perf(ai): decouple AI suite from startup and enable lazy loading across services and CLI

- Decouple connapp.ai from get_parser() via DeferredAIProxy to queue plugin modifications without importing connpy.ai or Pydantic v2 upfront.
- Guard connapp.start() finally block to invoke ai.cleanup() only if connpy.ai was actively imported.
- Defer CopilotInterface and AIService imports in core.py to execute strictly upon receiving Copilot hotkey (\x00).
- Convert heavy services (ai, sync, users, import_export, system, execution) into lazy properties on ServiceProvider and services/__init__.py.
- Implement O(1) globals() dict caching in module __getattr__ (PEP 562) across connpy/__init__.py and services/__init__.py.
- Defer Google Drive SDK in sync_service.py and interactive CLI components (inquirer) across handlers.
- Reduced CLI startup & parser initialization latency from ~1.47s to ~183ms (~87% improvement).
This commit is contained in:
fluzzi32 committed 2026-08-11 19:31:55 -03:00
1 parent bf5bddda5a
commit 835283eba6
8 files changed
+72 -20

No files matched your search

+3 -2
View File
@@ -712,13 +712,14 @@ class node:
def _copilot_handler(self, config):
"""Unified copilot handler for local session."""
from .cli.terminal_ui import CopilotInterface
from .services.ai_service import AIService
import asyncio
import os
async def handler(buffer, node_info, stream, child_fd, cmd_byte_positions=None):
try:
from .cli.terminal_ui import CopilotInterface
from .services.ai_service import AIService
interface = CopilotInterface(
config,
history=getattr(stream, 'copilot_history', None),