mward4/deepseek-harness
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 32 Wiki Activity

Merge master after separating plugin-manager fix

Browse Source
This commit is contained in:
fz committed 2026-10-12 13:04:25 +08:00
2 parents e51226401f a993a97c69
commit 88a0be8486
2277 files changed
+51396 -10652
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

scripts/smoke-python-runtime.py
+99 -14
View File
@@ -104,6 +104,8 @@ RESTART_SECOND_PROMPT = "Complete the second isolated Python SDK process turn."
RESTART_SECOND_TEXT = "PROCESS_TWO_OK"
RESTART_FIRST_SESSION_ID = "process-one"
RESTART_SECOND_SESSION_ID = "process-two"
RESUME_PROMPT = "Continue the first Python SDK session after restarting the runtime."
RESUME_TEXT = "RESUMED_PROCESS_OK"
RECOVERY_PROMPT = "Run three todo updates; the scheduler will reject the second call."
RECOVERY_CONTINUE_PROMPT = "Continue after the scheduler failure without repeating any tools."
RECOVERY_TEXT = "SCHEDULER_RECOVERY_OK"
@@ -429,6 +431,7 @@ def completion_chunks(body: dict[str, object]) -> list[dict[str, object]]:
MCP_PROMPT,
RESTART_FIRST_PROMPT,
RESTART_SECOND_PROMPT,
RESUME_PROMPT,
PROFILE_PLUGIN_PROMPT,
AUTHORING_PROMPT,
DYNAMIC_TOOLS_PROMPT,
@@ -457,6 +460,8 @@ def completion_chunks(body: dict[str, object]) -> list[dict[str, object]]:
return tool_call_chunks("advanced-native", "snapshot_double", {"value": 21})
if prompt == RESTART_FIRST_PROMPT:
return text_chunks(RESTART_FIRST_TEXT)
if prompt == RESUME_PROMPT:
return text_chunks(RESUME_TEXT)
if prompt == DYNAMIC_TOOLS_PROMPT:
assert_advertised_tool(body, "read")
if "snapshot_ping" in advertised_tool_names(body):
@@ -884,10 +889,18 @@ def main() -> None:
def smoke_sdk_authoring(base_url: str, executable: Path, update_snapshots: bool) -> None:
"""Query the bundled Python and switch skills without replacing that environment."""
with tempfile.TemporaryDirectory(prefix="dsh-sdk-authoring-cache-") as cache_directory:
_smoke_sdk_authoring_cached(base_url, executable, update_snapshots, Path(cache_directory))
def _smoke_sdk_authoring_cached(base_url: str, executable: Path, update_snapshots: bool, cache: Path) -> None:
from deepseek_harness import DeepSeekHarness
resources = executable.with_name(executable.name.removeprefix("deepseek-harness-sdk-runtime-").removesuffix(".exe"))
manifest = json.loads((resources / "primary-runtime/runtime.json").read_text())
cache_environment = {**os.environ, "DSH_RESOURCE_CACHE": str(cache), "DSH_RUNTIME_DOWNLOAD": "primary"}
primary = Path(json.loads(subprocess.check_output([str(executable)], encoding="utf-8", env=cache_environment)))
resources = primary.parent
manifest = json.loads((primary / "runtime.json").read_text())
for mode in ("default", "replacement", "disabled"):
with tempfile.TemporaryDirectory(prefix="dsh-sdk-authoring-") as temporary:
root = Path(temporary).resolve()
@@ -903,7 +916,7 @@ def smoke_sdk_authoring(base_url: str, executable: Path, update_snapshots: bool)
provider="deepseek-official", model="smoke-model", cwd=str(root),
dsh_bin=str(executable), dsh_home=str(home), patches=(str(patch),),
api_key="sk-keyless-smoke", base_url=base_url,
env={"DSH_PERMISSION_MODE": "danger-full-access", "DSH_TELEMETRY_DISABLED": "1"},
env={"DSH_PERMISSION_MODE": "danger-full-access", "DSH_TELEMETRY_DISABLED": "1", "DSH_RESOURCE_CACHE": str(cache)},
request_timeout_seconds=60,
) as harness:
result = harness.run(AUTHORING_PROMPT, session_id="authoring")
@@ -958,7 +971,9 @@ def smoke_sdk_office(executable: Path) -> None:
shutil.copytree(source, destination)
else:
shutil.copy2(source, destination)
office = root / f"{stem}-office"
cache_environment = {**os.environ, "DSH_RESOURCE_CACHE": str(root / "cache")}
office = Path(json.loads(subprocess.check_output([str(relocated)], encoding="utf-8", env={**cache_environment, "DSH_RUNTIME_DOWNLOAD": "office"})))
assert not (office.parent / "primary").exists(), "Office download unexpectedly prepared the authoring environment"
adapter = office / "node_modules/@deepseek-ai/libreoffice-kit/package.json"
native = stem.removeprefix("deepseek-harness-sdk-runtime-").replace("win-", "win32-").replace("macos-", "darwin-")
declared = json.loads(adapter.read_text(encoding="utf-8")).get("optionalDependencies", {})
@@ -997,7 +1012,7 @@ def smoke_sdk_office(executable: Path) -> None:
patches=(str(patch),),
api_key="sk-keyless-smoke",
base_url="http://127.0.0.1:9",
env={"DSH_PERMISSION_MODE": "danger-full-access", "DSH_TELEMETRY_DISABLED": "1"},
env={"DSH_PERMISSION_MODE": "danger-full-access", "DSH_TELEMETRY_DISABLED": "1", "DSH_RESOURCE_CACHE": str(root / "cache")},
# The startup plugin awaits a converter with a 120-second deadline before JSON-RPC is ready.
initialize_timeout_seconds=180,
request_timeout_seconds=180,
@@ -1007,10 +1022,32 @@ def smoke_sdk_office(executable: Path) -> None:
if result["backend"] != expected_backend:
raise AssertionError(f"Office conversion did not use {expected_backend}: {result}")
if not result["moduleUrl"].startswith(office.as_uri() + "/"):
raise AssertionError(f"Office module was not loaded from the relocated wheel: {result}")
raise AssertionError(f"Office module was not loaded from the downloaded sidecar: {result}")
pdf = output.read_bytes()
if len(pdf) < 100 or not pdf.startswith(b"%PDF-"):
raise AssertionError(f"Office conversion produced an invalid PDF at {output}")
workbook = root / "input.xlsx"
with zipfile.ZipFile(workbook, "w", compression=zipfile.ZIP_DEFLATED) as archive:
archive.writestr("[Content_Types].xml", '<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types"><Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/><Default Extension="xml" ContentType="application/xml"/><Override PartName="/xl/workbook.xml" ContentType="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet.main+xml"/><Override PartName="/xl/worksheets/sheet1.xml" ContentType="application/vnd.openxmlformats-officedocument.spreadsheetml.worksheet+xml"/></Types>')
archive.writestr("_rels/.rels", '<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="xl/workbook.xml"/></Relationships>')
archive.writestr("xl/workbook.xml", '<workbook xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"><sheets><sheet name="Sheet1" sheetId="1" r:id="rId1"/></sheets></workbook>')
archive.writestr("xl/_rels/workbook.xml.rels", '<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/worksheet" Target="worksheets/sheet1.xml"/></Relationships>')
archive.writestr("xl/worksheets/sheet1.xml", '<worksheet xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main"><sheetData><row r="1"><c r="A1"><f>1+2</f><v>0</v></c></row></sheetData></worksheet>')
subprocess.run([sys.executable, "-c", """
import sys, zipfile
from pathlib import Path
from xml.etree import ElementTree
from deepseek_harness.office import convert, render_images, recalculate
root = Path(sys.argv[1])
assert convert(root / 'document.docx', root / 'python.pdf').backend == sys.argv[2]
rendered = render_images(root / 'python.pdf', root / 'images', pages=[1])
assert rendered.images and rendered.images[0].path.read_bytes().startswith(b'\\x89PNG')
recalculate(root / 'input.xlsx', root / 'calculated.xlsx')
with zipfile.ZipFile(root / 'calculated.xlsx') as archive:
sheet = ElementTree.fromstring(archive.read('xl/worksheets/sheet1.xml'))
assert sheet.find('.//{*}c[@r="A1"]/{*}v').text == '3'
""", str(root), expected_backend], env={name: value for name, value in cache_environment.items() if name != "DSH_OFFICE_SIDECAR"}, check=True, timeout=180)
assert not (office.parent / "primary").exists(), "Python Office API unexpectedly downloaded authoring resources"
print(f"smoke-python-runtime: relocated Office {result['backend']} DOCX produced {len(pdf)} PDF bytes")
@@ -1336,6 +1373,8 @@ def smoke_sdk_dynamic_tools(base_url: str, executable: Path, update_snapshots: b
"DSH_PERMISSION_MODE": "danger-full-access", "DSH_TELEMETRY_DISABLED": "1",
"HOME": str(user_home), "USERPROFILE": str(user_home),
"DSH_AGENTS_HOME": str(user_home / ".agents"),
# Pin the tool baseline independently of optional resources in the host cache.
"DSH_PRIMARY_RUNTIME": "",
},
api_key="sk-keyless-smoke", base_url=base_url, request_timeout_seconds=60,
) as harness:
@@ -1675,7 +1714,7 @@ def smoke_sdk_snapshot_mode(
def smoke_sdk_restart_snapshot(base_url: str, executable: Path, update_snapshots: bool) -> None:
"""Snapshot two isolated sessions across complete SDK runtime restarts."""
"""Check isolated sessions and explicit history restoration across runtime restarts."""
from deepseek_harness import DeepSeekHarness
with tempfile.TemporaryDirectory(prefix="dsh-sdk-restart-") as temporary:
@@ -1685,8 +1724,8 @@ def smoke_sdk_restart_snapshot(base_url: str, executable: Path, update_snapshots
patch = write_advanced_profile_patch(root, "restart.patch.yml", sessions)
first_request = len(MockModelHandler.requests)
def run(prompt: str, session_id: str) -> "RunResult":
with DeepSeekHarness(
def connect() -> DeepSeekHarness:
return DeepSeekHarness(
provider="deepseek-official",
model="smoke-model",
cwd=str(root),
@@ -1700,11 +1739,12 @@ def smoke_sdk_restart_snapshot(base_url: str, executable: Path, update_snapshots
api_key="sk-keyless-smoke",
base_url=base_url,
request_timeout_seconds=60,
) as harness:
return harness.run(prompt, session_id=session_id)
)
first = run(RESTART_FIRST_PROMPT, RESTART_FIRST_SESSION_ID)
second = run(RESTART_SECOND_PROMPT, RESTART_SECOND_SESSION_ID)
with connect() as harness:
first = harness.run(RESTART_FIRST_PROMPT, session_id=RESTART_FIRST_SESSION_ID)
with connect() as harness:
second = harness.run(RESTART_SECOND_PROMPT, session_id=RESTART_SECOND_SESSION_ID)
requests = MockModelHandler.requests[first_request:]
if len(requests) != 2:
raise AssertionError(f"restart snapshot expected two model requests: {requests}")
@@ -1740,6 +1780,45 @@ def smoke_sdk_restart_snapshot(base_url: str, executable: Path, update_snapshots
native_writer_output=True,
)
before_resume = logs[RESTART_FIRST_SESSION_ID]
with connect() as harness:
session = harness.resume_session(RESTART_FIRST_SESSION_ID)
if session.id != RESTART_FIRST_SESSION_ID:
raise AssertionError(f"resume changed the saved identity: {session.id}")
if len(MockModelHandler.requests) != first_request + 2:
raise AssertionError("resume sent a model request before receiving new input")
resumed = session.run(RESUME_PROMPT)
requests = MockModelHandler.requests[first_request + 2:]
if len(requests) != 1 or resumed.final_response != RESUME_TEXT:
raise AssertionError(f"resumed turn failed: {resumed.final_response!r}, {requests}")
messages = requests[0].get("messages")
if not isinstance(messages, list):
raise AssertionError(f"resumed model request has no messages: {requests[0]}")
for role, expected in (
("user", RESTART_FIRST_PROMPT),
("assistant", RESTART_FIRST_TEXT),
("user", RESUME_PROMPT),
):
if not any(
isinstance(message, dict)
and message.get("role") == role
and expected in message_text(message.get("content"))
for message in messages
):
raise AssertionError(f"resumed model history has no {role} message: {expected}")
logs = read_session_logs(sessions)
if set(logs) != expected_ids:
raise AssertionError(f"resume created an unexpected durable session: {sorted(logs)}")
restored = logs[RESTART_FIRST_SESSION_ID]
if restored[:len(before_resume)] != before_resume:
raise AssertionError("resume replaced the saved session log instead of appending")
turns = [record for record in restored if record.get("type") == "turn/end"]
if [record["data"]["turn"] for record in turns] != [1, 2]:
raise AssertionError(f"resume did not continue the saved turn sequence: {turns}")
if RESUME_TEXT not in render_jsonl(restored):
raise AssertionError("resumed response was not appended to the saved session")
def smoke_sdk_scheduler_recovery(base_url: str, executable: Path, update_snapshots: bool) -> None:
"""Keep a failed tool turn usable through the real Messages serializer and SDK."""
@@ -2322,11 +2401,17 @@ def build_dynamic_tools_snapshot_files(
assert changes == [[], [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "snapshot_ping"}}], []], changes
assert requests[1]["messages"][:len(requests[0]["messages"])] == requests[0]["messages"], "tool addition changed earlier messages"
replacements = [(str(cwd), "{{cwd}}"), (result.session_id, "{{session}}")]
header_references = {header["seq"]: f"{{{{header:{index + 1}}}}}" for index, header in enumerate(headers)}
snapshot_events = [
{**event, "data": {**event["data"], "headerSeq": header_references[event["data"]["headerSeq"]]}}
if event["type"] == "developer/message" and "headerSeq" in event["data"] else event
for event in events if event["type"] in {"request/header", "request/context", "developer/message"}
]
evidence = {
"finalResponse": result.final_response,
"declarations": declarations,
"requestToolChanges": changes,
"events": [event for event in events if event["type"] in {"request/header", "request/context", "developer/message"}],
"events": snapshot_events,
"toolCalls": calls,
"pingResult": ping,
}