | |
| ok, lines = codegen_preflight.preflight_report() | |
| assert ok == all("[OK ]" in line for line in lines) | |
| def test_a_down_toolchain_takes_the_summary_down_with_it() -> None: | |
| """One dead gate must not be averaged away into an all-clear.""" | |
| original = codegen_preflight.CHECKS | |
| codegen_preflight.CHECKS = [ | |
| ("armed", lambda: (True, "fine")), | |
| ("dead", lambda: (False, "NOT FOUND")), | |
| ] | |
| try: | |
| ok, lines = codegen_preflight.preflight_report() | |
| assert ok is False | |
| assert any("[DOWN]" in line for line in lines) | |
| finally: | |
| codegen_preflight.CHECKS = original | |