[verified] Enforce delegation adoption at runtime consumers

This commit is contained in:
2026-10-11 07:11:40 +00:00
parent 9367d5a3a1
commit 6cb2317667
19 changed files with 1112 additions and 47 deletions
@@ -0,0 +1,150 @@
import importlib.util
import sys
from pathlib import Path
from tempfile import TemporaryDirectory
import unittest
ROOT = Path(__file__).resolve().parents[3]
PATH = ROOT / "harness/agents/impl/skills/gitea-issue-implementation/scripts/implementation_delivery.py"
INTEGRATION = ROOT / "harness/tests/common/test_delegation_consumers_integration.py"
def load(path, name):
spec = importlib.util.spec_from_file_location(name, path)
module = importlib.util.module_from_spec(spec)
sys.modules[name] = module
spec.loader.exec_module(module)
return module
class RecordingAdapter:
def __init__(self, head, *, remote_head=None, pr=None, mutate=None):
self.head = head
self.remote_head = remote_head if remote_head is not None else head
self.pr = pr if pr is not None else {"number": 17, "head": head, "state": "open"}
self.mutate = mutate
self.calls = []
def push(self, head):
self.calls.append(("push", head))
if self.mutate:
self.mutate("push")
def read_remote_head(self):
self.calls.append(("read_remote_head",))
if self.mutate:
self.mutate("read_remote_head")
return self.remote_head
def create_or_reuse_pr(self, head):
self.calls.append(("create_or_reuse_pr", head))
if self.mutate:
self.mutate("create_or_reuse_pr")
return self.pr["number"]
def read_pr(self, number):
self.calls.append(("read_pr", number))
return self.pr
class ImplementationDeliveryTests(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.delivery = load(PATH, "implementation_delivery_under_test")
cls.evidence = load(INTEGRATION, "implementation_delivery_evidence")
def setUp(self):
self.tmp = TemporaryDirectory(dir="/opt/data/cache/scratch")
self.ev = self.evidence.RealEvidence(Path(self.tmp.name))
self.ev.set_stage("implementation-delivery")
def tearDown(self):
self.tmp.cleanup()
def test_real_git_delivery_pushes_once_creates_once_and_returns_verified_readback(self):
adapter = RecordingAdapter(self.ev.revision)
result = self.delivery.deliver_implementation(
adapter, head=self.ev.revision, declaration=self.ev.declaration()
)
self.assertEqual(result, {
"remote_head": self.ev.revision,
"pr": {"number": 17, "head": self.ev.revision, "state": "open"},
})
self.assertEqual(
[call[0] for call in adapter.calls],
["push", "read_remote_head", "create_or_reuse_pr", "read_pr"],
)
def test_explicit_none_is_checked_and_exact_readback_is_required(self):
adapter = RecordingAdapter(self.ev.revision)
result = self.delivery.deliver_implementation(
adapter, head=self.ev.revision, declaration={"mode": "none"}
)
self.assertEqual(result["remote_head"], self.ev.revision)
self.assertEqual(result["pr"]["state"], "open")
def test_missing_or_malformed_declaration_has_no_effects(self):
for declaration in (None, {"mode": "bad"}):
adapter = RecordingAdapter(self.ev.revision)
with self.subTest(declaration=declaration), self.assertRaises(self.delivery.AdoptionError):
self.delivery.deliver_implementation(adapter, head=self.ev.revision, declaration=declaration)
self.assertEqual(adapter.calls, [])
def test_remote_drift_prevents_pr_creation(self):
adapter = RecordingAdapter(self.ev.revision, remote_head="0" * 40)
with self.assertRaises(self.delivery.AdoptionError):
self.delivery.deliver_implementation(adapter, head=self.ev.revision, declaration=self.ev.declaration())
self.assertEqual([call[0] for call in adapter.calls], ["push", "read_remote_head"])
def test_evidence_mutation_during_remote_readback_prevents_pr_creation(self):
def mutate(phase):
if phase == "read_remote_head":
self.ev.bundle.joinpath("src/a.py").write_text("VALUE = 2\n")
adapter = RecordingAdapter(self.ev.revision, mutate=mutate)
with self.assertRaises(self.delivery.AdoptionError):
self.delivery.deliver_implementation(adapter, head=self.ev.revision, declaration=self.ev.declaration())
self.assertEqual([call[0] for call in adapter.calls], ["push", "read_remote_head"])
def test_missing_adapter_capability_prevents_push(self):
class Incomplete:
def read_remote_head(self):
return self.ev.revision
with self.assertRaises(self.delivery.AdoptionError):
self.delivery.deliver_implementation(Incomplete(), head=self.ev.revision, declaration={"mode": "none"})
def test_pr_readback_mismatch_does_not_complete(self):
for pr in (
{"number": 17, "head": "0" * 40, "state": "open"},
{"number": 17, "head": self.ev.revision, "state": "closed"},
{"number": True, "head": self.ev.revision, "state": "open"},
):
adapter = RecordingAdapter(self.ev.revision, pr=pr)
with self.subTest(pr=pr), self.assertRaises(self.delivery.AdoptionError):
self.delivery.deliver_implementation(adapter, head=self.ev.revision, declaration=self.ev.declaration())
def test_workflow_can_load_from_external_cwd_without_pythonpath(self):
import os
import subprocess
code = "import importlib.util,json; s=importlib.util.spec_from_file_location('delivery'," + repr(str(PATH)) + "); m=importlib.util.module_from_spec(s); s.loader.exec_module(m); head='a'*40; A=type('A',(),{'push':lambda self,h:None,'read_remote_head':lambda self:head,'create_or_reuse_pr':lambda self,h:17,'read_pr':lambda self,n:{'number':n,'head':head,'state':'open'}}); print(json.dumps(m.deliver_implementation(A(),head=head,declaration={'mode':'none'})))"
env = os.environ.copy(); env.pop('PYTHONPATH', None)
proc = subprocess.run([sys.executable, '-c', code], cwd=self.tmp.name, env=env,
capture_output=True, text=True, timeout=10)
self.assertEqual(proc.returncode, 0, 'canonical delivery must load independently of cwd')
self.assertIn('remote_head', proc.stdout)
def test_uncertain_write_is_not_retried(self):
class FailingAdapter(RecordingAdapter):
def create_or_reuse_pr(self, head):
self.calls.append(("create_or_reuse_pr", head))
raise RuntimeError("transport uncertain")
adapter = FailingAdapter(self.ev.revision)
with self.assertRaises(RuntimeError):
self.delivery.deliver_implementation(adapter, head=self.ev.revision, declaration=self.ev.declaration())
self.assertEqual([call[0] for call in adapter.calls], ["push", "read_remote_head", "create_or_reuse_pr"])
if __name__ == "__main__":
unittest.main()
+3 -3
View File
@@ -11,7 +11,7 @@ class ReworkCommentTests(unittest.TestCase):
"result":"partial","findings":[{"id":"F-1","status":"resolved","classification":"valid_fix","changes":"src/main.py","verification":"pytest -q (0), pass","reason":"bounded fix"}],
"local_checks":"pytest -q (0), pass","runtime_browser":"해당 없음·미검증","commit_status":"success","actions_runs":"없음"}
self.write("input.json", self.data)
self.invoke("render", "--input", self.p("input.json"), "--comment", self.p("comment.md"), "--expected", self.p("expected.json"))
self.invoke("render", "--no-delegated-evidence", "--input", self.p("input.json"), "--comment", self.p("comment.md"), "--expected", self.p("expected.json"))
self.expected = json.loads(self.d.joinpath("expected.json").read_text())
def tearDown(self): self.t.cleanup()
@@ -47,7 +47,7 @@ class ReworkCommentTests(unittest.TestCase):
self.setUp_render_again(); self.readback(); c=self.read("comments.json"); c["comments"].append(dict(c["comments"][0],id=502)); c["total_count"]=2; self.write("comments.json",c); self.assertNotEqual(self.verify("reuse",1).returncode,0)
def setUp_render_again(self):
self.invoke("render","--input",self.p("input.json"),"--comment",self.p("comment.md"),"--expected",self.p("expected.json")); self.expected=self.read("expected.json")
self.invoke("render","--no-delegated-evidence","--input",self.p("input.json"),"--comment",self.p("comment.md"),"--expected",self.p("expected.json")); self.expected=self.read("expected.json")
def test_create_reuse_cardinality(self):
self.readback(); self.assertEqual(self.verify("create",0).returncode,0); self.assertEqual(self.read("receipt.json")["status"],"verified")
original=self.read("comments.json")["comments"][0]
@@ -59,7 +59,7 @@ class ReworkCommentTests(unittest.TestCase):
def test_unresolved_delivery_and_incomplete_or_wrong_scope_readback_fail(self):
bad=dict(self.data); bad["result"]="rework_delivered"; bad["findings"]=[dict(self.data["findings"][0],status="unresolved")]
self.write("bad-input.json",bad)
self.assertNotEqual(self.invoke("render","--input",self.p("bad-input.json"),"--comment",self.p("bad.md"),"--expected",self.p("bad.json")).returncode,0)
self.assertNotEqual(self.invoke("render","--no-delegated-evidence","--input",self.p("bad-input.json"),"--comment",self.p("bad.md"),"--expected",self.p("bad.json")).returncode,0)
self.readback(); comments=self.read("comments.json"); comments["complete"]=False; self.write("comments.json",comments); self.assertNotEqual(self.verify().returncode,0)
self.readback(); pr=self.read("pr.json"); pr["repository"]="other"; self.write("pr.json",pr); self.assertNotEqual(self.verify().returncode,0)