forked from HAL9000/cleveragents-core
167 lines
5.2 KiB
Python
167 lines
5.2 KiB
Python
"""ASV benchmarks for Plan domain model validation and serialization.
|
|
|
|
Measures the performance of:
|
|
- Plan model construction (Pydantic validation)
|
|
- Plan.as_cli_dict() serialization
|
|
- ProjectLink alias validation
|
|
- PlanInvariant construction
|
|
- Phase transition validation via can_transition()
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
try:
|
|
from cleveragents.domain.models.core.plan import (
|
|
AutomationLevel,
|
|
InvariantScope,
|
|
NamespacedName,
|
|
Plan,
|
|
PlanIdentity,
|
|
PlanInvariant,
|
|
PlanPhase,
|
|
ProcessingState,
|
|
ProjectLink,
|
|
can_transition,
|
|
)
|
|
except ModuleNotFoundError:
|
|
sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "src"))
|
|
from cleveragents.domain.models.core.plan import (
|
|
AutomationLevel,
|
|
InvariantScope,
|
|
NamespacedName,
|
|
Plan,
|
|
PlanIdentity,
|
|
PlanInvariant,
|
|
PlanPhase,
|
|
ProcessingState,
|
|
ProjectLink,
|
|
can_transition,
|
|
)
|
|
|
|
|
|
def _make_plan() -> Plan:
|
|
"""Create a fully-populated Plan for benchmarking."""
|
|
return Plan(
|
|
identity=PlanIdentity(plan_id="01ARZ3NDEKTSV4RRFFQ69G5FAV"),
|
|
namespaced_name=NamespacedName(
|
|
server=None, namespace="local", name="bench-plan"
|
|
),
|
|
description="Benchmark plan for validation and serialization",
|
|
action_name="local/bench-action",
|
|
definition_of_done="All benchmarks pass",
|
|
phase=PlanPhase.STRATEGIZE,
|
|
state=ProcessingState.QUEUED,
|
|
automation_level=AutomationLevel.REVIEW_BEFORE_APPLY,
|
|
automation_profile_name="org/bench-profile",
|
|
effective_profile_snapshot={
|
|
"max_file_changes": 50,
|
|
"require_tests": True,
|
|
},
|
|
project_links=[
|
|
ProjectLink(project_name="local/project-a", alias="main"),
|
|
ProjectLink(project_name="local/project-b", alias="lib", read_only=True),
|
|
],
|
|
invariants=[
|
|
PlanInvariant(
|
|
text="Global invariant",
|
|
scope=InvariantScope.GLOBAL,
|
|
source_name="system",
|
|
),
|
|
PlanInvariant(
|
|
text="Project invariant",
|
|
scope=InvariantScope.PROJECT,
|
|
source_name="my-project",
|
|
),
|
|
PlanInvariant(
|
|
text="Plan invariant",
|
|
scope=InvariantScope.PLAN,
|
|
source_name=None,
|
|
),
|
|
],
|
|
strategy_actor="openai/gpt-4",
|
|
execution_actor="openai/gpt-4",
|
|
created_by="bench-user",
|
|
tags=["benchmark", "test"],
|
|
reusable=True,
|
|
read_only=False,
|
|
)
|
|
|
|
|
|
class PlanValidationSuite:
|
|
"""Benchmark Plan model construction (Pydantic validation)."""
|
|
|
|
def time_plan_construction(self) -> None:
|
|
"""Benchmark creating a fully-populated Plan object."""
|
|
_make_plan()
|
|
|
|
def time_plan_minimal_construction(self) -> None:
|
|
"""Benchmark creating a minimal Plan object."""
|
|
Plan(
|
|
identity=PlanIdentity(plan_id="01ARZ3NDEKTSV4RRFFQ69G5FAV"),
|
|
namespaced_name=NamespacedName(
|
|
server=None, namespace="local", name="min-plan"
|
|
),
|
|
description="Minimal plan",
|
|
action_name="local/min-action",
|
|
phase=PlanPhase.STRATEGIZE,
|
|
state=ProcessingState.QUEUED,
|
|
)
|
|
|
|
def time_project_link_with_alias(self) -> None:
|
|
"""Benchmark ProjectLink construction with alias validation."""
|
|
ProjectLink(
|
|
project_name="local/my-project",
|
|
alias="my-alias",
|
|
read_only=False,
|
|
)
|
|
|
|
def time_plan_invariant_construction(self) -> None:
|
|
"""Benchmark PlanInvariant construction with text validation."""
|
|
PlanInvariant(
|
|
text="No secrets in code",
|
|
scope=InvariantScope.GLOBAL,
|
|
source_name="system",
|
|
)
|
|
|
|
|
|
class PlanSerializationSuite:
|
|
"""Benchmark Plan serialization methods."""
|
|
|
|
def setup(self) -> None:
|
|
"""Create a plan for serialization benchmarks."""
|
|
self.plan = _make_plan()
|
|
|
|
def time_as_cli_dict(self) -> None:
|
|
"""Benchmark Plan.as_cli_dict() ordered dict generation."""
|
|
self.plan.as_cli_dict()
|
|
|
|
def time_model_dump(self) -> None:
|
|
"""Benchmark Pydantic model_dump() serialization."""
|
|
self.plan.model_dump()
|
|
|
|
def time_model_dump_json(self) -> None:
|
|
"""Benchmark Pydantic model_dump_json() JSON serialization."""
|
|
self.plan.model_dump_json()
|
|
|
|
|
|
class PhaseTransitionSuite:
|
|
"""Benchmark phase transition validation."""
|
|
|
|
def time_valid_transition(self) -> None:
|
|
"""Benchmark checking a valid phase transition."""
|
|
can_transition(PlanPhase.STRATEGIZE, PlanPhase.EXECUTE)
|
|
|
|
def time_invalid_transition(self) -> None:
|
|
"""Benchmark checking an invalid phase transition."""
|
|
can_transition(PlanPhase.STRATEGIZE, PlanPhase.APPLY)
|
|
|
|
def time_all_transitions(self) -> None:
|
|
"""Benchmark checking all possible phase transitions."""
|
|
phases = list(PlanPhase)
|
|
for from_phase in phases:
|
|
for to_phase in phases:
|
|
can_transition(from_phase, to_phase)
|