feat(extensibility): implement Custom Sandbox Strategy Registration via SandboxStrategy Protocol
CI / benchmark-publish (pull_request) Has been skipped
CI / build (pull_request) Successful in 19s
CI / lint (pull_request) Successful in 23s
CI / quality (pull_request) Successful in 28s
CI / security (pull_request) Successful in 54s
CI / typecheck (pull_request) Successful in 1m25s
CI / e2e_tests (pull_request) Successful in 2m33s
CI / unit_tests (pull_request) Successful in 3m4s
CI / integration_tests (pull_request) Successful in 3m33s
CI / docker (pull_request) Successful in 1m7s
CI / coverage (pull_request) Successful in 5m50s
CI / lint (push) Successful in 18s
CI / build (push) Successful in 33s
CI / quality (push) Successful in 47s
CI / typecheck (push) Successful in 49s
CI / benchmark-regression (push) Has been skipped
CI / security (push) Successful in 52s
CI / e2e_tests (push) Successful in 2m9s
CI / unit_tests (push) Successful in 3m32s
CI / integration_tests (push) Successful in 3m41s
CI / docker (push) Successful in 56s
CI / coverage (push) Successful in 6m10s
CI / benchmark-publish (push) Has been cancelled
CI / benchmark-regression (pull_request) Successful in 40m21s

Implement SandboxStrategyProtocol, a 9-method @runtime_checkable Protocol enabling
third-party sandbox strategy registration. Includes:

- SandboxStrategyProtocol with create/read/write/diff/commit/rollback/checkpoint/
  restore_checkpoint/cleanup methods
- SandboxRef (frozen dataclass) and DiffView/DiffEntry (Pydantic models)
- SandboxStrategyRegistry with config-driven registration, Protocol validation,
  thread safety, and clear/list/has operations
- BuiltInSandboxStrategyAdapter wrapping existing Sandbox implementations to conform
  to the new Protocol
- CustomStrategyConfig for YAML/dict-based strategy registration
- SandboxFactory integration with custom_registry parameter,
  has_custom_strategy() and get_custom_strategy_class()
- 25 Behave BDD scenarios (85 steps) covering protocol, registry, adapter, config,
  and factory integration
- 8 Robot Framework integration tests with real filesystem operations
- ASV benchmarks for registry and adapter operations
- Developer documentation

ISSUES CLOSED: #586
This commit was merged in pull request #709.
This commit is contained in:
2026-03-12 01:30:46 +00:00
committed by Forgejo
parent 05503712ae
commit 2688c85769
12 changed files with 2555 additions and 1 deletions
+242
View File
@@ -0,0 +1,242 @@
"""ASV benchmarks for custom sandbox strategy registration.
Measures the performance of:
- SandboxRef creation
- DiffView / DiffEntry model creation
- SandboxStrategyRegistry registration and lookup
- BuiltInSandboxStrategyAdapter create/write/read cycle
- Protocol runtime checking via isinstance
"""
from __future__ import annotations
import importlib
import os
import sys
import tempfile
from datetime import datetime
from pathlib import Path
# Ensure the local *source* tree is importable even when ASV has an
# older build of the package installed.
_SRC = str(Path(__file__).resolve().parents[1] / "src")
if _SRC not in sys.path:
sys.path.insert(0, _SRC)
# Force-reload so ASV picks up the source tree version.
import cleveragents # noqa: E402
importlib.reload(cleveragents)
from cleveragents.domain.models.core.resource import ( # noqa: E402
PhysVirt,
Resource,
ResourceCapabilities,
SandboxStrategy as SandboxStrategyEnum,
)
from cleveragents.domain.models.core.sandbox_strategy import ( # noqa: E402
DiffEntry,
DiffView,
SandboxRef,
SandboxStrategyProtocol,
)
from cleveragents.infrastructure.sandbox.strategy_adapter import ( # noqa: E402
BuiltInSandboxStrategyAdapter,
)
from cleveragents.infrastructure.sandbox.strategy_registry import ( # noqa: E402
SandboxStrategyRegistry,
)
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
_COUNTER = 0
def _ulid(name: str) -> str:
global _COUNTER
_COUNTER += 1
base = abs(hash(name)) % (10**20)
raw = f"{_COUNTER:06d}{base:020d}"
charset = "0123456789ABCDEFGHJKMNPQRSTVWXYZ"
result = ""
for ch in raw:
result += charset[int(ch)]
return (result + "0" * 26)[:26]
class _MockStrategy:
"""Minimal strategy satisfying the Protocol for benchmarking."""
def create(self, plan_id, resource): # noqa: ANN001, ANN201
return None
def read(self, ref, path): # noqa: ANN001, ANN201
return b""
def write(self, ref, path, content): # noqa: ANN001, ANN201
return None
def diff(self, ref): # noqa: ANN001, ANN201
return None
def commit(self, ref): # noqa: ANN001, ANN201
pass
def rollback(self, ref): # noqa: ANN001, ANN201
pass
def checkpoint(self, ref, checkpoint_id): # noqa: ANN001, ANN201
pass
def restore_checkpoint(self, ref, checkpoint_id): # noqa: ANN001, ANN201
pass
def cleanup(self, ref): # noqa: ANN001, ANN201
pass
# ---------------------------------------------------------------------------
# SandboxRef benchmarks
# ---------------------------------------------------------------------------
class TimeSandboxRef:
"""Benchmark SandboxRef creation."""
def time_create_ref(self) -> None:
SandboxRef(
sandbox_id="bench-ref",
plan_id="bench-plan",
resource_id="bench-res",
created_at=datetime.now(),
)
def time_create_ref_with_metadata(self) -> None:
SandboxRef(
sandbox_id="bench-ref-meta",
plan_id="bench-plan",
resource_id="bench-res",
created_at=datetime.now(),
metadata={"key": "value", "num": 42},
)
# ---------------------------------------------------------------------------
# DiffEntry / DiffView benchmarks
# ---------------------------------------------------------------------------
class TimeDiffModels:
"""Benchmark DiffEntry and DiffView creation."""
def time_create_diff_entry(self) -> None:
DiffEntry(path="bench.txt", operation="modified")
def time_create_diff_entry_with_content(self) -> None:
DiffEntry(
path="bench.txt",
operation="modified",
before=b"old",
after=b"new",
)
def time_create_diff_view_10_entries(self) -> None:
entries = [
DiffEntry(path=f"file{i}.txt", operation="modified") for i in range(10)
]
DiffView(sandbox_id="bench-dv", entries=entries)
# ---------------------------------------------------------------------------
# Protocol checking benchmarks
# ---------------------------------------------------------------------------
class TimeProtocolCheck:
"""Benchmark SandboxStrategyProtocol isinstance checks."""
def setup(self) -> None:
self.good = _MockStrategy()
def time_isinstance_check_pass(self) -> None:
isinstance(self.good, SandboxStrategyProtocol)
def time_isinstance_check_fail(self) -> None:
isinstance("not a strategy", SandboxStrategyProtocol)
# ---------------------------------------------------------------------------
# SandboxStrategyRegistry benchmarks
# ---------------------------------------------------------------------------
class TimeStrategyRegistry:
"""Benchmark registry operations."""
def setup(self) -> None:
self.registry = SandboxStrategyRegistry(
allowed_prefixes=("cleveragents.",),
)
self.registry.register(
"bench_adapter",
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
def time_lookup_hit(self) -> None:
self.registry.get("bench_adapter")
def time_lookup_miss(self) -> None:
self.registry.get("nonexistent")
def time_has_check(self) -> None:
self.registry.has("bench_adapter")
def time_list_strategies(self) -> None:
self.registry.list_strategies()
# ---------------------------------------------------------------------------
# BuiltInSandboxStrategyAdapter benchmarks
# ---------------------------------------------------------------------------
class TimeAdapterLifecycle:
"""Benchmark adapter create/write/read cycle."""
def setup(self) -> None:
self.tmpdir = tempfile.mkdtemp(prefix="ca-bench-adapter-")
with open(os.path.join(self.tmpdir, "seed.txt"), "w") as f:
f.write("seed")
self.adapter = BuiltInSandboxStrategyAdapter(strategy_name="copy_on_write")
self.resource = Resource(
resource_id=_ulid("bench-adapter"),
name="bench-adapter",
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.COPY_ON_WRITE,
location=self.tmpdir,
capabilities=ResourceCapabilities(
readable=True,
writable=True,
sandboxable=True,
checkpointable=False,
),
)
def time_create_and_cleanup(self) -> None:
ref = self.adapter.create("bench-plan", self.resource)
self.adapter.cleanup(ref)
def time_write_read_cycle(self) -> None:
ref = self.adapter.create("bench-plan-wr", self.resource)
self.adapter.write(ref, "bench.txt", b"bench content")
self.adapter.read(ref, "bench.txt")
self.adapter.cleanup(ref)
def teardown(self) -> None:
import shutil
shutil.rmtree(self.tmpdir, ignore_errors=True)
+150
View File
@@ -0,0 +1,150 @@
# Custom Sandbox Strategy Registration
## Overview
CleverAgents supports custom sandbox strategies via the `SandboxStrategyProtocol`.
Custom strategies enable domain-specific isolation for specialized resource types
that are not covered by the built-in strategies (none, git\_worktree, copy\_on\_write,
transaction\_rollback).
## SandboxStrategy Protocol
All custom sandbox strategies must implement the 9-method `SandboxStrategyProtocol`:
```python
from cleveragents.domain.models.core.sandbox_strategy import (
SandboxStrategyProtocol,
SandboxRef,
DiffEntry,
DiffView,
)
from cleveragents.domain.models.core.resource import Resource
class MyCustomSandbox:
def create(self, plan_id: str, resource: Resource) -> SandboxRef: ...
def read(self, ref: SandboxRef, path: str) -> bytes: ...
def write(self, ref: SandboxRef, path: str, content: bytes) -> DiffEntry: ...
def diff(self, ref: SandboxRef) -> DiffView: ...
def commit(self, ref: SandboxRef) -> None: ...
def rollback(self, ref: SandboxRef) -> None: ...
def checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None: ...
def restore_checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None: ...
def cleanup(self, ref: SandboxRef) -> None: ...
```
## Supporting Models
### SandboxRef
Opaque reference returned by `create()` and passed to all other methods:
```python
@dataclass(frozen=True)
class SandboxRef:
sandbox_id: str
plan_id: str
resource_id: str
created_at: datetime
metadata: dict[str, object] = field(default_factory=dict)
```
### DiffEntry / DiffView
`DiffEntry` describes a single file change; `DiffView` aggregates entries:
```python
class DiffEntry(BaseModel):
path: str # relative file path
operation: str # "added", "modified", or "deleted"
before: bytes | None
after: bytes | None
class DiffView(BaseModel):
sandbox_id: str
entries: list[DiffEntry]
summary: str
```
## Config-Driven Registration
Custom strategies are registered via the `SandboxStrategyRegistry`:
```python
from cleveragents.infrastructure.sandbox.strategy_registry import (
SandboxStrategyRegistry,
)
registry = SandboxStrategyRegistry(
allowed_prefixes=("cleveragents.", "mycompany."),
)
# Single registration
registry.register(
"my_strategy",
"mycompany.sandbox.custom",
"MyCustomSandbox",
)
# Batch config-driven registration
configs = {
"redis_sandbox": {
"module": "mycompany.sandbox.redis",
"class": "RedisSandbox",
},
"s3_sandbox": {
"module": "mycompany.sandbox.s3",
"class": "S3Sandbox",
},
}
registry.register_all_from_config(configs)
```
## Factory Integration
The `SandboxFactory` can be initialized with a custom registry:
```python
from cleveragents.infrastructure.sandbox.factory import SandboxFactory
factory = SandboxFactory(custom_registry=registry)
# Check if custom strategy is available
if factory.has_custom_strategy("redis_sandbox"):
cls = factory.get_custom_strategy_class("redis_sandbox")
```
## Built-In Strategy Adapter
The `BuiltInSandboxStrategyAdapter` wraps existing `Sandbox` implementations
to conform to the `SandboxStrategyProtocol`:
```python
from cleveragents.infrastructure.sandbox.strategy_adapter import (
BuiltInSandboxStrategyAdapter,
)
adapter = BuiltInSandboxStrategyAdapter(strategy_name="copy_on_write")
ref = adapter.create("plan-001", resource)
adapter.write(ref, "file.txt", b"content")
data = adapter.read(ref, "file.txt")
adapter.cleanup(ref)
```
## Protocol Validation
All strategies are validated against the Protocol at registration time.
The registry checks that all 9 required methods are present as callable
attributes:
```python
# This will raise ProtocolMismatchError if the class
# doesn't implement all required methods:
registry.register("bad", "mymodule", "IncompleteClass")
```
## Security
Module imports are restricted to a configurable prefix allowlist
(default: `("cleveragents.",)`). Only modules whose fully-qualified
name starts with an allowed prefix may be dynamically imported.
@@ -0,0 +1,193 @@
@extensibility @custom_sandbox_strategy
Feature: Custom Sandbox Strategy Registration via SandboxStrategy Protocol
As a CleverAgents developer
I want to register custom sandbox strategies via configuration
So that specialized resource types can use domain-specific isolation
# ---------------------------------------------------------------------------
# SandboxRef model
# ---------------------------------------------------------------------------
@sandbox_ref
Scenario: SandboxRef is a frozen dataclass
Given a SandboxRef with sandbox_id "sb-001" and plan_id "plan-001"
Then the SandboxRef sandbox_id should be "sb-001"
And the SandboxRef plan_id should be "plan-001"
And the SandboxRef should be immutable
@sandbox_ref
Scenario: SandboxRef carries metadata
Given a SandboxRef with metadata key "backend" value "redis"
Then the SandboxRef metadata should contain key "backend"
# ---------------------------------------------------------------------------
# DiffView / DiffEntry models
# ---------------------------------------------------------------------------
@diff_view
Scenario: DiffEntry creation with required fields
Given a DiffEntry with path "src/main.py" and operation "modified"
Then the DiffEntry path should be "src/main.py"
And the DiffEntry operation should be "modified"
@diff_view
Scenario: DiffView aggregates entries
Given a DiffView with 3 entries
Then the DiffView should have 3 entries
And the DiffView should have a sandbox_id
@diff_view
Scenario: DiffEntry rejects empty path
When I attempt to create a DiffEntry with empty path
Then a sandbox strategy validation error should be raised
# ---------------------------------------------------------------------------
# SandboxStrategyProtocol
# ---------------------------------------------------------------------------
@sandbox_strategy_protocol
Scenario: SandboxStrategyProtocol is runtime_checkable
Given the SandboxStrategyProtocol is available
Then it should be a runtime_checkable Protocol
@sandbox_strategy_protocol
Scenario: A class implementing all 9 methods satisfies the Protocol
Given a mock class implementing all 9 SandboxStrategy methods
Then the mock class should satisfy SandboxStrategyProtocol
@sandbox_strategy_protocol
Scenario: A class missing methods does not satisfy the Protocol
Given a class missing the checkpoint method
Then the class should not satisfy SandboxStrategyProtocol
# ---------------------------------------------------------------------------
# SandboxStrategyRegistry
# ---------------------------------------------------------------------------
@strategy_registry
Scenario: Register a valid custom strategy
Given a SandboxStrategyRegistry with test allowed prefixes
And a valid custom strategy class in "cleveragents.infrastructure.sandbox.strategy_adapter"
When I register the strategy as "test_adapter"
Then the registry should contain strategy "test_adapter"
And listing strategies should include "test_adapter"
@strategy_registry
Scenario: Registering a non-protocol class raises ProtocolMismatchError
Given a SandboxStrategyRegistry with test allowed prefixes
When I attempt to register a non-protocol class as "bad_strategy"
Then a sandbox ProtocolMismatchError should be raised
@strategy_registry
Scenario: Register with empty name raises ValueError
Given a SandboxStrategyRegistry with test allowed prefixes
When I attempt to register a strategy with empty name
Then a sandbox ValueError should be raised
@strategy_registry
Scenario: Get unregistered strategy returns None
Given an empty SandboxStrategyRegistry
When I look up strategy "nonexistent"
Then the strategy lookup result should be None
@strategy_registry
Scenario: Register all from config dictionary
Given a SandboxStrategyRegistry with test allowed prefixes
And a config dictionary with 2 custom strategies
When I register all from config
Then 2 strategies should be registered
@strategy_registry
Scenario: Clear removes all strategies
Given a SandboxStrategyRegistry with one registered strategy
When I clear the sandbox strategy registry
Then the sandbox strategy registry should be empty
# ---------------------------------------------------------------------------
# BuiltInSandboxStrategyAdapter
# ---------------------------------------------------------------------------
@adapter
Scenario: Adapter create returns SandboxRef
Given a BuiltInSandboxStrategyAdapter for "none" strategy
And a test Resource with location
When I call adapter.create with plan "plan-001"
Then I should get a SandboxRef with plan_id "plan-001"
@adapter
Scenario: Adapter write creates a DiffEntry
Given a BuiltInSandboxStrategyAdapter for "copy_on_write" strategy
And a test Resource with a temporary directory
And the adapter sandbox is created for plan "plan-write"
When I call adapter.write with path "test.txt" and content "hello"
Then I should get a DiffEntry with operation "added"
@adapter
Scenario: Adapter read returns file content
Given a BuiltInSandboxStrategyAdapter for "copy_on_write" strategy
And a test Resource with a temporary directory
And the adapter sandbox is created for plan "plan-read"
And I write "data" to "read_test.txt" via the adapter
When I call adapter.read for "read_test.txt"
Then I should get content "data"
@adapter
Scenario: Adapter diff returns DiffView
Given a BuiltInSandboxStrategyAdapter for "none" strategy
And a test Resource with location
And the adapter sandbox is created for plan "plan-diff"
When I call adapter.diff
Then I should get a DiffView
@adapter
Scenario: Adapter checkpoint and restore
Given a BuiltInSandboxStrategyAdapter for "copy_on_write" strategy
And a test Resource with a temporary directory
And the adapter sandbox is created for plan "plan-cp"
And I write "v1" to "cp_test.txt" via the adapter
And I checkpoint as "cp-1"
And I write "v2" to "cp_test.txt" via the adapter
When I restore checkpoint "cp-1"
Then reading "cp_test.txt" should return "v1"
@adapter
Scenario: Adapter cleanup releases resources
Given a BuiltInSandboxStrategyAdapter for "none" strategy
And a test Resource with location
And the adapter sandbox is created for plan "plan-cleanup"
When I call adapter.cleanup
Then the adapter should have no tracked sandboxes
# ---------------------------------------------------------------------------
# CustomStrategyConfig
# ---------------------------------------------------------------------------
@config
Scenario: CustomStrategyConfig rejects empty name
When I attempt to create a CustomStrategyConfig with empty name
Then a sandbox ValueError should be raised
@config
Scenario: CustomStrategyConfig rejects empty module
When I attempt to create a CustomStrategyConfig with empty module
Then a sandbox ValueError should be raised
@config
Scenario: CustomStrategyConfig rejects empty class_name
When I attempt to create a CustomStrategyConfig with empty class_name
Then a sandbox ValueError should be raised
# ---------------------------------------------------------------------------
# Factory custom strategy integration
# ---------------------------------------------------------------------------
@factory_integration
Scenario: Factory reports custom strategy availability
Given a SandboxFactory with a custom registry containing "my_custom"
Then the factory should report "my_custom" as a custom strategy
And the factory should not report "nonexistent" as a custom strategy
@factory_integration
Scenario: Factory without registry reports no custom strategies
Given a SandboxFactory without a custom registry
Then the factory should not report "anything" as a custom strategy
@@ -0,0 +1,602 @@
"""Behave step definitions for Custom Sandbox Strategy Registration.
Covers SandboxRef, DiffView, DiffEntry, SandboxStrategyProtocol,
SandboxStrategyRegistry, BuiltInSandboxStrategyAdapter,
CustomStrategyConfig, and factory integration.
Based on issue #586.
"""
from __future__ import annotations
import os
import tempfile
from datetime import datetime
from typing import Any
from behave import given, then, when # type: ignore[import-untyped]
from behave.runner import Context # type: ignore[import-untyped]
from pydantic import ValidationError
from cleveragents.domain.models.core.resource import (
PhysVirt,
Resource,
ResourceCapabilities,
)
from cleveragents.domain.models.core.resource import (
SandboxStrategy as SandboxStrategyEnum,
)
from cleveragents.domain.models.core.sandbox_strategy import (
DiffEntry,
DiffView,
SandboxRef,
SandboxStrategyProtocol,
)
from cleveragents.infrastructure.plugins.exceptions import (
ProtocolMismatchError,
)
from cleveragents.infrastructure.sandbox.factory import SandboxFactory
from cleveragents.infrastructure.sandbox.strategy_adapter import (
BuiltInSandboxStrategyAdapter,
)
from cleveragents.infrastructure.sandbox.strategy_registry import (
CustomStrategyConfig,
SandboxStrategyRegistry,
)
# ---------------------------------------------------------------------------
# ULID helper
# ---------------------------------------------------------------------------
_COUNTER = 0
def _ulid(name: str) -> str:
global _COUNTER
_COUNTER += 1
base = abs(hash(name)) % (10**20)
raw = f"{_COUNTER:06d}{base:020d}"
charset = "0123456789ABCDEFGHJKMNPQRSTVWXYZ"
result = ""
for ch in raw:
result += charset[int(ch)]
return (result + "0" * 26)[:26]
def _make_resource(name: str, location: str | None = None) -> Resource:
return Resource(
resource_id=_ulid(name),
name=name,
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.NONE,
location=location,
capabilities=ResourceCapabilities(
readable=True, writable=True, sandboxable=True, checkpointable=False
),
)
# ---------------------------------------------------------------------------
# SandboxRef
# ---------------------------------------------------------------------------
@given('a SandboxRef with sandbox_id "{sid}" and plan_id "{pid}"')
def step_create_sandbox_ref(context: Context, sid: str, pid: str) -> None:
context.sandbox_ref = SandboxRef(
sandbox_id=sid,
plan_id=pid,
resource_id="res-001",
created_at=datetime.now(),
)
@then('the SandboxRef sandbox_id should be "{expected}"')
def step_sandbox_ref_sid(context: Context, expected: str) -> None:
assert context.sandbox_ref.sandbox_id == expected
@then('the SandboxRef plan_id should be "{expected}"')
def step_sandbox_ref_pid(context: Context, expected: str) -> None:
assert context.sandbox_ref.plan_id == expected
@then("the SandboxRef should be immutable")
def step_sandbox_ref_frozen(context: Context) -> None:
import dataclasses
assert dataclasses.is_dataclass(context.sandbox_ref)
try:
context.sandbox_ref.sandbox_id = "mutated" # type: ignore[misc]
raise AssertionError("Expected FrozenInstanceError")
except (AttributeError, dataclasses.FrozenInstanceError):
pass
@given('a SandboxRef with metadata key "{key}" value "{value}"')
def step_sandbox_ref_metadata(context: Context, key: str, value: str) -> None:
context.sandbox_ref = SandboxRef(
sandbox_id="sb-meta",
plan_id="plan-meta",
resource_id="res-meta",
created_at=datetime.now(),
metadata={key: value},
)
@then('the SandboxRef metadata should contain key "{key}"')
def step_sandbox_ref_meta_key(context: Context, key: str) -> None:
assert key in context.sandbox_ref.metadata
# ---------------------------------------------------------------------------
# DiffEntry / DiffView
# ---------------------------------------------------------------------------
@given('a DiffEntry with path "{path}" and operation "{op}"')
def step_create_diff_entry(context: Context, path: str, op: str) -> None:
context.diff_entry = DiffEntry(path=path, operation=op)
@then('the DiffEntry path should be "{expected}"')
def step_diff_entry_path(context: Context, expected: str) -> None:
assert context.diff_entry.path == expected
@then('the DiffEntry operation should be "{expected}"')
def step_diff_entry_op(context: Context, expected: str) -> None:
assert context.diff_entry.operation == expected
@given("a DiffView with {count:d} entries")
def step_create_diff_view(context: Context, count: int) -> None:
entries = [
DiffEntry(path=f"file{i}.txt", operation="modified") for i in range(count)
]
context.diff_view = DiffView(
sandbox_id="dv-001",
entries=entries,
)
@then("the DiffView should have {count:d} entries")
def step_diff_view_count(context: Context, count: int) -> None:
assert len(context.diff_view.entries) == count
@then("the DiffView should have a sandbox_id")
def step_diff_view_sid(context: Context) -> None:
assert context.diff_view.sandbox_id
@when("I attempt to create a DiffEntry with empty path")
def step_create_diff_entry_empty(context: Context) -> None:
context.sandbox_error = None
try:
DiffEntry(path="", operation="modified")
except ValidationError as exc:
context.sandbox_error = exc
@then("a sandbox strategy validation error should be raised")
def step_sandbox_validation_error(context: Context) -> None:
assert context.sandbox_error is not None
# ---------------------------------------------------------------------------
# SandboxStrategyProtocol
# ---------------------------------------------------------------------------
@given("the SandboxStrategyProtocol is available")
def step_protocol_available(context: Context) -> None:
context.protocol = SandboxStrategyProtocol
@then("it should be a runtime_checkable Protocol")
def step_protocol_runtime_checkable(context: Context) -> None:
assert hasattr(context.protocol, "__protocol_attrs__") or hasattr(
context.protocol, "_is_runtime_protocol"
)
class _FullMockStrategy:
"""Mock satisfying all 9 SandboxStrategy methods."""
def create(self, plan_id: str, resource: Any) -> Any:
return None
def read(self, ref: Any, path: str) -> bytes:
return b""
def write(self, ref: Any, path: str, content: bytes) -> Any:
return None
def diff(self, ref: Any) -> Any:
return None
def commit(self, ref: Any) -> None:
pass
def rollback(self, ref: Any) -> None:
pass
def checkpoint(self, ref: Any, checkpoint_id: str) -> None:
pass
def restore_checkpoint(self, ref: Any, checkpoint_id: str) -> None:
pass
def cleanup(self, ref: Any) -> None:
pass
class _PartialMockStrategy:
"""Missing checkpoint method."""
def create(self, plan_id: str, resource: Any) -> Any:
return None
def read(self, ref: Any, path: str) -> bytes:
return b""
def write(self, ref: Any, path: str, content: bytes) -> Any:
return None
def diff(self, ref: Any) -> Any:
return None
def commit(self, ref: Any) -> None:
pass
def rollback(self, ref: Any) -> None:
pass
# Missing: checkpoint, restore_checkpoint
def cleanup(self, ref: Any) -> None:
pass
@given("a mock class implementing all 9 SandboxStrategy methods")
def step_full_mock(context: Context) -> None:
context.mock_cls = _FullMockStrategy
@then("the mock class should satisfy SandboxStrategyProtocol")
def step_mock_satisfies(context: Context) -> None:
instance = context.mock_cls()
assert isinstance(instance, SandboxStrategyProtocol)
@given("a class missing the checkpoint method")
def step_partial_mock(context: Context) -> None:
context.partial_cls = _PartialMockStrategy
@then("the class should not satisfy SandboxStrategyProtocol")
def step_partial_fails(context: Context) -> None:
instance = context.partial_cls()
assert not isinstance(instance, SandboxStrategyProtocol)
# ---------------------------------------------------------------------------
# SandboxStrategyRegistry
# ---------------------------------------------------------------------------
@given("a SandboxStrategyRegistry with test allowed prefixes")
def step_registry_with_prefixes(context: Context) -> None:
context.strategy_registry = SandboxStrategyRegistry(
allowed_prefixes=("cleveragents.",)
)
@given("an empty SandboxStrategyRegistry")
def step_empty_registry(context: Context) -> None:
context.strategy_registry = SandboxStrategyRegistry(
allowed_prefixes=("cleveragents.",)
)
@given(
"a valid custom strategy class in "
'"cleveragents.infrastructure.sandbox.strategy_adapter"'
)
def step_valid_strategy_class(context: Context) -> None:
context.strategy_module = "cleveragents.infrastructure.sandbox.strategy_adapter"
context.strategy_class = "BuiltInSandboxStrategyAdapter"
@when('I register the strategy as "{name}"')
def step_register_strategy(context: Context, name: str) -> None:
context.strategy_registry.register(
name, context.strategy_module, context.strategy_class
)
@then('the registry should contain strategy "{name}"')
def step_registry_contains(context: Context, name: str) -> None:
assert context.strategy_registry.has(name)
@then('listing strategies should include "{name}"')
def step_registry_list(context: Context, name: str) -> None:
assert name in context.strategy_registry.list_strategies()
@when('I attempt to register a non-protocol class as "{name}"')
def step_register_nonprotocol(context: Context, name: str) -> None:
context.sandbox_error = None
try:
context.strategy_registry.register(
name,
"cleveragents.infrastructure.sandbox.protocol",
"SandboxError",
)
except ProtocolMismatchError as exc:
context.sandbox_error = exc
@then("a sandbox ProtocolMismatchError should be raised")
def step_sandbox_protocol_mismatch(context: Context) -> None:
assert isinstance(context.sandbox_error, ProtocolMismatchError)
@when("I attempt to register a strategy with empty name")
def step_register_empty_name(context: Context) -> None:
context.sandbox_error = None
try:
context.strategy_registry.register(
"",
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
except ValueError as exc:
context.sandbox_error = exc
@then("a sandbox ValueError should be raised")
def step_sandbox_value_error(context: Context) -> None:
assert isinstance(context.sandbox_error, (ValueError, type(None))) is False or (
context.sandbox_error is not None
)
assert context.sandbox_error is not None
@when('I look up strategy "{name}"')
def step_lookup_strategy(context: Context, name: str) -> None:
context.strategy_lookup_result = context.strategy_registry.get(name)
@then("the strategy lookup result should be None")
def step_strategy_result_none(context: Context) -> None:
assert context.strategy_lookup_result is None
@given("a config dictionary with 2 custom strategies")
def step_config_dict(context: Context) -> None:
context.config_dict = {
"adapter1": {
"module": "cleveragents.infrastructure.sandbox.strategy_adapter",
"class": "BuiltInSandboxStrategyAdapter",
},
"adapter2": {
"module": "cleveragents.infrastructure.sandbox.strategy_adapter",
"class": "BuiltInSandboxStrategyAdapter",
},
}
@when("I register all from config")
def step_register_all_config(context: Context) -> None:
context.registered = context.strategy_registry.register_all_from_config(
context.config_dict
)
@then("{count:d} strategies should be registered")
def step_count_registered(context: Context, count: int) -> None:
assert len(context.registered) == count
@given("a SandboxStrategyRegistry with one registered strategy")
def step_registry_one(context: Context) -> None:
context.strategy_registry = SandboxStrategyRegistry(
allowed_prefixes=("cleveragents.",)
)
context.strategy_registry.register(
"single",
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
@when("I clear the sandbox strategy registry")
def step_clear_sandbox_registry(context: Context) -> None:
context.strategy_registry.clear()
@then("the sandbox strategy registry should be empty")
def step_sandbox_registry_empty(context: Context) -> None:
assert len(context.strategy_registry.list_strategies()) == 0
# ---------------------------------------------------------------------------
# BuiltInSandboxStrategyAdapter
# ---------------------------------------------------------------------------
@given('a BuiltInSandboxStrategyAdapter for "{strategy}" strategy')
def step_adapter(context: Context, strategy: str) -> None:
context.adapter = BuiltInSandboxStrategyAdapter(strategy_name=strategy) # type: ignore[arg-type]
@given("a test Resource with location")
def step_resource_with_location(context: Context) -> None:
context.tmpdir = tempfile.mkdtemp(prefix="ca-test-")
context.test_resource = _make_resource("test-res", location=context.tmpdir)
@given("a test Resource with a temporary directory")
def step_resource_with_tmpdir(context: Context) -> None:
context.tmpdir = tempfile.mkdtemp(prefix="ca-test-cow-")
# Create a seed file so CoW sandbox has something to copy
with open(os.path.join(context.tmpdir, "seed.txt"), "w") as f:
f.write("seed")
context.test_resource = Resource(
resource_id=_ulid("cow-res"),
name="cow-res",
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.COPY_ON_WRITE,
location=context.tmpdir,
capabilities=ResourceCapabilities(
readable=True, writable=True, sandboxable=True, checkpointable=False
),
)
@when('I call adapter.create with plan "{plan_id}"')
def step_adapter_create(context: Context, plan_id: str) -> None:
context.adapter_ref = context.adapter.create(plan_id, context.test_resource)
@given('the adapter sandbox is created for plan "{plan_id}"')
def step_adapter_create_given(context: Context, plan_id: str) -> None:
context.adapter_ref = context.adapter.create(plan_id, context.test_resource)
@then('I should get a SandboxRef with plan_id "{expected}"')
def step_adapter_ref_check(context: Context, expected: str) -> None:
assert isinstance(context.adapter_ref, SandboxRef)
assert context.adapter_ref.plan_id == expected
@when('I call adapter.write with path "{path}" and content "{content}"')
def step_adapter_write(context: Context, path: str, content: str) -> None:
context.write_result = context.adapter.write(
context.adapter_ref, path, content.encode()
)
@given('I write "{content}" to "{path}" via the adapter')
def step_adapter_write_given(context: Context, content: str, path: str) -> None:
context.adapter.write(context.adapter_ref, path, content.encode())
@then('I should get a DiffEntry with operation "{expected}"')
def step_adapter_write_op(context: Context, expected: str) -> None:
assert isinstance(context.write_result, DiffEntry)
assert context.write_result.operation == expected
@when('I call adapter.read for "{path}"')
def step_adapter_read(context: Context, path: str) -> None:
context.read_result = context.adapter.read(context.adapter_ref, path)
@then('I should get content "{expected}"')
def step_adapter_read_content(context: Context, expected: str) -> None:
assert context.read_result == expected.encode()
@when("I call adapter.diff")
def step_adapter_diff(context: Context) -> None:
context.diff_result = context.adapter.diff(context.adapter_ref)
@then("I should get a DiffView")
def step_adapter_diff_check(context: Context) -> None:
assert isinstance(context.diff_result, DiffView)
@given('I checkpoint as "{cp_id}"')
def step_adapter_checkpoint(context: Context, cp_id: str) -> None:
context.adapter.checkpoint(context.adapter_ref, cp_id)
@when('I restore checkpoint "{cp_id}"')
def step_adapter_restore(context: Context, cp_id: str) -> None:
context.adapter.restore_checkpoint(context.adapter_ref, cp_id)
@then('reading "{path}" should return "{expected}"')
def step_adapter_read_after_restore(context: Context, path: str, expected: str) -> None:
data = context.adapter.read(context.adapter_ref, path)
assert data == expected.encode(), f"Expected {expected!r}, got {data!r}"
@when("I call adapter.cleanup")
def step_adapter_cleanup(context: Context) -> None:
context.adapter.cleanup(context.adapter_ref)
@then("the adapter should have no tracked sandboxes")
def step_adapter_no_sandboxes(context: Context) -> None:
assert len(context.adapter._sandboxes) == 0
# ---------------------------------------------------------------------------
# CustomStrategyConfig
# ---------------------------------------------------------------------------
@when("I attempt to create a CustomStrategyConfig with empty name")
def step_config_empty_name(context: Context) -> None:
context.sandbox_error = None
try:
CustomStrategyConfig(name="", module="mod", class_name="Cls")
except ValueError as exc:
context.sandbox_error = exc
@when("I attempt to create a CustomStrategyConfig with empty module")
def step_config_empty_module(context: Context) -> None:
context.sandbox_error = None
try:
CustomStrategyConfig(name="x", module="", class_name="Cls")
except ValueError as exc:
context.sandbox_error = exc
@when("I attempt to create a CustomStrategyConfig with empty class_name")
def step_config_empty_class(context: Context) -> None:
context.sandbox_error = None
try:
CustomStrategyConfig(name="x", module="mod", class_name="")
except ValueError as exc:
context.sandbox_error = exc
# ---------------------------------------------------------------------------
# Factory custom strategy integration
# ---------------------------------------------------------------------------
@given('a SandboxFactory with a custom registry containing "{name}"')
def step_factory_with_registry(context: Context, name: str) -> None:
registry = SandboxStrategyRegistry(allowed_prefixes=("cleveragents.",))
registry.register(
name,
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
context.factory = SandboxFactory(custom_registry=registry)
@given("a SandboxFactory without a custom registry")
def step_factory_no_registry(context: Context) -> None:
context.factory = SandboxFactory()
@then('the factory should report "{name}" as a custom strategy')
def step_factory_has_custom(context: Context, name: str) -> None:
assert context.factory.has_custom_strategy(name)
@then('the factory should not report "{name}" as a custom strategy')
def step_factory_no_custom(context: Context, name: str) -> None:
assert not context.factory.has_custom_strategy(name)
+69
View File
@@ -0,0 +1,69 @@
*** Settings ***
Documentation Integration tests for custom sandbox strategy registration.
... Covers SandboxStrategyProtocol, SandboxStrategyRegistry,
... BuiltInSandboxStrategyAdapter, CustomStrategyConfig, and
... SandboxFactory custom strategy integration.
... Based on issue #586.
Resource ${CURDIR}/common.resource
Suite Setup Setup Test Environment
Suite Teardown Cleanup Test Environment
*** Variables ***
${HELPER_SCRIPT} robot/helper_custom_sandbox_strategy.py
*** Test Cases ***
SandboxRef Creation And Immutability
[Documentation] Verify SandboxRef creation, properties, and frozen semantics
[Tags] sandbox strategy ref
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} sandbox-ref-lifecycle cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} sandbox-ref-lifecycle-ok
DiffView And DiffEntry Models
[Documentation] Verify DiffView and DiffEntry creation and validation
[Tags] sandbox strategy diff
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} diff-view-lifecycle cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} diff-view-lifecycle-ok
SandboxStrategyProtocol Runtime Check
[Documentation] Verify SandboxStrategyProtocol is runtime_checkable with 9 methods
[Tags] sandbox strategy protocol
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} protocol-check cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} protocol-check-ok
SandboxStrategyRegistry Lifecycle
[Documentation] Verify registry register, lookup, config-driven, protocol validation, clear
[Tags] sandbox strategy registry
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} registry-lifecycle cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} registry-lifecycle-ok
BuiltInSandboxStrategyAdapter Full Lifecycle
[Documentation] Verify adapter create/read/write/diff/cleanup with real filesystem
[Tags] sandbox strategy adapter
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} adapter-lifecycle cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} adapter-lifecycle-ok
Adapter Checkpoint And Restore
[Documentation] Verify adapter checkpoint save and restore operations
[Tags] sandbox strategy adapter checkpoint
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} adapter-checkpoint cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} adapter-checkpoint-ok
CustomStrategyConfig Validation
[Documentation] Verify CustomStrategyConfig rejects empty name, module, class_name
[Tags] sandbox strategy config
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} config-validation cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} config-validation-ok
SandboxFactory Custom Strategy Integration
[Documentation] Verify SandboxFactory with custom registry lookup
[Tags] sandbox strategy factory
${result}= Run Process ${PYTHON} ${HELPER_SCRIPT} factory-custom-integration cwd=${WORKSPACE}
Should Be Equal As Integers ${result.rc} 0
Should Contain ${result.stdout} factory-custom-integration-ok
+439
View File
@@ -0,0 +1,439 @@
"""Helper utilities for custom sandbox strategy Robot integration tests.
Covers end-to-end integration of:
- SandboxStrategyProtocol and runtime checking
- SandboxStrategyRegistry registration and lookup
- BuiltInSandboxStrategyAdapter full lifecycle
- CustomStrategyConfig validation
- SandboxFactory custom strategy integration
Based on issue #586.
"""
from __future__ import annotations
import os
import sys
import tempfile
from datetime import datetime
from cleveragents.domain.models.core.resource import (
PhysVirt,
Resource,
ResourceCapabilities,
)
from cleveragents.domain.models.core.resource import (
SandboxStrategy as SandboxStrategyEnum,
)
from cleveragents.domain.models.core.sandbox_strategy import (
DiffEntry,
DiffView,
SandboxRef,
SandboxStrategyProtocol,
)
from cleveragents.infrastructure.plugins.exceptions import (
ProtocolMismatchError,
)
from cleveragents.infrastructure.sandbox.factory import SandboxFactory
from cleveragents.infrastructure.sandbox.strategy_adapter import (
BuiltInSandboxStrategyAdapter,
)
from cleveragents.infrastructure.sandbox.strategy_registry import (
CustomStrategyConfig,
SandboxStrategyRegistry,
)
# ---------------------------------------------------------------------------
# ULID helper
# ---------------------------------------------------------------------------
_CTR = 0
def _ulid(name: str) -> str:
global _CTR
_CTR += 1
base = abs(hash(name)) % (10**20)
raw = f"{_CTR:06d}{base:020d}"
charset = "0123456789ABCDEFGHJKMNPQRSTVWXYZ"
result = ""
for ch in raw:
result += charset[int(ch)]
return (result + "0" * 26)[:26]
def _make_resource(name: str, location: str) -> Resource:
return Resource(
resource_id=_ulid(name),
name=name,
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.NONE,
location=location,
capabilities=ResourceCapabilities(
readable=True,
writable=True,
sandboxable=True,
checkpointable=False,
),
)
# ---------------------------------------------------------------------------
# Test: SandboxRef creation and immutability
# ---------------------------------------------------------------------------
def _sandbox_ref_lifecycle() -> None:
"""Integration test: SandboxRef creation and properties."""
ref = SandboxRef(
sandbox_id="ref-001",
plan_id="plan-001",
resource_id="res-001",
created_at=datetime.now(),
metadata={"backend": "test"},
)
assert ref.sandbox_id == "ref-001"
assert ref.plan_id == "plan-001"
assert ref.metadata["backend"] == "test"
# Frozen check
try:
ref.sandbox_id = "mutated" # type: ignore[misc]
print("FAIL: Expected FrozenInstanceError")
return
except (AttributeError, Exception):
pass
print("sandbox-ref-lifecycle-ok")
# ---------------------------------------------------------------------------
# Test: DiffView and DiffEntry
# ---------------------------------------------------------------------------
def _diff_view_lifecycle() -> None:
"""Integration test: DiffView and DiffEntry models."""
entry = DiffEntry(path="test.py", operation="added", after=b"content")
assert entry.path == "test.py"
assert entry.operation == "added"
view = DiffView(
sandbox_id="dv-001",
entries=[entry],
summary="1 file added",
)
assert len(view.entries) == 1
assert view.sandbox_id == "dv-001"
# Empty path should fail
try:
DiffEntry(path="", operation="modified")
print("FAIL: Expected ValidationError")
return
except Exception:
pass
print("diff-view-lifecycle-ok")
# ---------------------------------------------------------------------------
# Test: SandboxStrategyProtocol runtime check
# ---------------------------------------------------------------------------
def _protocol_check() -> None:
"""Integration test: Protocol runtime checking."""
class _Good:
def create(self, plan_id, resource):
return None
def read(self, ref, path):
return b""
def write(self, ref, path, content):
return None
def diff(self, ref):
return None
def commit(self, ref):
pass
def rollback(self, ref):
pass
def checkpoint(self, ref, checkpoint_id):
pass
def restore_checkpoint(self, ref, checkpoint_id):
pass
def cleanup(self, ref):
pass
class _Bad:
def create(self, plan_id, resource):
return None
good = _Good()
bad = _Bad()
assert isinstance(good, SandboxStrategyProtocol)
assert not isinstance(bad, SandboxStrategyProtocol)
print("protocol-check-ok")
# ---------------------------------------------------------------------------
# Test: SandboxStrategyRegistry
# ---------------------------------------------------------------------------
def _registry_lifecycle() -> None:
"""Integration test: Registry registration, lookup, config, clear."""
registry = SandboxStrategyRegistry(allowed_prefixes=("cleveragents.",))
# Register
cls = registry.register(
"adapter",
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
assert cls is BuiltInSandboxStrategyAdapter
assert registry.has("adapter")
assert "adapter" in registry.list_strategies()
# Get
assert registry.get("adapter") is BuiltInSandboxStrategyAdapter
assert registry.get("nonexistent") is None
# Non-protocol rejection
try:
registry.register(
"bad",
"cleveragents.infrastructure.sandbox.protocol",
"SandboxError",
)
print("FAIL: Expected ProtocolMismatchError")
return
except ProtocolMismatchError:
pass
# Config-driven registration
configs = {
"cfg1": {
"module": "cleveragents.infrastructure.sandbox.strategy_adapter",
"class": "BuiltInSandboxStrategyAdapter",
},
}
registered = registry.register_all_from_config(configs)
assert len(registered) == 1
# Clear
registry.clear()
assert len(registry.list_strategies()) == 0
print("registry-lifecycle-ok")
# ---------------------------------------------------------------------------
# Test: BuiltInSandboxStrategyAdapter full lifecycle
# ---------------------------------------------------------------------------
def _adapter_lifecycle() -> None:
"""Integration test: Adapter create/read/write/diff/commit/cleanup."""
with tempfile.TemporaryDirectory() as tmpdir:
# Seed a file
with open(os.path.join(tmpdir, "seed.txt"), "w") as f:
f.write("seed")
adapter = BuiltInSandboxStrategyAdapter(strategy_name="copy_on_write")
resource = Resource(
resource_id=_ulid("adapter-res"),
name="adapter-res",
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.COPY_ON_WRITE,
location=tmpdir,
capabilities=ResourceCapabilities(
readable=True,
writable=True,
sandboxable=True,
checkpointable=False,
),
)
# Create
ref = adapter.create("plan-adapter", resource)
assert isinstance(ref, SandboxRef)
assert ref.plan_id == "plan-adapter"
# Write
entry = adapter.write(ref, "new_file.txt", b"hello")
assert isinstance(entry, DiffEntry)
assert entry.operation == "added"
# Read
data = adapter.read(ref, "new_file.txt")
assert data == b"hello"
# Diff
view = adapter.diff(ref)
assert isinstance(view, DiffView)
# Cleanup
adapter.cleanup(ref)
assert ref.sandbox_id not in adapter._sandboxes
print("adapter-lifecycle-ok")
# ---------------------------------------------------------------------------
# Test: Adapter checkpoint/restore
# ---------------------------------------------------------------------------
def _adapter_checkpoint() -> None:
"""Integration test: Adapter checkpoint and restore."""
with tempfile.TemporaryDirectory() as tmpdir:
with open(os.path.join(tmpdir, "seed.txt"), "w") as f:
f.write("seed")
adapter = BuiltInSandboxStrategyAdapter(strategy_name="copy_on_write")
resource = Resource(
resource_id=_ulid("cp-res"),
name="cp-res",
resource_type_name="fs-directory",
classification=PhysVirt.PHYSICAL,
sandbox_strategy=SandboxStrategyEnum.COPY_ON_WRITE,
location=tmpdir,
capabilities=ResourceCapabilities(
readable=True,
writable=True,
sandboxable=True,
checkpointable=False,
),
)
ref = adapter.create("plan-cp", resource)
adapter.write(ref, "cp.txt", b"v1")
adapter.checkpoint(ref, "cp-1")
adapter.write(ref, "cp.txt", b"v2")
# Verify v2
assert adapter.read(ref, "cp.txt") == b"v2"
# Restore
adapter.restore_checkpoint(ref, "cp-1")
assert adapter.read(ref, "cp.txt") == b"v1"
adapter.cleanup(ref)
print("adapter-checkpoint-ok")
# ---------------------------------------------------------------------------
# Test: CustomStrategyConfig validation
# ---------------------------------------------------------------------------
def _config_validation() -> None:
"""Integration test: CustomStrategyConfig validation."""
# Valid
cfg = CustomStrategyConfig(
name="test", module="cleveragents.test", class_name="TestCls"
)
assert cfg.name == "test"
# Empty name
try:
CustomStrategyConfig(name="", module="mod", class_name="Cls")
print("FAIL: Expected ValueError for empty name")
return
except ValueError:
pass
# Empty module
try:
CustomStrategyConfig(name="x", module="", class_name="Cls")
print("FAIL: Expected ValueError for empty module")
return
except ValueError:
pass
# Empty class
try:
CustomStrategyConfig(name="x", module="mod", class_name="")
print("FAIL: Expected ValueError for empty class")
return
except ValueError:
pass
print("config-validation-ok")
# ---------------------------------------------------------------------------
# Test: SandboxFactory custom strategy integration
# ---------------------------------------------------------------------------
def _factory_custom_integration() -> None:
"""Integration test: SandboxFactory with custom strategy registry."""
registry = SandboxStrategyRegistry(allowed_prefixes=("cleveragents.",))
registry.register(
"my_custom",
"cleveragents.infrastructure.sandbox.strategy_adapter",
"BuiltInSandboxStrategyAdapter",
)
factory = SandboxFactory(custom_registry=registry)
assert factory.has_custom_strategy("my_custom")
assert not factory.has_custom_strategy("nonexistent")
cls = factory.get_custom_strategy_class("my_custom")
assert cls is BuiltInSandboxStrategyAdapter
# Factory without registry
plain_factory = SandboxFactory()
assert not plain_factory.has_custom_strategy("anything")
assert plain_factory.get_custom_strategy_class("anything") is None
print("factory-custom-integration-ok")
# ---------------------------------------------------------------------------
# Main dispatcher
# ---------------------------------------------------------------------------
_TESTS = {
"sandbox-ref-lifecycle": _sandbox_ref_lifecycle,
"diff-view-lifecycle": _diff_view_lifecycle,
"protocol-check": _protocol_check,
"registry-lifecycle": _registry_lifecycle,
"adapter-lifecycle": _adapter_lifecycle,
"adapter-checkpoint": _adapter_checkpoint,
"config-validation": _config_validation,
"factory-custom-integration": _factory_custom_integration,
}
def main() -> None:
if len(sys.argv) < 2:
print(f"Usage: {sys.argv[0]} <test-name>")
print(f"Available: {', '.join(sorted(_TESTS))}")
sys.exit(1)
test_name = sys.argv[1]
if test_name not in _TESTS:
print(f"Unknown test: {test_name}")
print(f"Available: {', '.join(sorted(_TESTS))}")
sys.exit(1)
_TESTS[test_name]()
if __name__ == "__main__":
main()
@@ -248,6 +248,14 @@ from cleveragents.domain.models.core.safety_profile import (
SafetyProfileRef, SafetyProfileRef,
resolve_safety_profile, resolve_safety_profile,
) )
from cleveragents.domain.models.core.sandbox_strategy import (
DiffEntry as SandboxDiffEntry,
)
from cleveragents.domain.models.core.sandbox_strategy import (
DiffView,
SandboxRef,
SandboxStrategyProtocol,
)
from cleveragents.domain.models.core.session import ( from cleveragents.domain.models.core.session import (
MessageRole, MessageRole,
Session, Session,
@@ -354,6 +362,7 @@ __all__ = [
"DiffBuilder", "DiffBuilder",
"DiffEntry", "DiffEntry",
"DiffSerializer", "DiffSerializer",
"DiffView",
"DoDCriterion", "DoDCriterion",
"DoDEvaluator", "DoDEvaluator",
"DoDResult", "DoDResult",
@@ -453,7 +462,10 @@ __all__ = [
"SafetyProfile", "SafetyProfile",
"SafetyProfileProvenance", "SafetyProfileProvenance",
"SafetyProfileRef", "SafetyProfileRef",
"SandboxDiffEntry",
"SandboxRef",
"SandboxStrategy", "SandboxStrategy",
"SandboxStrategyProtocol",
"ScoredFragment", "ScoredFragment",
"ServiceRetryPolicy", "ServiceRetryPolicy",
"ServiceRetryPolicyRegistry", "ServiceRetryPolicyRegistry",
@@ -0,0 +1,235 @@
"""Custom sandbox strategy Protocol and supporting models.
Defines the ``SandboxStrategyProtocol`` that custom sandbox strategies must
implement, along with ``SandboxRef`` and ``DiffView`` value objects used in
the Protocol method signatures.
The Protocol specifies 9 methods covering the full sandbox lifecycle:
create, read, write, diff, commit, rollback, checkpoint,
restore_checkpoint, cleanup.
Custom strategies are registered via configuration::
sandbox.custom_strategies.<name>.module = "my_module.path"
sandbox.custom_strategies.<name>.class = "MySandboxStrategy"
See Also:
- :class:`~cleveragents.infrastructure.sandbox.protocol.Sandbox`
The existing infrastructure-level sandbox protocol.
- :class:`~cleveragents.domain.models.core.resource.SandboxStrategy`
The StrEnum defining built-in strategy names.
- Specification § Custom Sandbox Strategies (lines 46171-46186)
Based on issue #586.
"""
from __future__ import annotations
from dataclasses import dataclass, field
from datetime import datetime
from typing import Protocol, runtime_checkable
from pydantic import BaseModel, ConfigDict, Field
from cleveragents.domain.models.core.resource import Resource
# ---------------------------------------------------------------------------
# SandboxRef — opaque reference to a sandbox instance
# ---------------------------------------------------------------------------
@dataclass(frozen=True)
class SandboxRef:
"""Opaque reference to a sandbox instance.
Returned by :meth:`SandboxStrategyProtocol.create` and passed to
all subsequent operations. The ``sandbox_id`` is a unique
identifier (typically a ULID); ``metadata`` allows strategy
implementations to carry extra state across calls.
Attributes:
sandbox_id: Unique identifier for this sandbox instance.
plan_id: The plan that owns this sandbox.
resource_id: The resource being sandboxed.
created_at: When the sandbox was created.
metadata: Strategy-specific opaque data.
"""
sandbox_id: str
plan_id: str
resource_id: str
created_at: datetime
metadata: dict[str, object] = field(default_factory=dict)
# ---------------------------------------------------------------------------
# DiffView — structured view of sandbox changes
# ---------------------------------------------------------------------------
class DiffEntry(BaseModel):
"""A single file-level diff entry within a :class:`DiffView`.
Attributes:
path: Relative path of the changed file.
operation: Type of change (``added``, ``modified``, ``deleted``).
before: Content before the change (``None`` for additions).
after: Content after the change (``None`` for deletions).
"""
path: str = Field(..., min_length=1, description="Relative file path")
operation: str = Field(
...,
description="Change type: 'added', 'modified', or 'deleted'",
)
before: bytes | None = Field(
default=None,
description="Content before change (None for additions)",
)
after: bytes | None = Field(
default=None,
description="Content after change (None for deletions)",
)
model_config = ConfigDict(frozen=True)
class DiffView(BaseModel):
"""Structured view of all changes in a sandbox.
Produced by :meth:`SandboxStrategyProtocol.diff`.
Attributes:
sandbox_id: The sandbox these diffs belong to.
entries: Individual file-level diff entries.
summary: Human-readable summary of the changes.
"""
sandbox_id: str = Field(..., description="Sandbox identifier")
entries: list[DiffEntry] = Field(
default_factory=list,
description="Per-file diff entries",
)
summary: str = Field(
default="",
description="Human-readable change summary",
)
model_config = ConfigDict(frozen=True)
# ---------------------------------------------------------------------------
# SandboxStrategyProtocol — the 9-method custom strategy interface
# ---------------------------------------------------------------------------
@runtime_checkable
class SandboxStrategyProtocol(Protocol):
"""Protocol for custom sandbox isolation strategies.
All custom sandbox strategies must implement these 9 methods.
Strategies are registered via configuration and resolved at
runtime by the :class:`SandboxStrategyRegistry`.
Lifecycle::
ref = strategy.create(plan_id, resource) # create sandbox
data = strategy.read(ref, "file.txt") # read from sandbox
change = strategy.write(ref, "f.txt", b"x") # write to sandbox
view = strategy.diff(ref) # inspect changes
strategy.checkpoint(ref, "cp-1") # save checkpoint
strategy.restore_checkpoint(ref, "cp-1") # restore checkpoint
strategy.commit(ref) # apply changes
strategy.rollback(ref) # discard changes
strategy.cleanup(ref) # release resources
"""
def create(self, plan_id: str, resource: Resource) -> SandboxRef:
"""Create a new sandbox for the given plan and resource.
Args:
plan_id: Identifier of the plan requesting the sandbox.
resource: The resource to sandbox.
Returns:
An opaque :class:`SandboxRef` for subsequent operations.
"""
...
def read(self, ref: SandboxRef, path: str) -> bytes:
"""Read file content from the sandbox.
Args:
ref: Sandbox reference from :meth:`create`.
path: Relative path within the sandbox.
Returns:
Raw file content as bytes.
"""
...
def write(self, ref: SandboxRef, path: str, content: bytes) -> DiffEntry:
"""Write content to a file in the sandbox.
Args:
ref: Sandbox reference from :meth:`create`.
path: Relative path within the sandbox.
content: Raw content to write.
Returns:
A :class:`DiffEntry` describing the change.
"""
...
def diff(self, ref: SandboxRef) -> DiffView:
"""Compute a structured diff of all sandbox changes.
Args:
ref: Sandbox reference from :meth:`create`.
Returns:
A :class:`DiffView` with per-file entries.
"""
...
def commit(self, ref: SandboxRef) -> None:
"""Apply sandbox changes to the original resource.
Args:
ref: Sandbox reference from :meth:`create`.
"""
...
def rollback(self, ref: SandboxRef) -> None:
"""Discard all sandbox changes.
Args:
ref: Sandbox reference from :meth:`create`.
"""
...
def checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None:
"""Save a named checkpoint of the current sandbox state.
Args:
ref: Sandbox reference from :meth:`create`.
checkpoint_id: Unique identifier for the checkpoint.
"""
...
def restore_checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None:
"""Restore the sandbox to a previously saved checkpoint.
Args:
ref: Sandbox reference from :meth:`create`.
checkpoint_id: Identifier of the checkpoint to restore.
"""
...
def cleanup(self, ref: SandboxRef) -> None:
"""Release all sandbox resources and artefacts.
Args:
ref: Sandbox reference from :meth:`create`.
"""
...
@@ -6,6 +6,7 @@ and multiple strategy implementations (git worktree, filesystem copy, no-op).
Stage B3 of the implementation plan. Updated in TASK-006 (B4 rework). Stage B3 of the implementation plan. Updated in TASK-006 (B4 rework).
Updated in M4 to add checkpoint/rollback hooks. Updated in M4 to add checkpoint/rollback hooks.
Updated in M6 to add sandbox boundary algebra (#548). Updated in M6 to add sandbox boundary algebra (#548).
Updated in M6+ to add custom sandbox strategy registration (#586).
""" """
from cleveragents.infrastructure.sandbox.boundary import ( from cleveragents.infrastructure.sandbox.boundary import (
@@ -40,14 +41,23 @@ from cleveragents.infrastructure.sandbox.protocol import (
SandboxError, SandboxError,
SandboxStatus, SandboxStatus,
) )
from cleveragents.infrastructure.sandbox.strategy_adapter import (
BuiltInSandboxStrategyAdapter,
)
from cleveragents.infrastructure.sandbox.strategy_registry import (
CustomStrategyConfig,
SandboxStrategyRegistry,
)
from cleveragents.infrastructure.sandbox.transaction_sandbox import TransactionSandbox from cleveragents.infrastructure.sandbox.transaction_sandbox import TransactionSandbox
__all__ = [ __all__ = [
"BoundaryCache", "BoundaryCache",
"BuiltInSandboxStrategyAdapter",
"CheckpointManager", "CheckpointManager",
"Checkpointable", "Checkpointable",
"CommitResult", "CommitResult",
"CopyOnWriteSandbox", "CopyOnWriteSandbox",
"CustomStrategyConfig",
"GitMergeStrategy", "GitMergeStrategy",
"GitWorktreeSandbox", "GitWorktreeSandbox",
"JsonMergeStrategy", "JsonMergeStrategy",
@@ -63,6 +73,7 @@ __all__ = [
"SandboxFactory", "SandboxFactory",
"SandboxManager", "SandboxManager",
"SandboxStatus", "SandboxStatus",
"SandboxStrategyRegistry",
"SequentialMergeStrategy", "SequentialMergeStrategy",
"TransactionSandbox", "TransactionSandbox",
"compute_sandbox_domains", "compute_sandbox_domains",
@@ -6,12 +6,13 @@ available, this factory will accept ``Resource`` objects directly; until
then it operates on raw parameters. then it operates on raw parameters.
Stage B3.6 of the implementation plan. Updated in TASK-006 (B4 rework). Stage B3.6 of the implementation plan. Updated in TASK-006 (B4 rework).
Updated in M6+ to support custom sandbox strategy resolution (#586).
""" """
from __future__ import annotations from __future__ import annotations
import logging import logging
from typing import Literal from typing import TYPE_CHECKING, Literal
from cleveragents.infrastructure.sandbox.copy_on_write import CopyOnWriteSandbox from cleveragents.infrastructure.sandbox.copy_on_write import CopyOnWriteSandbox
from cleveragents.infrastructure.sandbox.git_worktree import GitWorktreeSandbox from cleveragents.infrastructure.sandbox.git_worktree import GitWorktreeSandbox
@@ -21,6 +22,11 @@ from cleveragents.infrastructure.sandbox.protocol import (
) )
from cleveragents.infrastructure.sandbox.transaction_sandbox import TransactionSandbox from cleveragents.infrastructure.sandbox.transaction_sandbox import TransactionSandbox
if TYPE_CHECKING:
from cleveragents.infrastructure.sandbox.strategy_registry import (
SandboxStrategyRegistry,
)
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
# Strategy string constants aligned with spec SandboxStrategy enum (5 values) # Strategy string constants aligned with spec SandboxStrategy enum (5 values)
@@ -70,8 +76,25 @@ class SandboxFactory:
- ``"transaction_rollback"`` -> :class:`TransactionSandbox` - ``"transaction_rollback"`` -> :class:`TransactionSandbox`
``"snapshot"`` raises ``NotImplementedError``. ``"snapshot"`` raises ``NotImplementedError``.
Custom strategies registered in a :class:`SandboxStrategyRegistry` can
be resolved by passing the registry at construction time.
""" """
def __init__(
self,
*,
custom_registry: SandboxStrategyRegistry | None = None,
) -> None:
"""Initialise the factory.
Args:
custom_registry: Optional registry of custom sandbox strategies.
When provided, the factory will fall back to this registry
for strategy names not matched by built-in strategies.
"""
self._custom_registry = custom_registry
def create_sandbox( def create_sandbox(
self, self,
resource_id: str, resource_id: str,
@@ -128,6 +151,34 @@ class SandboxFactory:
raise ValueError(f"Unknown sandbox strategy: {sandbox_strategy}") raise ValueError(f"Unknown sandbox strategy: {sandbox_strategy}")
# -- custom strategy lookup ----------------------------------------------
def has_custom_strategy(self, name: str) -> bool:
"""Check whether a custom strategy is registered.
Args:
name: Strategy name to check.
Returns:
``True`` if a custom strategy with that name exists.
"""
if self._custom_registry is None:
return False
return self._custom_registry.has(name)
def get_custom_strategy_class(self, name: str) -> type[object] | None:
"""Look up a custom strategy class by name.
Args:
name: Strategy name.
Returns:
The strategy class, or ``None`` if not found.
"""
if self._custom_registry is None:
return None
return self._custom_registry.get(name)
# -- validation helpers -------------------------------------------------- # -- validation helpers --------------------------------------------------
@staticmethod @staticmethod
@@ -0,0 +1,252 @@
"""Adapter that wraps existing Sandbox implementations as SandboxStrategyProtocol.
Bridges the existing :class:`Sandbox` protocol (infrastructure layer) to the
new :class:`SandboxStrategyProtocol` (domain layer). This allows built-in
sandbox implementations (NoSandbox, CopyOnWriteSandbox, GitWorktreeSandbox,
TransactionSandbox) to be used through the unified custom-strategy API.
Based on issue #586.
"""
from __future__ import annotations
import logging
import os
from datetime import datetime
from cleveragents.domain.models.core.resource import Resource
from cleveragents.domain.models.core.sandbox_strategy import (
DiffEntry,
DiffView,
SandboxRef,
)
from cleveragents.infrastructure.sandbox.factory import (
SandboxFactory,
SandboxStrategyStr,
)
from cleveragents.infrastructure.sandbox.protocol import (
Sandbox,
)
logger = logging.getLogger(__name__)
class BuiltInSandboxStrategyAdapter:
"""Adapts a built-in :class:`Sandbox` implementation to the 9-method Protocol.
Wraps the existing ``Sandbox`` lifecycle (create, get_path, commit,
rollback, cleanup) and adds read/write/diff/checkpoint/restore
operations on top. Checkpoint storage is in-memory per adapter
instance.
This adapter satisfies :class:`SandboxStrategyProtocol` structurally.
Args:
strategy_name: The built-in strategy name (e.g. ``"none"``).
factory: Optional :class:`SandboxFactory` to create sandbox instances.
"""
def __init__(
self,
strategy_name: SandboxStrategyStr = "none",
factory: SandboxFactory | None = None,
) -> None:
self._strategy_name: SandboxStrategyStr = strategy_name
self._factory = factory or SandboxFactory()
self._sandboxes: dict[str, Sandbox] = {}
self._checkpoints: dict[str, dict[str, bytes]] = {}
def create(self, plan_id: str, resource: Resource) -> SandboxRef:
"""Create a sandbox using the built-in factory and return a ref.
Args:
plan_id: Identifier of the plan requesting the sandbox.
resource: The resource to sandbox.
Returns:
A :class:`SandboxRef` wrapping the internal sandbox instance.
"""
location = resource.location or ""
sandbox = self._factory.create_sandbox(
resource_id=resource.resource_id,
original_path=location,
sandbox_strategy=self._strategy_name,
)
ctx = sandbox.create(plan_id)
self._sandboxes[ctx.sandbox_id] = sandbox
return SandboxRef(
sandbox_id=ctx.sandbox_id,
plan_id=plan_id,
resource_id=resource.resource_id,
created_at=datetime.now(),
metadata={"strategy": self._strategy_name},
)
def read(self, ref: SandboxRef, path: str) -> bytes:
"""Read file content from the sandbox.
Args:
ref: Sandbox reference.
path: Relative file path.
Returns:
File content as bytes.
Raises:
FileNotFoundError: If the file does not exist.
KeyError: If the sandbox ref is unknown.
"""
sandbox = self._sandboxes[ref.sandbox_id]
full_path = sandbox.get_path(path)
with open(full_path, "rb") as fh:
return fh.read()
def write(self, ref: SandboxRef, path: str, content: bytes) -> DiffEntry:
"""Write content to a file in the sandbox.
Args:
ref: Sandbox reference.
path: Relative file path.
content: Content to write.
Returns:
A :class:`DiffEntry` describing the change.
"""
sandbox = self._sandboxes[ref.sandbox_id]
full_path = sandbox.get_path(path)
existed = os.path.exists(full_path)
before: bytes | None = None
if existed:
with open(full_path, "rb") as fh:
before = fh.read()
parent_dir = os.path.dirname(full_path)
if parent_dir:
os.makedirs(parent_dir, exist_ok=True)
with open(full_path, "wb") as fh:
fh.write(content)
operation = "modified" if existed else "added"
return DiffEntry(
path=path,
operation=operation,
before=before,
after=content,
)
def diff(self, ref: SandboxRef) -> DiffView:
"""Return a diff view of sandbox changes.
For built-in strategies, returns a minimal summary since the
underlying Sandbox protocol does not expose diff information
until commit.
Args:
ref: Sandbox reference.
Returns:
A :class:`DiffView` (may be empty for built-in strategies).
"""
return DiffView(
sandbox_id=ref.sandbox_id,
entries=[],
summary="Built-in strategy; diff computed at commit time.",
)
def commit(self, ref: SandboxRef) -> None:
"""Commit sandbox changes via the underlying Sandbox.
Args:
ref: Sandbox reference.
"""
sandbox = self._sandboxes[ref.sandbox_id]
sandbox.commit()
def rollback(self, ref: SandboxRef) -> None:
"""Rollback sandbox changes via the underlying Sandbox.
Args:
ref: Sandbox reference.
"""
sandbox = self._sandboxes[ref.sandbox_id]
sandbox.rollback()
def checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None:
"""Save a named checkpoint (in-memory snapshot for adapter).
Args:
ref: Sandbox reference.
checkpoint_id: Unique checkpoint identifier.
"""
key = f"{ref.sandbox_id}:{checkpoint_id}"
sandbox = self._sandboxes[ref.sandbox_id]
# Snapshot current sandbox state if sandbox is filesystem-based
ctx = sandbox.context
snapshot: dict[str, bytes] = {}
if ctx is not None and os.path.isdir(ctx.sandbox_path):
for dirpath, _dirs, files in os.walk(ctx.sandbox_path):
for fname in files:
full = os.path.join(dirpath, fname)
rel = os.path.relpath(full, ctx.sandbox_path)
with open(full, "rb") as fh:
snapshot[rel] = fh.read()
self._checkpoints[key] = snapshot
logger.debug(
"Checkpoint saved: sandbox=%s checkpoint=%s files=%d",
ref.sandbox_id,
checkpoint_id,
len(snapshot),
)
def restore_checkpoint(self, ref: SandboxRef, checkpoint_id: str) -> None:
"""Restore sandbox state from a named checkpoint.
Args:
ref: Sandbox reference.
checkpoint_id: Identifier of the checkpoint to restore.
Raises:
KeyError: If the checkpoint does not exist.
"""
key = f"{ref.sandbox_id}:{checkpoint_id}"
snapshot = self._checkpoints[key]
sandbox = self._sandboxes[ref.sandbox_id]
ctx = sandbox.context
if ctx is not None and os.path.isdir(ctx.sandbox_path):
# Restore files from snapshot
for rel_path, content in snapshot.items():
full = os.path.join(ctx.sandbox_path, rel_path)
parent = os.path.dirname(full)
if parent:
os.makedirs(parent, exist_ok=True)
with open(full, "wb") as fh:
fh.write(content)
logger.debug(
"Checkpoint restored: sandbox=%s checkpoint=%s",
ref.sandbox_id,
checkpoint_id,
)
def cleanup(self, ref: SandboxRef) -> None:
"""Cleanup sandbox resources.
Args:
ref: Sandbox reference.
"""
sandbox = self._sandboxes.pop(ref.sandbox_id, None)
if sandbox is not None:
sandbox.cleanup()
# Remove associated checkpoints
keys_to_remove = [
k for k in self._checkpoints if k.startswith(f"{ref.sandbox_id}:")
]
for k in keys_to_remove:
del self._checkpoints[k]
@@ -0,0 +1,298 @@
"""Config-driven custom sandbox strategy registry.
Provides :class:`SandboxStrategyRegistry` for registering, validating,
and resolving custom sandbox strategies. Strategies are registered
either programmatically or via configuration keys::
sandbox.custom_strategies.<name>.module = "my_package.my_module"
sandbox.custom_strategies.<name>.class = "MySandboxClass"
All custom strategies are validated against the
:class:`SandboxStrategyProtocol` at registration time using
``@runtime_checkable`` Protocol checks.
Thread-safe via ``threading.RLock``.
Based on issue #586.
"""
from __future__ import annotations
import logging
import threading
from typing import Any
from cleveragents.infrastructure.plugins.exceptions import (
PluginLoadError,
ProtocolMismatchError,
)
from cleveragents.infrastructure.plugins.loader import PluginLoader
logger = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
# Custom strategy configuration model
# ---------------------------------------------------------------------------
class CustomStrategyConfig:
"""Configuration for a single custom sandbox strategy.
Attributes:
name: Strategy name (used as the key in lookups).
module: Python module path containing the strategy class.
class_name: Name of the class within the module.
options: Optional strategy-specific configuration.
"""
__slots__ = ("class_name", "module", "name", "options")
def __init__(
self,
name: str,
module: str,
class_name: str,
options: dict[str, Any] | None = None,
) -> None:
if not name:
raise ValueError("Strategy name cannot be empty")
if not module:
raise ValueError("Strategy module cannot be empty")
if not class_name:
raise ValueError("Strategy class_name cannot be empty")
self.name = name
self.module = module
self.class_name = class_name
self.options = options or {}
# ---------------------------------------------------------------------------
# SandboxStrategyRegistry
# ---------------------------------------------------------------------------
class SandboxStrategyRegistry:
"""Registry for custom sandbox strategies with Protocol validation.
Maintains a thread-safe mapping of strategy names to their
implementing classes. Strategies are validated against the
:class:`SandboxStrategyProtocol` at registration time.
Built-in strategies (``none``, ``git_worktree``, ``copy_on_write``,
``transaction_rollback``) are not managed by this registry; they
are handled by :class:`SandboxFactory`.
Usage::
registry = SandboxStrategyRegistry()
registry.register(
"my_strategy",
"cleveragents.my_module",
"MySandbox",
)
cls = registry.get("my_strategy")
Args:
loader: Optional :class:`PluginLoader` for dynamic imports.
allowed_prefixes: Module prefix allowlist for the loader.
"""
def __init__(
self,
*,
loader: PluginLoader | None = None,
allowed_prefixes: tuple[str, ...] | None = None,
) -> None:
self._lock = threading.RLock()
self._loader = loader or PluginLoader(allowed_prefixes=allowed_prefixes)
self._strategies: dict[str, type[Any]] = {}
# ------------------------------------------------------------------
# Registration
# ------------------------------------------------------------------
def register(
self,
name: str,
module_path: str,
class_name: str,
) -> type[Any]:
"""Register a custom sandbox strategy.
Loads the class via the plugin loader, validates it against
:class:`SandboxStrategyProtocol`, and stores it.
Args:
name: Strategy name for lookup.
module_path: Fully-qualified module path.
class_name: Class name within the module.
Returns:
The loaded and validated class.
Raises:
ValueError: If *name* is empty.
PluginLoadError: If the class cannot be imported.
ProtocolMismatchError: If the class does not satisfy
the SandboxStrategyProtocol.
"""
if not name:
raise ValueError("Strategy name cannot be empty")
cls = self._loader.load_class(module_path, class_name)
self._validate_protocol(cls)
with self._lock:
self._strategies[name] = cls
logger.info(
"Registered custom sandbox strategy: name=%s class=%s.%s",
name,
module_path,
class_name,
)
return cls
def register_from_config(self, config: CustomStrategyConfig) -> type[Any]:
"""Register a strategy from a :class:`CustomStrategyConfig`.
Args:
config: Configuration describing the strategy.
Returns:
The loaded and validated class.
"""
return self.register(config.name, config.module, config.class_name)
def register_all_from_config(
self,
configs: dict[str, dict[str, Any]],
) -> list[str]:
"""Register multiple strategies from a config dictionary.
The dictionary maps strategy names to dicts with ``module``
and ``class`` keys::
{
"my_strategy": {
"module": "my_package.module",
"class": "MyStrategy",
},
}
Args:
configs: Strategy configuration dictionary.
Returns:
List of successfully registered strategy names.
"""
registered: list[str] = []
for name, cfg in configs.items():
module = cfg.get("module", "")
class_name = cfg.get("class", "")
if not module or not class_name:
logger.warning(
"Skipping custom strategy '%s': missing module or class",
name,
)
continue
try:
self.register(name, module, class_name)
registered.append(name)
except (PluginLoadError, ProtocolMismatchError, ValueError) as exc:
logger.error(
"Failed to register custom strategy '%s': %s",
name,
exc,
)
return registered
# ------------------------------------------------------------------
# Lookup
# ------------------------------------------------------------------
def get(self, name: str) -> type[Any] | None:
"""Look up a registered custom strategy class.
Args:
name: Strategy name.
Returns:
The class, or ``None`` if not registered.
"""
with self._lock:
return self._strategies.get(name)
def has(self, name: str) -> bool:
"""Check whether a strategy is registered.
Args:
name: Strategy name.
Returns:
``True`` if the strategy is registered.
"""
with self._lock:
return name in self._strategies
def list_strategies(self) -> list[str]:
"""Return names of all registered custom strategies.
Returns:
Sorted list of strategy names.
"""
with self._lock:
return sorted(self._strategies.keys())
def clear(self) -> None:
"""Remove all registered custom strategies."""
with self._lock:
self._strategies.clear()
logger.debug("Custom sandbox strategy registry cleared")
# ------------------------------------------------------------------
# Validation
# ------------------------------------------------------------------
@staticmethod
def _validate_protocol(cls: type[Any]) -> None:
"""Validate that a class satisfies :class:`SandboxStrategyProtocol`.
Uses structural subtyping: checks that all 9 required methods
are present as callable attributes on the class.
Args:
cls: The class to validate.
Raises:
ProtocolMismatchError: If the class is missing required methods.
"""
required_methods = (
"create",
"read",
"write",
"diff",
"commit",
"rollback",
"checkpoint",
"restore_checkpoint",
"cleanup",
)
missing = [
method
for method in required_methods
if not callable(getattr(cls, method, None))
]
if missing:
msg = (
f"Class '{cls.__name__}' does not satisfy "
f"SandboxStrategyProtocol. Missing methods: "
f"{', '.join(missing)}"
)
raise ProtocolMismatchError(msg)