"""Unit tests for Phase-3 results assembly (harness.results), plan-phase3-results-ux.md §4.2 / R1/R3.
Covers JUnit parsing, stage roll-up, the tier→rung derivation (the documented mapping the level
depends on), and full results.json assembly incl. the U0 gate cases. Pure / tmp-file only. Run cold:
cc-ci-run -m pytest tests/unit/test_results.py -q
"""
from __future__ import annotations
import json
import os
import sys
sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "runner"))
from harness import results as R # noqa: E402
JUNIT_PASS = """
"""
JUNIT_MIXED = """
trace
"""
def _write(tmp_path, name, content):
p = tmp_path / name
p.write_text(content)
return str(p)
def test_parse_junit_pass(tmp_path):
rows = R.parse_junit(_write(tmp_path, "p.xml", JUNIT_PASS))
assert len(rows) == 2
assert {r["status"] for r in rows} == {"pass"}
assert rows[1]["ms"] == 1500
def test_parse_junit_mixed(tmp_path):
rows = R.parse_junit(_write(tmp_path, "m.xml", JUNIT_MIXED))
by = {r["name"]: r["status"] for r in rows}
assert by == {"test_ok": "pass", "test_bad": "fail", "test_skipped": "skip"}
def test_parse_junit_missing_file_is_empty():
assert R.parse_junit("/nonexistent/x.xml") == []
def test_collect_stages_orders_and_rolls_up(tmp_path):
recs = [
{
"tier": "install",
"source": "generic",
"file": "g/test_install.py",
"rc": 0,
"junit": _write(tmp_path, "i.xml", JUNIT_PASS),
},
{
"tier": "custom",
"source": "cc-ci",
"file": "c/test_x.py",
"rc": 1,
"junit": _write(tmp_path, "c.xml", JUNIT_MIXED),
},
]
stages = R.collect_stages(recs)
assert [s["name"] for s in stages] == ["install", "custom"] # install before custom
assert stages[0]["status"] == "pass"
assert stages[1]["status"] == "fail" # the failure in JUNIT_MIXED
assert len(stages[1]["tests"]) == 3
def test_collect_stages_synthesizes_when_no_junit():
recs = [
{
"tier": "install",
"source": "generic",
"file": "g/test_install.py",
"rc": 1,
"junit": None,
}
]
stages = R.collect_stages(recs)
assert stages[0]["status"] == "fail"
assert len(stages[0]["tests"]) == 1
# ---- derive_rungs: the documented mapping ----
def _results(**kw):
base = {
"install": "pass",
"upgrade": "pass",
"backup": "pass",
"restore": "pass",
"custom": "pass",
}
base.update(kw)
return base
def test_derive_rungs_full_climb_four_essential():
rungs = R.derive_rungs(_results(), backup_capable=True, has_custom=True)
# only the four essential rungs — integration/recipe-local are optional, not produced here.
assert rungs == {
"install": "pass",
"upgrade": "pass",
"backup_restore": "pass",
"functional": "pass",
}
def test_derive_rungs_stateless_backup_and_functional_na():
rungs = R.derive_rungs(
_results(backup="skip", restore="skip", custom="skip"),
backup_capable=False,
has_custom=False,
)
assert rungs["backup_restore"] == "na"
assert rungs["functional"] == "na"
assert "integration" not in rungs and "recipe_local" not in rungs
def test_derive_rungs_functional_fail():
rungs = R.derive_rungs(_results(custom="fail"), backup_capable=True, has_custom=True)
assert rungs["functional"] == "fail"
# ---- build_results: end-to-end incl level + flags ----
def test_build_results_level_and_flags(tmp_path):
recs = [
{
"tier": "install",
"source": "generic",
"file": "g/test_install.py",
"rc": 0,
"junit": _write(tmp_path, "i.xml", JUNIT_PASS),
},
{
"tier": "custom",
"source": "cc-ci",
"file": "c/test_func.py",
"rc": 0,
"junit": _write(tmp_path, "c.xml", JUNIT_PASS),
},
]
data = R.build_results(
recipe="hedgedoc",
version="1.2.3",
pr="7",
ref="deadbeefcafe0000",
records=recs,
results=_results(),
backup_capable=True,
clean_teardown=True,
no_secret_leak=True,
finished_ts=1234.0,
)
# all four essential rungs pass → full climb to L4 (the top), no cap
assert data["level"] == 4
assert data["level_cap_reason"] == ""
assert data["recipe"] == "hedgedoc"
assert data["ref"] == "deadbeefcafe"
assert data["flags"] == {"clean_teardown": True, "no_secret_leak": True}
assert [s["name"] for s in data["stages"]] == ["install", "custom"]
def test_build_results_capped_at_L1_on_upgrade_fail(tmp_path):
recs = [
{
"tier": "install",
"source": "generic",
"file": "g/test_install.py",
"rc": 0,
"junit": _write(tmp_path, "i.xml", JUNIT_PASS),
}
]
data = R.build_results(
recipe="x",
version=None,
pr="0",
ref=None,
records=recs,
results=_results(upgrade="fail"),
backup_capable=True,
clean_teardown=True,
no_secret_leak=True,
finished_ts=0.0,
)
assert data["level"] == 1
assert "L2" in data["level_cap_reason"]
# ---- skips: intentional (declared) vs unintentional (everything else skipped) ----
def _rungs(**kw):
base = {
"install": "pass",
"upgrade": "pass",
"backup_restore": "pass",
"functional": "pass",
}
base.update(kw)
return base
def test_skips_intentional_vs_unintentional():
rungs = _rungs(backup_restore="na", functional="na")
sk = R.skips(rungs, {"backup_restore": "stateless static server"})
# backup_restore is declared (intentional, with reason); functional skipped but not declared.
assert sk["intentional"] == {"backup_restore": "stateless static server"}
assert sk["unintentional"] == ["functional"]
def test_skips_none_declared_all_unintentional():
rungs = _rungs(backup_restore="na")
sk = R.skips(rungs, None)
assert sk["intentional"] == {}
assert sk["unintentional"] == ["backup_restore"]
def test_skips_declaration_only_counts_when_actually_skipped():
# backup_restore actually ran (pass) → not a skip, so a declaration for it is simply inert.
rungs = _rungs(backup_restore="pass")
sk = R.skips(rungs, {"backup_restore": "reason"})
assert "backup_restore" not in sk["intentional"]
assert "backup_restore" not in sk["unintentional"]
def test_build_results_threads_expected_na(tmp_path):
# Mirrors custom-html-tiny post-change: install + a passing functional (custom) test, but no
# backup surface (backup_restore declared intentionally skipped).
recs = [
{
"tier": "install",
"source": "generic",
"file": "g/test_install.py",
"rc": 0,
"junit": _write(tmp_path, "i.xml", JUNIT_PASS),
},
{
"tier": "custom",
"source": "cc-ci",
"file": "c/test_serves_content.py",
"rc": 0,
"junit": _write(tmp_path, "c.xml", JUNIT_PASS),
},
]
data = R.build_results(
recipe="custom-html-tiny",
version="1.1.0",
pr="0",
ref=None,
records=recs,
results=_results(backup="skip", restore="skip"), # custom=pass (default) → functional pass
backup_capable=False, # no backupbot label → backup_restore skipped (N/A)
clean_teardown=True,
no_secret_leak=True,
finished_ts=0.0,
expected_na={"backup_restore": "stateless static file server"},
)
# backup_restore skip still caps at L2 (never inflates) — even though functional passes above it,
# the skip caps the climb — but it's the declared (intentional) rung that capped.
assert data["level"] == 2
assert "L3" in data["level_cap_reason"]
assert data["level_cap_rung"] == "backup_restore"
assert data["rungs"]["functional"] == "pass"
assert data["skips"]["intentional"]["backup_restore"] == "stateless static file server"
assert (
data["skips"]["unintentional"] == []
) # backup_restore declared; functional passed → clean
def test_write_results_roundtrip(tmp_path):
data = {"run_id": "42", "level": 3, "stages": []}
path = R.write_results(data, runs_dir_override=str(tmp_path))
assert path.endswith("/42/results.json")
with open(path) as f:
assert json.load(f)["level"] == 3