split_coder.py

#!/usr/bin/env python3
"""Split a monolithic Python program into an executable, source-complete package.

The splitter preserves every top-level statement, stores the real source in the
owning module, and generates a small runtime that initializes those statements
in their original order.  The ordered initialization is important for legacy
monoliths whose modules form cycles or whose registries are built by top-level
loops.  The generated package never imports or reads the original monolith.
"""

from __future__ import annotations

import argparse
import ast
import hashlib
import json
import os
import py_compile
import re
import shutil
import subprocess
import symtable
import sys
import tempfile
import textwrap
import unicodedata
from collections import defaultdict
from dataclasses import asdict, dataclass, field
from pathlib import Path

GENERATOR_VERSION = "3.1"
GENERATED_HEADER = (
    "# Auto-generated by split_coder.py v3.1 — do not edit manually.\n"
    "# Re-run run_split.py to regenerate.\n"
    "# flake8: noqa: F821\n"
    "# ruff: noqa: F401, F821\n"
)


@dataclass
class NodeInfo:
    """One source module top-level statement."""

    order: int
    name: str
    kind: str
    lineno: int
    end_lineno: int
    source_start: int
    source_end: int
    source_hash: str = ""
    target_module: str = ""
    bound_names: list[str] = field(default_factory=list)
    bases: list[str] = field(default_factory=list)
    generated_ast_index: int = -1


@dataclass
class ManifestEntry:
    order: int
    name: str
    kind: str
    lineno: int
    end_lineno: int
    source_start: int
    source_end: int
    source_hash: str
    target_module: str
    bound_names: list[str] = field(default_factory=list)
    generated_ast_index: int = -1


@dataclass
class SplitManifest:
    source_path: str
    source_hash: str
    output_dir: str
    package_name: str
    generator_version: str = GENERATOR_VERSION
    source_statement_count: int = 0
    nodes: list[ManifestEntry] = field(default_factory=list)
    layout_config: str = ""

    FILENAME = ".split_manifest.json"

    def save(self, output_dir: Path) -> None:
        (output_dir / self.FILENAME).write_text(
            json.dumps(asdict(self), indent=2, ensure_ascii=False),
            encoding="utf-8",
        )

    @classmethod
    def load(cls, output_dir: Path) -> SplitManifest | None:
        path = output_dir / cls.FILENAME
        if not path.exists():
            return None
        try:
            data = json.loads(path.read_text(encoding="utf-8"))
            if str(data.get("generator_version", "")) != GENERATOR_VERSION:
                return None
            data["nodes"] = [ManifestEntry(**row) for row in data.get("nodes", [])]
            return cls(**data)
        except Exception:
            return None


@dataclass(frozen=True)
class ImportAlias:
    name: str
    asname: str | None = None

    @property
    def bound_name(self) -> str:
        return self.asname or self.name.split(".")[0]


@dataclass
class ImportStatement:
    kind: str
    lineno: int
    end_lineno: int
    bound_names: set[str] = field(default_factory=set)


# Runtime families used by both the maintained layout and generic auto-layout.
# Keep state, locks, helpers and their owner together instead of hiding private
# names in a generic constants module or accepting an unclassified owner.
SHARED_RUNTIME_LAYOUT: dict[str, list[str]] = {
    "config/bootstrap.py": ["sync_embedded_liquid_kernel_bundle"],
    "llm/gate.py": [
        "SessionLLMGate", "LLMGateBusy", "TodoLLMBudgetExhausted",
        "~^_?SESSION_LLM_(?:CONCURRENCY|GATE)", "DEFAULT_SESSION_LLM_CONCURRENCY",
        "_session_llm_concurrency_env_default", "set_session_llm_concurrency_on_runtime",
        "~^_?TODO_LLM_", "_todo_llm_timeout_env_default", "todo_llm_budget_scope",
        "_consume_todo_llm_budget", "todo_llm_budget_active", "todo_llm_skip_reason",
    ],
    "ide/toolchains.py": [
        "~^_?IDE_TOOLCHAIN_", "~^_ide_toolchain_", "_probe_ide_toolchains",
        "_refresh_ide_toolchains_in_background",
    ],
    "agent/context.py": ["_READ_DISPATCH_CONTEXT", "compact_successful_write_history"],
    "agent/evidence.py": ["EvidenceRetrievalController", "_EVIDENCE_CONTROLLER_INIT_LOCK"],
    "agent/process.py": ["_TOOL_CANCEL_LOCAL", "_TOOL_ORPHAN_LOCK"],
    "agent/tools.py": ["bash_operation_is_observation", "bash_operation_is_marked_observation"],
    "skills/store.py": ["~^_skill_(?:term|recall)_"],
    "utils/text.py": [
        "~^_TODO_SHORT_ID_", "normalize_todo_short_id", "strip_todo_short_id_tag",
        "split_todo_status_text_with_id", "is_todo_transport_metadata",
    ],
    "utils/json_utils.py": [
        "decode_model_json_object", "standard_model_json_object",
        "semantic_json_repair_request", "validate_semantic_json_repair",
    ],
    "llm/utils.py": ["evolution_worker_chat_messages"],
    "utils/files.py": ["_start_offline_js_warmup"],
}


# Exact symbols reflect the current Clouds_Coder architecture.  Regex rules are
# deliberately ordered from specific subsystems to the general constants bin.
DEFAULT_LAYOUT: dict[str, list[str]] = {
    "__init__.py": [],
    "app/__init__.py": [],
    "app/main.py": ["main", "~^_main_guard_"],
    "app/context.py": ["AppContext"],
    "app/services.py": ["TelemetryStore", "ApplicationRegistry"],

    "admin/__init__.py": [],
    "admin/auth.py": ["AdminAuthError", "AdminAuthStore", "trusted_client_ip"],
    "admin/config.py": ["~^_admin_"],
    "admin/constants.py": ["~^ADMIN_"],

    "config/__init__.py": [],
    "config/paths.py": [
        "SCRIPT_DIR", "WORKDIR", "CODES_ROOT", "LLM_CONFIG_PATH", "REPO_ROOT",
        "_resolve_default_agent_workdir", "_migrate_legacy_runtime_roots", "detect_repo_root",
        "_is_installed_python_runtime", "_runtime_storage_mode", "_runtime_tree_has_content",
        "_copy_runtime_tree_with_crypto_migration", "_merge_legacy_codes_root",
    ],
    "config/settings.py": [
        "~^(?:normalize|extract|infer|load|looks_like|parse|merge|resolve|select|supported|backend|runtime_environment|model_language|default_multimodal|set_web_search)_",
        "_to_bool_like", "_detect_os_shell_instruction", "_count_skill_markdown_files",
        "extract_single_no_plan_todo_settings", "_single_no_plan_todo_setting_sections",
        "_single_no_plan_todo_setting_present", "user_memory_enabled_from_mode",
        "task_complexity_rank", "task_complexity_at_least", "max_task_complexity",
        "agent_language_preference_payload",
        "probe_and_merge_model_profiles", "apply_model_runtime_settings",
        "model_runtime_settings_for", "apply_model_option_runtime_fields",
    ],

    "web/__init__.py": [],
    "web/assets.py": ["INDEX_HTML", "APP_CSS", "APP_JS", "APP_TS"],
    "web/skills_assets.py": ["SKILLS_INDEX_HTML", "SKILLS_EXTRA_CSS", "SKILLS_APP_JS"],
    "web/admin_assets.py": ["ADMIN_INDEX_HTML", "ADMIN_CSS", "ADMIN_JS"],

    "ide/__init__.py": [],
    "ide/assets.py": ["IDE_INDEX_HTML", "IDE_CSS", "IDE_JS"],
    "ide/auth.py": ["IDEAuthError", "IDEAuthStore"],
    "ide/errors.py": ["IDECapabilityError", "IDEFileConflict"],
    "ide/events.py": ["ide_public_operation_data"],
    "ide/preview.py": [
        "normalize_rel_preview_path", "is_code_preview_candidate", "preview_kind_for_path",
        "normalize_markdown_preview_text", "build_code_preview_rows",
        "workspace_file_revision_map", "workspace_revision_delta",
        "~^_preview_markdown_",
    ],
    "ide/sandbox.py": [
        "WINDOWS_JOB_SANDBOX_MARKER",
        "~^_IDE_SANDBOX_", "~^_WINDOWS_LOW_INTEGRITY_",
        "~^_(?:is_windows_job_sandbox_prefix|windows_|popen_windows_sandboxed|run_windows_sandboxed_command|detect_ide_sandbox_backend)$",
    ],
    "ide/handler.py": ["IdeHandler"],

    "utils/__init__.py": [],
    "utils/http.py": [
        "_URL_OPEN_ORIGINAL", "_HTTP_SSL_CONTEXT", "_shared_http_ssl_context", "urlopen",
        "json_response_bytes", "read_http_json_body", "close_if_http_request_body_unread",
    ],
    "utils/errors.py": ["~(?:Error|Exception)$"],
    "utils/text.py": [
        "trim", "display_clean", "short_title_from", "filter_runtime_noise_lines",
        "normalize_embedded_newlines", "_map_todo_status_token", "split_todo_status_text",
        "extract_todo_rows_from_text", "decode_structured_todo_container",
        "infer_todo_status_from_text", "split_structured_todo_content", "normalize_work_text",
        "_fmt_export_ts", "_html_esc", "_text_to_minimal_pdf", "make_unified_diff",
        "make_numbered_diff", "render_numbered_diff_text", "_skip_row", "_row_is_hot",
        "_hotspot_index", "_compress_rows_keep_hotspot", "_focused_diff_rows_from_opcodes",
        "is_synthetic_public_progress",
        "safe_utf8_bytes", "escape_invalid_utf8_text", "sanitize_utf8_surrogates",
        "decode_utf8_replace", "MAX_TOOL_OUTPUT", "SOCKET_NOISE_LINE_PATTERNS",
    ],
    "utils/json_utils.py": [
        "json_dumps", "parse_json_object", "extract_json_object_from_text",
        "parse_tool_arguments", "repair_truncated_json_object", "parse_tool_arguments_with_error",
        "_is_valid_json_object", "_scan_top_level_json_objects", "reconstruct_streamed_tool_args",
        "_json_default_copy", "_read_json_file", "_write_json_file", "JSON_FSYNC_ENABLED",
    ],
    "utils/files.py": [
        "safe_path", "try_read_text", "_normalize_js_lib_asset_ref",
        "~^(?:_?sha256|_?safe_js|_?download_http|offline_js|load_offline_js|ensure_offline_js|cache_external_js|is_external_js|match_offline_js|_normalize_external_js|_resolve_js|_discover_extra_js|_offline_js|_archive_member|_path_size|_extract_archive|_package_|_postprocess_offline|_ensure_offline|_render_offline_js)",
    ],
    "utils/media.py": [
        "guess_mime_from_name", "guess_ext_from_mime", "_convert_image_to_safe_format",
        "_capability_probe_png_bytes", "_capability_probe_audio_bytes",
        "_capability_probe_video_bytes",
    ],
    "utils/compress.py": ["compress_text_blob", "decompress_text_blob"],
    "utils/crypto.py": ["CryptoBox"],
    "utils/sqlite.py": [
        "_ClosingSQLiteConnection", "_connect_sqlite", "sqlite_failure_diagnostics",
    ],
    "utils/misc.py": [
        "now_ts", "make_id", "sanitize_profile_id", "user_id_from_ip", "_meta_string_list",
        "_module_exists", "is_benign_socket_error", "_socket_error_code",
        "_log_benign_socket_error_limited", "swallow_benign_socket_error",
        "normalize_timeout_seconds", "detect_local_lan_ip", "detect_local_lan_ip_cached",
        "_LOCAL_LAN_IP_CACHE", "_benign_socket_log_lock", "_benign_socket_log_state",
    ],

    "llm/__init__.py": [],
    "llm/constants.py": [
        "DEFAULT_OLLAMA_BASE_URL", "DEFAULT_OLLAMA_MODEL", "OPENAI_COMPAT_PROVIDER_NAMES",
        "OPENAI_LIKE_PROVIDER_NAMES", "~^(?:EFFORT_|TASK_LEVEL_EFFORT|ROLE_EFFORT|COORDINATION_EFFORT)",
    ],
    "llm/client.py": ["OllamaError", "OllamaClient"],
    "llm/utils.py": [
        "~^(?:probe_ollama|list_ollama|resolve_ollama|infer_thinking|split_thinking|strip_thinking|check_ollama|list_loaded_ollama|wake_ollama|try_pull_ollama|ordered_model|pick_working_ollama|extract_base_url|complete_chat_endpoint|normalize_openai|is_openai|openai_compat|extract_openai|clamp_effort|model_reasoning|resolve_reasoning|_is_http_url|_resolve_local_path)",
        "_OLLAMA_TAG_CACHE", "_OLLAMA_TAG_CACHE_LOCK",
        "_fetch_ollama_models_cached", "anthropic_model_list_url",
        "_PROVIDER_MODEL_CACHE_LOCK", "_PROVIDER_MODEL_CACHE",
        "_fetch_provider_models_cached", "probe_provider_models",
    ],

    "agent/__init__.py": [],
    "agent/events.py": ["EventHub"],
    "agent/todo.py": ["TodoManager"],
    "agent/tasks.py": ["TaskManager"],
    "agent/background.py": ["BackgroundManager"],
    "agent/bus.py": ["MessageBus"],
    "agent/worktree.py": ["WorktreeManager"],
    "agent/process.py": [
        "subprocess_text_encodings", "decode_subprocess_bytes", "run_subprocess_text",
        "windows_utf8_shell_command", "shell_process_invocation", "join_shell_task_command",
        "UserProcessManager",
    ],
    "agent/tools.py": [
        "tool_def", "TOOLS", "TOOL_REQUIRED_ARGS", "TOOL_SPEC_BY_NAME", "TOOL_NAME_FUZZY_MAP",
        "is_todo_resume_tool_name", "canonicalize_tool_name", "filter_tool_specs_for_runtime",
        "_ask_user_option_rows", "_ask_user_option_value",
        "DEVELOPER_TOOL_DROP", "AGENT_TOOL_ALLOWLIST",
    ],
    "agent/errors.py": ["CircuitBreakerTriggered"],

    "collaboration/__init__.py": [],
    "collaboration/core.py": [
        "_now", "_json", "_load_json", "_b64_token", "_branch_label", "_digest",
        "_password_hash", "_normalize_ip", "_normalize_name",
        "_COLLAB_PUBLIC_SECRET_PATTERNS", "_collaboration_public_text",
        "_collaboration_task_objective", "_collaboration_task_title",
        "_collaboration_task_key", "_collaboration_plan_steps", "CollaborationError",
        "CollaborationPrincipal", "_normalize_operation", "operation_input_length",
        "apply_text_operation", "transform_text_operation", "CollaborationStore",
        "CollaborationWriteCoordinator",
    ],
    "collaboration/watcher.py": [
        "collaboration_file_watcher_loop", "collaboration_watcher_health",
    ],

    "skills/__init__.py": [],
    "skills/embedded.py": [
        "EMBEDDED_SKILLS_ARCHIVE_B64", "EMBEDDED_SKILLS_ARCHIVE_SHA256",
        "EMBEDDED_SKILLS_ARCHIVE_FILES", "EMBEDDED_CLAWHUB_SKILLS_ARCHIVE_B64",
        "BUILTIN_CLAWHUB_SKILLS_VERSION", "MCP_BUILDER_SKILL_MD", "SKILL_PROTOCOL_SPECS",
        "SKILL_PROTOCOL_LOCAL", "SKILL_PROTOCOL_CLAWHUB", "SKILL_PROTOCOL_HTTP_JSON",
    ],
    "skills/provisioning.py": ["~^(?:ensure_|detect_upload|_render_cap|_write_text|_skill_knowledge|analyze_skill|_sanitize_skill|_build_skills)"],
    "skills/studio.py": [
        "SkillsStudioError", "_studio_slug", "_studio_hash", "_studio_cookie_value",
        "SkillsStudioStore",
    ],
    "skills/store.py": ["_BUILTIN_SKILLS", "SkillStore"],

    "mcp/__init__.py": [],
    "mcp/constants.py": ["~^MCP_", "~^_MCP_"],
    "mcp/driver.py": ["~^_?mcp_", "MCPServerProcess", "MCPManager"],
    "mcp/service.py": ["McpServiceHandler"],

    "session/__init__.py": [],
    "session/state.py": ["SessionState"],
    "session/manager.py": ["SessionCreationLimitExceeded", "SessionManager"],

    "rag/__init__.py": [],
    "rag/constants.py": ["~^(?:RAG_|CODE_LIBRARY_|WEB_SEARCH_|USER_MEMORY_)"],
    "rag/assets.py": [
        "RAG_ADMIN_INDEX_HTML", "RAG_ADMIN_CSS", "RAG_ADMIN_JS",
        "CODE_ADMIN_INDEX_HTML", "CODE_ADMIN_CSS", "CODE_ADMIN_JS",
    ],
    "rag/web_search.py": ["~^_agent_web_", "AgentWebHTMLParser", "AgentWebSearchEngine"],
    "rag/parsers.py": [
        "~^_rag_(?:safe|detect|cjk|is_noise|entity|filter|filename|apply|choose|tokenize|expand|extract|classify|chunk)",
        "~^_code_(?:language|is_test)", "_CallCollector", "_ALGO_COMPLEXITY_RE", "_ALGO_STEP_RE",
        "_ALGO_MATH_VARS", "_ALGO_DOC_KEYWORDS", "_detect_algo_chunk", "EvidenceRecord",
        "CodeContentParser", "RAGContentParser",
    ],
    "rag/index.py": ["TFGraphIDFIndex", "CodeGraphIndex", "~^_code_(?:module|choose|query)"],
    "rag/store.py": [
        "RAGLibraryStore", "WikiStore", "UserMemoryStore", "UserInteractionOptimizer",
        "UserIntentProfiler", "WorkflowMemoryStore", "CodeLibraryStore",
    ],
    "rag/ingestion.py": [
        "RAGIngestionService", "CodeIngestionService", "_rag_parse_file_worker",
        "~^_rag_(?:trigram|jaccard|mmr|embed|window|focused|query|parse_segments|boundary)",
    ],

    "server/__init__.py": [],
    "server/http.py": [
        "AgentHTTPServer", "Handler", "admin_language_payload",
        "_UI_TRUNCATION_MARKER", "_ui_trim_text", "_bounded_ui_value",
        "_bounded_ui_row", "_bounded_ui_rows", "_enforce_ui_payload_budget",
        "_apply_lite_snapshot_bounds",
    ],
    "server/skills.py": ["SkillsHandler"],
    "server/rag_admin.py": ["_RagAdminAuthMixin", "RagAdminHandler", "CodeAdminHandler"],

    # Specific regex buckets above run before this general constant rule.
    "config/constants.py": [
        "~^[A-Z][A-Z0-9_]{2,}$", "_SHELL_AUTO_CONFIRM_PATTERNS", "_TOOL_TIMEOUT_MAP",
        "_DEFAULT_TOOL_TIMEOUT", "_DEFAULT_SHELL_TIMEOUT_MODE_RAW",
    ],
    "config/bootstrap.py": [
        "_EMBEDDED_LIQUID_KERNEL_PACKAGE_B64", "_ensure_embedded_liquid_kernel_package",
        "_liquid_kernel_history_present", "prepare_liquid_kernel_runtime",
        "_persist_liquid_kernel_bootstrap",
        "_ensure_liquid_kernel_sqlite_lifecycle",
    ],
    "_unclassified.py": [],
}

# Place runtime-family rules before broad constant/settings regex buckets.
# Existing modules retain their original symbol rules as well.
DEFAULT_LAYOUT = {
    **SHARED_RUNTIME_LAYOUT,
    **{module: [*SHARED_RUNTIME_LAYOUT.get(module, []), *patterns]
       for module, patterns in DEFAULT_LAYOUT.items()},
}


class _TopLevelBindingCollector(ast.NodeVisitor):
    """Collect names stored by one statement without entering nested scopes."""

    def __init__(self) -> None:
        self.names: list[str] = []
        self._seen: set[str] = set()

    def add(self, name: str) -> None:
        if name and name not in self._seen:
            self._seen.add(name)
            self.names.append(name)

    def visit_Name(self, node: ast.Name) -> None:
        if isinstance(node.ctx, (ast.Store, ast.Del)):
            self.add(node.id)

    def visit_FunctionDef(self, node: ast.FunctionDef) -> None:
        self.add(node.name)

    def visit_AsyncFunctionDef(self, node: ast.AsyncFunctionDef) -> None:
        self.add(node.name)

    def visit_ClassDef(self, node: ast.ClassDef) -> None:
        self.add(node.name)

    def visit_Lambda(self, node: ast.Lambda) -> None:
        return

    # Python 3 comprehensions own a nested scope; their loop variables do not
    # survive as module attributes.
    def visit_ListComp(self, node: ast.ListComp) -> None:
        return

    def visit_SetComp(self, node: ast.SetComp) -> None:
        return

    def visit_DictComp(self, node: ast.DictComp) -> None:
        return

    def visit_GeneratorExp(self, node: ast.GeneratorExp) -> None:
        return

    def visit_Import(self, node: ast.Import) -> None:
        for alias in node.names:
            self.add(alias.asname or alias.name.split(".")[0])

    def visit_ImportFrom(self, node: ast.ImportFrom) -> None:
        for alias in node.names:
            if alias.name != "*":
                self.add(alias.asname or alias.name)

    def visit_ExceptHandler(self, node: ast.ExceptHandler) -> None:
        # An ``except ... as name`` binding is deleted when the handler exits;
        # it is not a persistent module export.
        for child in node.body:
            self.visit(child)


def top_level_bound_names(node: ast.AST) -> list[str]:
    collector = _TopLevelBindingCollector()
    collector.visit(node)
    return collector.names


class ArchitectureAnalyzer:
    def __init__(self, source_path: Path) -> None:
        self.source_path = source_path
        self.source_text = ""
        self.source_lines: list[str] = []
        self.tree: ast.Module | None = None
        self.nodes: list[NodeInfo] = []
        self.import_statements: list[ImportStatement] = []

    def analyze(self) -> list[NodeInfo]:
        self.source_text = self.source_path.read_text(encoding="utf-8-sig")
        self.source_lines = self.source_text.splitlines(keepends=True)
        print(f"  Parsing {len(self.source_lines):,} lines with AST...")
        self.tree = ast.parse(self.source_text, filename=str(self.source_path))
        self.nodes = []
        self.import_statements = []
        previous_end = 0
        body = list(self.tree.body)
        for order, raw_node in enumerate(body):
            node_end = int(getattr(raw_node, "end_lineno", None) or raw_node.lineno)
            source_end = node_end
            if order == len(body) - 1:
                source_end = len(self.source_lines)
            source_start = previous_end + 1
            previous_end = node_end
            info = self._make_info(order, raw_node, source_start, source_end)
            self.nodes.append(info)
            if isinstance(raw_node, (ast.Import, ast.ImportFrom, ast.Try)) and info.kind == "import":
                self.import_statements.append(
                    ImportStatement(info.kind, info.lineno, info.end_lineno, set(info.bound_names))
                )
        return self.nodes

    def _make_info(self, order: int, node: ast.AST, source_start: int, source_end: int) -> NodeInfo:
        bound = top_level_bound_names(node)
        bases: list[str] = []
        if isinstance(node, ast.ClassDef):
            name, kind = node.name, "class"
            bases = [self._name_of(base) for base in node.bases]
        elif isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
            name, kind = node.name, "function"
        elif isinstance(node, (ast.Import, ast.ImportFrom)):
            name, kind = f"_import_{node.lineno}", "import"
        elif isinstance(node, ast.Try) and self._is_import_try(node):
            name, kind = f"_try_import_{node.lineno}", "import"
        elif isinstance(node, (ast.Assign, ast.AnnAssign, ast.AugAssign)):
            name = bound[0] if bound else f"_assignment_{node.lineno}"
            kind = "constant" if re.match(r"^[A-Z][A-Z0-9_]{2,}$", name) else "assignment"
        elif isinstance(node, ast.Expr):
            name, kind = self._expression_name(node), "expression"
        elif isinstance(node, ast.If) and self._is_main_guard(node):
            name, kind = f"_main_guard_{node.lineno}", "main_guard"
        else:
            name = f"_{type(node).__name__.lower()}_{node.lineno}"
            kind = "statement"
        lineno = int(getattr(node, "lineno", source_start))
        end_lineno = int(getattr(node, "end_lineno", lineno) or lineno)
        chunk = self.get_source_lines(source_start, source_end)
        return NodeInfo(
            order=order,
            name=name,
            kind=kind,
            lineno=lineno,
            end_lineno=end_lineno,
            source_start=source_start,
            source_end=source_end,
            source_hash=hashlib.sha256(chunk.encode("utf-8")).hexdigest()[:16],
            bound_names=bound,
            bases=bases,
        )

    @staticmethod
    def _name_of(node: ast.AST) -> str:
        if isinstance(node, ast.Name):
            return node.id
        if isinstance(node, ast.Attribute):
            return node.attr
        try:
            return ast.unparse(node)
        except Exception:
            return ""

    @staticmethod
    def _is_import_try(node: ast.Try) -> bool:
        return any(isinstance(child, (ast.Import, ast.ImportFrom)) for child in ast.walk(node))

    @staticmethod
    def _is_main_guard(node: ast.If) -> bool:
        test = node.test
        return bool(
            isinstance(test, ast.Compare)
            and isinstance(test.left, ast.Name)
            and test.left.id == "__name__"
            and len(test.ops) == 1
            and isinstance(test.ops[0], ast.Eq)
            and len(test.comparators) == 1
            and isinstance(test.comparators[0], ast.Constant)
            and test.comparators[0].value == "__main__"
        )

    def _expression_name(self, node: ast.Expr) -> str:
        value = node.value
        if isinstance(value, ast.Call):
            try:
                raw = ast.unparse(value.func)
            except Exception:
                raw = "call"
            anchor = self._slug(raw) or "call"
            return f"_call_{anchor}_{node.lineno}"
        if isinstance(value, ast.Constant) and isinstance(value.value, str):
            return f"_module_docstring_{node.lineno}"
        return f"_expression_{node.lineno}"

    @staticmethod
    def _slug(value: object) -> str:
        raw = str(value or "").strip()
        norm = unicodedata.normalize("NFKD", raw).encode("ascii", "ignore").decode("ascii")
        norm = re.sub(r"[^a-zA-Z0-9]+", "_", norm).strip("_").lower()
        return norm[:48]

    def get_source_lines(self, start: int, end: int) -> str:
        return "".join(self.source_lines[start - 1:end])

    def get_node_source(self, node: NodeInfo) -> str:
        return self.get_source_lines(node.source_start, node.source_end)


class AutoLayoutGenerator:
    """Architecture heuristics used for new symbols not in the explicit map."""

    CLASS_RULES: list[tuple[re.Pattern[str], str]] = [
        (re.compile(r"^IDEAuth"), "ide/auth.py"),
        (re.compile(r"^IDE.*(?:Error|Conflict)$"), "ide/errors.py"),
        (re.compile(r"^IDE"), "ide/handler.py"),
        (re.compile(r"(?:Handler|HTTPServer)$"), "server/http.py"),
        (re.compile(r"^(?:RAG|Code).*(?:Store|Index)$"), "rag/store.py"),
        (re.compile(r"^(?:RAG|Code).*Ingestion"), "rag/ingestion.py"),
        (re.compile(r"^AgentWeb"), "rag/web_search.py"),
        (re.compile(r"^Ollama"), "llm/client.py"),
        (re.compile(r"^Session"), "session/manager.py"),
        (re.compile(r"^Skill"), "skills/store.py"),
    ]

    def generate(self, nodes: list[NodeInfo]) -> dict[str, list[str]]:
        layout: dict[str, list[str]] = {key: [] for key in DEFAULT_LAYOUT if key.endswith("__init__.py")}
        for node in nodes:
            if node.kind == "import":
                continue
            layout.setdefault(self._classify(node), []).append(node.name)
        return layout

    def _classify(self, node: NodeInfo) -> str:
        name = node.name
        for module, patterns in SHARED_RUNTIME_LAYOUT.items():
            if any(re.search(pattern[1:], name) if pattern.startswith("~") else name == pattern
                   for pattern in patterns):
                return module
        if name == "main" or node.kind == "main_guard":
            return "app/main.py"
        if name == "agent_language_preference_payload":
            return "config/settings.py"
        if name in {
            "_EMBEDDED_LIQUID_KERNEL_PACKAGE_B64", "_ensure_embedded_liquid_kernel_package",
            "_liquid_kernel_history_present", "prepare_liquid_kernel_runtime",
            "_persist_liquid_kernel_bootstrap",
        }:
            return "config/bootstrap.py"
        if name == "admin_language_payload" or name.startswith("_bounded_ui_") or name in {
            "_UI_TRUNCATION_MARKER", "_ui_trim_text", "_enforce_ui_payload_budget",
            "_apply_lite_snapshot_bounds",
        }:
            return "server/http.py"
        if name == "EvidenceRecord":
            return "rag/parsers.py"
        if name == "is_synthetic_public_progress":
            return "utils/text.py"
        if name == "ide_public_operation_data":
            return "ide/events.py"
        if name.startswith("_ask_user_"):
            return "agent/tools.py"
        if (
            name.startswith("_preview_markdown_")
            or name.startswith("workspace_")
            or name in {
                "normalize_rel_preview_path", "is_code_preview_candidate",
                "preview_kind_for_path", "normalize_markdown_preview_text",
                "build_code_preview_rows",
            }
        ):
            return "ide/preview.py"
        if (
            name.startswith(("_IDE_SANDBOX_", "_WINDOWS_LOW_INTEGRITY_", "_windows_"))
            or name in {
                "WINDOWS_JOB_SANDBOX_MARKER", "_is_windows_job_sandbox_prefix",
                "_popen_windows_sandboxed", "_run_windows_sandboxed_command",
                "_detect_ide_sandbox_backend",
            }
        ):
            return "ide/sandbox.py"
        if node.kind == "class":
            for regex, module in self.CLASS_RULES:
                if regex.search(name):
                    return module
        if name.startswith("_rag_") or name.startswith("RAG"):
            return "rag/parsers.py" if node.kind == "function" else "rag/constants.py"
        if name.startswith("_agent_web_"):
            return "rag/web_search.py"
        if name.startswith(("mcp_", "_mcp_", "MCP")):
            return "mcp/driver.py"
        if name.startswith("ensure_") and "skill" in name:
            return "skills/provisioning.py"
        if node.kind == "constant":
            return "config/constants.py"
        return "_unclassified.py"


class ModuleRouter:
    def __init__(self, layout: dict[str, list[str]]) -> None:
        self.layout = layout
        self.fallback = AutoLayoutGenerator()
        self.rules: list[tuple[str, list[tuple[str, object]]]] = []
        for module, patterns in layout.items():
            compiled: list[tuple[str, object]] = []
            for pattern in patterns:
                if pattern.startswith("~"):
                    compiled.append(("regex", re.compile(pattern[1:])))
                else:
                    compiled.append(("exact", pattern))
            self.rules.append((module, compiled))

    def route(self, node: NodeInfo) -> str:
        if node.kind == "import":
            return "_imports.py"
        for module, matchers in self.rules:
            if any(kind == "exact" and node.name == matcher for kind, matcher in matchers):
                return module
        for module, matchers in self.rules:
            for kind, matcher in matchers:
                if kind == "regex" and matcher.search(node.name):
                    return module
        return self.fallback._classify(node)

    def assign_all(self, nodes: list[NodeInfo]) -> None:
        for node in nodes:
            node.target_module = self.route(node)
        for index, node in enumerate(nodes):
            if node.target_module != "_unclassified.py" or node.kind not in {"expression", "statement"}:
                continue
            nearby = self._context_module(nodes, index)
            if nearby:
                node.target_module = nearby

    @staticmethod
    def _context_module(nodes: list[NodeInfo], index: int) -> str:
        candidates: list[tuple[int, int, str]] = []
        current = nodes[index]
        for direction, bias in ((-1, 0), (1, 1)):
            cursor = index + direction
            while 0 <= cursor < len(nodes):
                other = nodes[cursor]
                if other.kind != "import" and other.target_module != "_unclassified.py":
                    gap = (
                        max(0, current.lineno - other.end_lineno)
                        if direction < 0 else max(0, other.lineno - current.end_lineno)
                    )
                    candidates.append((gap, bias, other.target_module))
                    break
                cursor += direction
        if not candidates:
            return ""
        if len(candidates) == 2 and candidates[0][2] == candidates[1][2]:
            return candidates[0][2]
        candidates.sort()
        return candidates[0][2] if candidates[0][0] <= 240 else ""


class DependencyAnalyzer:
    def __init__(self, analyzer: ArchitectureAnalyzer, nodes: list[NodeInfo]) -> None:
        self.analyzer = analyzer
        self.nodes = nodes
        self.symbol_to_module: dict[str, str] = {}
        for node in nodes:
            if node.kind == "import":
                continue
            for name in node.bound_names:
                self.symbol_to_module[name] = node.target_module

    def compute_dependency_map(self, module: str, nodes: list[NodeInfo]) -> dict[str, set[str]]:
        deps: dict[str, set[str]] = defaultdict(set)
        for node in nodes:
            for name in self._referenced_names(node):
                owner = self.symbol_to_module.get(name)
                if owner and owner not in {module, "_imports.py"}:
                    deps[owner].add(name)
        return deps

    def _referenced_names(self, node: NodeInfo) -> set[str]:
        source = self.analyzer.get_node_source(node)
        try:
            table = symtable.symtable(source, str(self.analyzer.source_path), "exec")
        except SyntaxError:
            return set()
        out: set[str] = set()
        self._walk_symbols(table, out)
        return out

    def _walk_symbols(self, table: symtable.SymbolTable, out: set[str]) -> None:
        table_type = table.get_type()
        for symbol in table.get_symbols():
            if not symbol.is_referenced() or symbol.is_imported() or symbol.is_parameter():
                continue
            if table_type == "module" or symbol.is_global() or symbol.is_free() or symbol.is_nonlocal():
                out.add(symbol.get_name())
        for child in table.get_children():
            self._walk_symbols(child, out)


class CodeGenerator:
    RUNTIME_MODULE = "_runtime.py"
    IMPORTS_MODULE = "_imports.py"

    def __init__(self, analyzer: ArchitectureAnalyzer, package_name: str) -> None:
        self.analyzer = analyzer
        self.package_name = package_name

    def generate_source_module(self, module_path: str, nodes: list[NodeInfo]) -> str:
        parts = [GENERATED_HEADER.rstrip(), "", "from __future__ import annotations", ""]
        for ast_index, node in enumerate(nodes, start=1):
            node.generated_ast_index = ast_index
            parts.append(
                f"# split-source: order={node.order} original-lines={node.source_start}-{node.source_end} hash={node.source_hash}"
            )
            source = self._rewrite_context_sensitive_source(
                node,
                self.analyzer.get_node_source(node),
            )
            parts.append(source.rstrip("\n"))
            parts.append("")
        return "\n".join(parts).rstrip() + "\n"

    @staticmethod
    def _rewrite_context_sensitive_source(node: NodeInfo, source: str) -> str:
        """Make monolith-only ``__file__`` operations package-native."""
        if node.name == "_admin_supervised_restart":
            source = source.replace(
                "command = [sys.executable, script_path, *_admin_config_to_argv(config)]",
                "command = [*_SPLIT_ENTRY_COMMAND, *_admin_config_to_argv(config)]",
            )
            source = source.replace(
                "[sys.executable, script_path, *_admin_config_to_argv(fallback_config)]",
                "[*_SPLIT_ENTRY_COMMAND, *_admin_config_to_argv(fallback_config)]",
            )
            source = source.replace(
                "    script_path = str(Path(__file__).resolve())\n",
                "",
            )
        if node.name == "AppContext":
            source = source.replace(
                'zf.writestr("Clouds_Coder.py", Path(__file__).read_text(encoding="utf-8"))',
                "for _split_rel, _split_data in _split_package_source_files():\n"
                "                zf.writestr(f\"Code_Structure/{_split_rel}\", _split_data)",
            )
            source = source.replace(
                '"Run: python Clouds_Coder.py --host 0.0.0.0 --port 8080\\n"',
                '"Run: python -m Code_Structure --host 0.0.0.0 --port 8080\\n"',
            )
        return source

    def runtime_plan(self, nodes: list[NodeInfo]) -> list[dict]:
        return [
            {
                "order": node.order,
                "module": node.target_module,
                "ast_index": node.generated_ast_index,
                "exports": node.bound_names,
                "source_hash": node.source_hash,
                "original_lines": [node.source_start, node.source_end],
            }
            for node in sorted(nodes, key=lambda item: item.order)
        ]

    def generate_runtime(self, nodes: list[NodeInfo]) -> str:
        plan_repr = repr(self.runtime_plan(nodes))
        source_filename = self.analyzer.source_path.name
        template = r'''__HEADER__
from __future__ import annotations

import ast
import builtins
import importlib.machinery
import sys
import threading
import types
from pathlib import Path

_PACKAGE = __package__
_PACKAGE_ROOT = Path(__file__).resolve().parent
# Some legacy source fragments use absolute imports for support packages that
# are relocated inside the generated package (for example
# ``from liquid_kernel import ...``).  The monolith resolves these from its own
# directory; make the generated package root an import location before
# replaying the source statements so the same imports work when the package is
# imported from its parent directory or copied into an isolated environment.
_PACKAGE_ROOT_TEXT = str(_PACKAGE_ROOT)
if _PACKAGE_ROOT_TEXT not in sys.path:
    sys.path.insert(0, _PACKAGE_ROOT_TEXT)
_SOURCE_FILENAME = __SOURCE_FILENAME__
_PLAN = __PLAN__
_LOCK = threading.RLock()
_READY = False
_SHARED = {
    "__builtins__": builtins.__dict__,
    "__name__": _PACKAGE,
    "__package__": "",
    # Relocated code should resolve paths from the generated package, never
    # from the original monolith's directory.  The original filename remains
    # available above only as a provenance label in the runtime plan.
    "__file__": str(_PACKAGE_ROOT / "__main__.py"),
}


def _split_package_source_files():
    """Return the generated package source tree for source-bundle export."""
    rows = []
    for path in sorted(_PACKAGE_ROOT.rglob("*")):
        if not path.is_file() or "__pycache__" in path.parts:
            continue
        if path.suffix not in {".py", ".md", ".json"}:
            continue
        rows.append((path.relative_to(_PACKAGE_ROOT).as_posix(), path.read_bytes()))
    return rows


_SPLIT_PACKAGE_NAME = _PACKAGE
_SPLIT_PACKAGE_ROOT = _PACKAGE_ROOT
_SPLIT_ENTRY_COMMAND = (sys.executable, "-m", _PACKAGE)
_SHARED.update({
    "_SPLIT_PACKAGE_NAME": _SPLIT_PACKAGE_NAME,
    "_SPLIT_PACKAGE_ROOT": _SPLIT_PACKAGE_ROOT,
    "_SPLIT_ENTRY_COMMAND": _SPLIT_ENTRY_COMMAND,
    "_split_package_source_files": _split_package_source_files,
})


class _SplitSourceModule(types.ModuleType):
    """A module view over the shared legacy-global namespace."""

    def __getattribute__(self, name):
        namespace = types.ModuleType.__getattribute__(self, "__dict__")
        exports = namespace.get("_split_export_names", ())
        shared = namespace.get("_split_shared", {})
        if name in exports and name in shared:
            return shared[name]
        return types.ModuleType.__getattribute__(self, name)

    def __setattr__(self, name, value):
        namespace = types.ModuleType.__getattribute__(self, "__dict__")
        if name in namespace.get("_split_export_names", ()):
            namespace.get("_split_shared", {})[name] = value
        types.ModuleType.__setattr__(self, name, value)

    def __dir__(self):
        namespace = types.ModuleType.__getattribute__(self, "__dict__")
        return sorted(set(namespace) | set(namespace.get("_split_export_names", ())))


def _ensure_package(relative: str):
    parent = sys.modules[_PACKAGE]
    if not relative:
        return parent
    current_name = _PACKAGE
    current_path = _PACKAGE_ROOT
    for part in relative.split("/"):
        current_name += "." + part
        current_path /= part
        module = sys.modules.get(current_name)
        if module is None:
            module = types.ModuleType(current_name)
            module.__file__ = str(current_path / "__init__.py")
            module.__package__ = current_name
            module.__path__ = [str(current_path)]
            spec = importlib.machinery.ModuleSpec(current_name, loader=None, is_package=True)
            spec.submodule_search_locations = [str(current_path)]
            module.__spec__ = spec
            sys.modules[current_name] = module
            setattr(parent, part, module)
        parent = module
    return parent


def _ensure_source_module(relative_file: str):
    relative = relative_file[:-3] if relative_file.endswith(".py") else relative_file
    parts = relative.split("/")
    package_rel = "/".join(parts[:-1])
    parent = _ensure_package(package_rel)
    full_name = _PACKAGE + "." + ".".join(parts)
    module = sys.modules.get(full_name)
    if module is None:
        module = _SplitSourceModule(full_name)
        module.__file__ = str(_PACKAGE_ROOT / relative_file)
        module.__package__ = full_name.rpartition(".")[0]
        module.__spec__ = importlib.machinery.ModuleSpec(full_name, loader=None, origin=module.__file__)
        module._split_export_names = set()
        module._split_shared = _SHARED
        sys.modules[full_name] = module
        setattr(parent, parts[-1], module)
    return module


def bootstrap():
    global _READY
    if _READY:
        return _SHARED
    with _LOCK:
        if _READY:
            return _SHARED

        source_modules = {}
        parsed = {}
        for entry in _PLAN:
            relative = entry["module"]
            source_modules[relative] = _ensure_source_module(relative)

        for relative, module in source_modules.items():
            path = _PACKAGE_ROOT / relative
            text = path.read_text(encoding="utf-8")
            parsed[relative] = (path, ast.parse(text, filename=str(path)))

        future_flags = __import__("__future__").annotations.compiler_flag
        for entry in _PLAN:
            relative = entry["module"]
            module = source_modules[relative]
            path, tree = parsed[relative]
            ast_index = int(entry["ast_index"])
            if ast_index < 0 or ast_index >= len(tree.body):
                raise ImportError(
                    f"split plan is stale for {relative}: AST index {ast_index} is unavailable"
                )
            source_node = tree.body[ast_index]
            _SHARED["__name__"] = module.__name__
            try:
                code = compile(
                    ast.Module(body=[source_node], type_ignores=[]),
                    str(path),
                    "exec",
                    flags=future_flags,
                    dont_inherit=True,
                )
                exec(code, _SHARED, _SHARED)
            except BaseException as exc:
                start, end = entry.get("original_lines", [0, 0])
                raise ImportError(
                    f"failed to initialize split statement {entry['order']} "
                    f"from original lines {start}-{end} in {relative}"
                ) from exc
            exports = module._split_export_names
            for name in entry.get("exports", ()):
                exports.add(name)

        _SHARED["__name__"] = _PACKAGE
        for module in source_modules.values():
            names = sorted(module._split_export_names)
            module.__all__ = names
            for name in names:
                if name in _SHARED:
                    types.ModuleType.__setattr__(module, name, _SHARED[name])
        _READY = True
        return _SHARED


def export_into(relative_file: str, namespace: dict):
    """Compatibility helper for unusual direct-loader integrations."""
    bootstrap()
    relative = relative_file[:-3] if relative_file.endswith(".py") else relative_file
    full_name = _PACKAGE + "." + relative.replace("/", ".")
    module = sys.modules[full_name]
    for name in module.__all__:
        namespace[name] = getattr(module, name)
    return namespace
'''
        return (
            template.replace("__HEADER__", GENERATED_HEADER.rstrip())
            .replace("__SOURCE_FILENAME__", repr(source_filename))
            .replace("__PLAN__", plan_repr)
            .rstrip()
            + "\n"
        )

    def generate_root_init(self) -> str:
        return (
            GENERATED_HEADER
            + '"""Architecture-aware split of the original Clouds_Coder application."""\n\n'
            + "from ._runtime import bootstrap as _bootstrap\n\n"
            + "_bootstrap()\n"
            + "del _bootstrap\n"
        )

    def generate_package_init(self, relative_dir: str) -> str:
        dotted = relative_dir.replace("/", ".")
        return GENERATED_HEADER + f'"""{dotted} subsystem."""\n'

    def generate_main_wrapper(self) -> str:
        return (
            GENERATED_HEADER
            + "from .app.main import main\n\n"
            + 'if __name__ == "__main__":\n'
            + "    main()\n"
        )


class FrameworkReportGenerator:
    def __init__(
        self,
        source_path: Path,
        output_dir: Path,
        modules: dict[str, list[NodeInfo]],
        dependency_analyzer: DependencyAnalyzer,
    ) -> None:
        self.source_path = source_path
        self.output_dir = output_dir
        self.modules = modules
        self.dependency_analyzer = dependency_analyzer

    def generate(self) -> str:
        source_modules = sorted(self.modules)
        real_nodes = [node for rows in self.modules.values() for node in rows]
        unclassified = self.modules.get("_unclassified.py", [])
        lines = [
            f"# {self.output_dir.name} Framework",
            "",
            "## Overview",
            "",
            f"- Source snapshot: `{self.source_path.name}` ({len(real_nodes)} top-level statements)",
            f"- Generated source modules: {len(source_modules)}",
            f"- Unclassified statements: {len(unclassified)}",
            "- Execution model: real source fragments initialized in original top-level order",
            "- Runtime dependency on original monolith: none",
            "",
            "The generated modules contain the actual Python source. `_runtime.py` only preserves the original",
            "global initialization order and shared-global semantics required by this legacy monolith's circular",
            "dependency graph; it does not import or read `Clouds_Coder.py`.",
            "",
            "## Package Tree",
            "",
            "```text",
        ]
        paths = set(source_modules) | {"__init__.py", "__main__.py", "_runtime.py"}
        lines.extend(self._tree_lines(sorted(paths)))
        lines.extend([
            "```",
            "",
            "## Module Summary",
            "",
            "| Module | Statements | Exported names | Dependencies | Original line span |",
            "| --- | ---: | ---: | --- | --- |",
        ])
        for module in source_modules:
            nodes = self.modules[module]
            exports = {name for node in nodes for name in node.bound_names}
            deps = self.dependency_analyzer.compute_dependency_map(module, nodes)
            dep_text = ", ".join(f"`{name}`" for name in sorted(deps)) or "—"
            span = f"{min(n.source_start for n in nodes)}–{max(n.source_end for n in nodes)}"
            lines.append(f"| `{module}` | {len(nodes)} | {len(exports)} | {dep_text} | {span} |")
        if unclassified:
            lines.extend(["", "## Unclassified Statements", ""])
            for node in unclassified:
                lines.append(
                    f"- `{node.name}` ({node.kind}, original lines {node.source_start}-{node.source_end})"
                )
        lines.extend(["", "## Source Mapping", ""])
        for module in source_modules:
            lines.extend([f"### `{module}`", ""])
            for node in self.modules[module]:
                exports = ", ".join(f"`{name}`" for name in node.bound_names) or "—"
                lines.append(
                    f"- order {node.order}: `{node.name}` ({node.kind}), lines "
                    f"{node.source_start}-{node.source_end}, exports {exports}"
                )
            lines.append("")
        return "\n".join(lines).rstrip() + "\n"

    def _tree_lines(self, paths: list[str]) -> list[str]:
        tree: dict[str, dict] = {}
        for path in paths:
            cursor = tree
            for part in path.split("/"):
                cursor = cursor.setdefault(part, {})
        out = [f"{self.output_dir.name}/"]

        def render(branch: dict[str, dict], prefix: str = "") -> None:
            items = sorted(branch.items(), key=lambda row: (not bool(row[1]), row[0]))
            for index, (name, children) in enumerate(items):
                last = index == len(items) - 1
                out.append(prefix + ("└── " if last else "├── ") + name)
                if children:
                    render(children, prefix + ("    " if last else "│   "))

        render(tree)
        return out


class FileWriter:
    def __init__(self, output_dir: Path, dry_run: bool = False) -> None:
        self.output_dir = output_dir
        self.dry_run = dry_run
        self.written: list[str] = []
        self.skipped: list[str] = []

    def write(self, relative: str, content: str) -> bool:
        path = self.output_dir / relative
        if self.dry_run:
            print(f"  [DRY-RUN] Would write: {relative} ({len(content):,} bytes)")
            self.written.append(relative)
            return True
        path.parent.mkdir(parents=True, exist_ok=True)
        if path.exists() and path.read_text(encoding="utf-8") == content:
            self.skipped.append(relative)
            return False
        path.write_text(content, encoding="utf-8")
        self.written.append(relative)
        return True


def _generated_file(path: Path) -> bool:
    if not path.is_file():
        return False
    try:
        return path.read_text(encoding="utf-8", errors="ignore").startswith("# Auto-generated by split_coder.py")
    except Exception:
        return False


def _remove_stale_generated_files(output_dir: Path, expected: set[str]) -> list[str]:
    removed: list[str] = []
    if not output_dir.exists():
        return removed
    for path in sorted(output_dir.rglob("*.py"), key=lambda item: len(item.parts), reverse=True):
        if "__pycache__" in path.parts:
            continue
        relative = path.relative_to(output_dir).as_posix()
        if relative not in expected and _generated_file(path):
            path.unlink()
            removed.append(relative)
    for directory in sorted(
        (item for item in output_dir.rglob("*") if item.is_dir()),
        key=lambda item: len(item.parts),
        reverse=True,
    ):
        if directory.name == "__pycache__":
            shutil.rmtree(directory, ignore_errors=True)
            continue
        try:
            directory.rmdir()
        except OSError:
            pass
    return sorted(removed)


def _clear_pycache(root: Path) -> list[str]:
    removed: list[str] = []
    if not root.exists():
        return removed
    for path in sorted(root.rglob("__pycache__"), key=lambda item: len(item.parts), reverse=True):
        if path.is_dir():
            removed.append(path.relative_to(root).as_posix())
            shutil.rmtree(path, ignore_errors=True)
    return removed


def _collect_source_defined_names(path: Path) -> set[str]:
    tree = ast.parse(path.read_text(encoding="utf-8-sig"), filename=str(path))
    names: set[str] = set()
    for node in tree.body:
        if isinstance(node, (ast.Import, ast.ImportFrom)) or (
            isinstance(node, ast.Try) and ArchitectureAnalyzer._is_import_try(node)
        ):
            continue
        names.update(top_level_bound_names(node))
    return names


def run_self_check(
    source_path: Path,
    output_dir: Path,
    *,
    expect_report: bool = True,
    report_name: str = "FRAMEWORK.md",
) -> bool:
    print("\n[SELF-CHECK] Running source-complete split validation...")
    ok = True
    manifest = SplitManifest.load(output_dir)
    if manifest is None:
        print("  [FAIL] missing or incompatible manifest")
        return False
    try:
        source_tree = ast.parse(source_path.read_text(encoding="utf-8-sig"), filename=str(source_path))
        if len(source_tree.body) != len(manifest.nodes):
            ok = False
            print(
                f"  [FAIL] top-level statement coverage: source={len(source_tree.body)} "
                f"manifest={len(manifest.nodes)}"
            )
        else:
            print(f"  [OK] all {len(source_tree.body)} top-level statements are mapped")

        source_names = _collect_source_defined_names(source_path)
        manifest_names = {
            name
            for node in manifest.nodes
            if node.kind != "import"
            for name in node.bound_names
        }
        if source_names != manifest_names:
            ok = False
            missing = sorted(source_names - manifest_names)
            extra = sorted(manifest_names - source_names)
            print(f"  [FAIL] exported-name coverage mismatch: missing={missing[:10]} extra={extra[:10]}")
        else:
            print(f"  [OK] all {len(source_names)} source-defined names are mapped")

        expected_source_modules = sorted({node.target_module for node in manifest.nodes})
        compile_paths = sorted(output_dir.rglob("*.py"))
        compile_errors: list[str] = []
        for path in compile_paths:
            if "__pycache__" in path.parts:
                continue
            try:
                py_compile.compile(str(path), doraise=True)
            except Exception as exc:
                compile_errors.append(f"{path.relative_to(output_dir)}: {exc}")
        if compile_errors:
            ok = False
            print(f"  [FAIL] py_compile errors: {compile_errors[0]}")
        else:
            print(f"  [OK] py_compile passed for {len(compile_paths)} Python files")

        bridge = output_dir / "_source_bridge.py"
        runtime_text = (output_dir / CodeGenerator.RUNTIME_MODULE).read_text(encoding="utf-8")
        if bridge.exists() or "_split_source_symbol" in runtime_text or "spec_from_file_location" in runtime_text:
            ok = False
            print("  [FAIL] legacy monolith source bridge is still present")
        else:
            print("  [OK] generated package has no monolith-loading source bridge")

        missing_real_source: list[str] = []
        for relative in expected_source_modules:
            path = output_dir / relative
            text = path.read_text(encoding="utf-8") if path.exists() else ""
            if not text or "# split-source:" not in text:
                missing_real_source.append(relative)
        if missing_real_source:
            ok = False
            print(f"  [FAIL] modules without real source fragments: {missing_real_source[:8]}")
        else:
            print(f"  [OK] real source is present in all {len(expected_source_modules)} source modules")

        if expect_report and not (output_dir / report_name).is_file():
            ok = False
            print(f"  [FAIL] missing report: {report_name}")
        elif expect_report:
            print(f"  [OK] report exists: {report_name}")

        package_name = output_dir.name
        if not package_name.isidentifier():
            ok = False
            print(f"  [FAIL] output directory is not an importable package name: {package_name!r}")
        else:
            import_code = textwrap.dedent(
                f"""
                import importlib
                import pkgutil
                import sys
                from pathlib import Path

                sys.path.insert(0, {str(output_dir.parent)!r})
                pkg = importlib.import_module({package_name!r})
                manifest = {json.dumps([asdict(node) for node in manifest.nodes], ensure_ascii=False)!r}
                import json
                rows = json.loads(manifest)
                modules = sorted({{row['target_module'] for row in rows}})
                for relative in modules:
                    name = {package_name!r} + '.' + relative[:-3].replace('/', '.')
                    module = importlib.import_module(name)
                    for row in rows:
                        if row['target_module'] != relative or row['kind'] == 'import':
                            continue
                        for symbol in row.get('bound_names', []):
                            if not hasattr(module, symbol):
                                raise SystemExit(f'missing runtime export: {{name}}.{{symbol}}')
                discovered = [m.name for m in pkgutil.walk_packages(pkg.__path__, pkg.__name__ + '.')]
                for name in discovered:
                    importlib.import_module(name)
                main_mod = importlib.import_module({package_name!r} + '.app.main')
                if not callable(main_mod.main):
                    raise SystemExit('app.main.main is not callable')
                print('ok', len(modules), len(discovered))
                """
            ).strip()
            proc = subprocess.run(
                [sys.executable, "-c", import_code],
                cwd=str(output_dir.parent),
                text=True,
                encoding="utf-8",
                errors="replace",
                env={**os.environ, "PYTHONIOENCODING": "utf-8", "PYTHONUTF8": "1"},
                capture_output=True,
                timeout=180,
            )
            if proc.returncode:
                ok = False
                detail = (proc.stderr or proc.stdout).strip().splitlines()
                print("  [FAIL] package import/export validation failed")
                if detail:
                    print(f"         {detail[-1]}")
            else:
                print(f"  [OK] package import/export validation passed ({proc.stdout.strip()})")

            help_proc = subprocess.run(
                [sys.executable, "-m", package_name, "--help"],
                cwd=str(output_dir.parent),
                text=True,
                encoding="utf-8",
                errors="replace",
                env={**os.environ, "PYTHONIOENCODING": "utf-8", "PYTHONUTF8": "1"},
                capture_output=True,
                timeout=180,
            )
            if help_proc.returncode:
                ok = False
                detail = (help_proc.stderr or help_proc.stdout).strip().splitlines()
                print("  [FAIL] generated entry point --help failed")
                if detail:
                    print(f"         {detail[-1]}")
            else:
                first = next((line for line in help_proc.stdout.splitlines() if line.strip()), "help ok")
                print(f"  [OK] generated entry point works ({first.strip()})")

            with tempfile.TemporaryDirectory(prefix="split_coder_standalone_") as temporary:
                isolated_parent = Path(temporary)
                isolated_package = isolated_parent / package_name
                shutil.copytree(
                    output_dir,
                    isolated_package,
                    ignore=shutil.ignore_patterns("__pycache__", "*.pyc"),
                )
                standalone_code = textwrap.dedent(
                    f"""
                    import importlib
                    import sys
                    from pathlib import Path

                    sys.path[:] = [{str(isolated_parent)!r}] + [
                        row for row in sys.path
                        if row and Path(row).resolve() != Path({str(source_path.parent)!r}).resolve()
                    ]
                    pkg = importlib.import_module({package_name!r})
                    main_module = importlib.import_module({package_name!r} + '.app.main')
                    if not callable(main_module.main):
                        raise SystemExit('standalone main export is unavailable')
                    source_bundle = importlib.import_module({package_name!r} + '.app.context').AppContext.source_bundle
                    if not callable(source_bundle):
                        raise SystemExit('standalone source bundle export is unavailable')
                    print('standalone ok')
                    """
                ).strip()
                standalone_proc = subprocess.run(
                    [sys.executable, "-c", standalone_code],
                    cwd=str(isolated_parent),
                    text=True,
                    encoding="utf-8",
                    errors="replace",
                    env={**os.environ, "PYTHONIOENCODING": "utf-8", "PYTHONUTF8": "1"},
                    capture_output=True,
                    timeout=180,
                )
                if standalone_proc.returncode:
                    ok = False
                    detail = (standalone_proc.stderr or standalone_proc.stdout).strip().splitlines()
                    print("  [FAIL] isolated package still depends on the original monolith")
                    if detail:
                        print(f"         {detail[-1]}")
                else:
                    print("  [OK] isolated package imports without the original monolith")
    finally:
        removed = _clear_pycache(output_dir)
        print(f"  [CLEAN] removed {len(removed)} __pycache__ directories")
    print(f"[SELF-CHECK] {'PASSED' if ok else 'FAILED'}")
    return ok


class Splitter:
    def __init__(
        self,
        source_path: Path,
        output_dir: Path,
        layout: dict[str, list[str]],
        *,
        dry_run: bool = False,
        update_mode: bool = False,
        show_tree: bool = False,
        dump_layout: bool = False,
        report_name: str = "FRAMEWORK.md",
        generate_report: bool = True,
    ) -> None:
        self.source_path = source_path
        self.output_dir = output_dir
        self.layout = layout
        self.dry_run = dry_run
        self.update_mode = update_mode
        self.show_tree = show_tree
        self.dump_layout = dump_layout
        self.report_name = report_name
        self.generate_report = generate_report

    def run(self) -> None:
        print("\n" + "─" * 68)
        print(f"  split_coder.py v{GENERATOR_VERSION} — source-complete split of {self.source_path.name}")
        print(f"  Output: {self.output_dir}")
        print("─" * 68 + "\n")
        print("[1/7] Analyzing every top-level source statement...")
        analyzer = ArchitectureAnalyzer(self.source_path)
        nodes = analyzer.analyze()
        print(f"  Found {len(nodes)} top-level statements")

        old_manifest = SplitManifest.load(self.output_dir) if self.update_mode else None
        if old_manifest:
            print(f"[2/7] Loaded v{old_manifest.generator_version} manifest ({len(old_manifest.nodes)} statements)")
        else:
            print("[2/7] Building a fresh source map")

        print("[3/7] Routing statements by application architecture...")
        router = ModuleRouter(self.layout)
        router.assign_all(nodes)
        modules: dict[str, list[NodeInfo]] = defaultdict(list)
        for node in nodes:
            modules[node.target_module].append(node)
        unclassified = modules.get("_unclassified.py", [])
        print(f"  Routed into {len(modules)} source modules; unclassified={len(unclassified)}")
        if unclassified:
            print("  Unclassified sample: " + ", ".join(node.name for node in unclassified[:10]))
            print("  [WARNING] These statements are preserved in _unclassified.py; add ownership rules to complete the architecture map.")
        if self.dump_layout:
            print(json.dumps(
                {module: [node.name for node in rows] for module, rows in sorted(modules.items())},
                indent=2,
                ensure_ascii=False,
            ))

        print("[4/7] Computing the cross-module dependency map...")
        dependencies = DependencyAnalyzer(analyzer, nodes)
        generator = CodeGenerator(analyzer, self.output_dir.name)
        writer = FileWriter(self.output_dir, self.dry_run)

        print("[5/7] Writing real source modules and ordered runtime...")
        expected: set[str] = {"__init__.py", "__main__.py", CodeGenerator.RUNTIME_MODULE}
        for module, rows in sorted(modules.items()):
            writer.write(module, generator.generate_source_module(module, rows))
            expected.add(module)
        writer.write(CodeGenerator.RUNTIME_MODULE, generator.generate_runtime(nodes))
        writer.write("__init__.py", generator.generate_root_init())
        writer.write("__main__.py", generator.generate_main_wrapper())

        package_dirs = {
            "/".join(module.split("/")[:-1])
            for module in modules
            if "/" in module
        }
        for directory in sorted(package_dirs):
            relative = f"{directory}/__init__.py"
            writer.write(relative, generator.generate_package_init(directory))
            expected.add(relative)

        print("[6/7] Writing framework report..." if self.generate_report else "[6/7] Framework report disabled")
        if self.generate_report:
            report = FrameworkReportGenerator(
                self.source_path,
                self.output_dir,
                modules,
                dependencies,
            ).generate()
            writer.write(self.report_name, report)

        print("[7/7] Saving manifest and removing stale generated modules...")
        manifest = SplitManifest(
            source_path=str(self.source_path),
            source_hash=hashlib.sha256(self.source_path.read_bytes()).hexdigest()[:16],
            output_dir=str(self.output_dir),
            package_name=self.output_dir.name,
            source_statement_count=len(nodes),
            nodes=[
                ManifestEntry(
                    order=node.order,
                    name=node.name,
                    kind=node.kind,
                    lineno=node.lineno,
                    end_lineno=node.end_lineno,
                    source_start=node.source_start,
                    source_end=node.source_end,
                    source_hash=node.source_hash,
                    target_module=node.target_module,
                    bound_names=node.bound_names,
                    generated_ast_index=node.generated_ast_index,
                )
                for node in nodes
            ],
        )
        removed: list[str] = []
        if not self.dry_run:
            manifest.save(self.output_dir)
            removed = _remove_stale_generated_files(self.output_dir, expected)

        print("\n" + "─" * 68)
        print(f"  Written : {len(writer.written)} files")
        print(f"  Unchanged: {len(writer.skipped)} files")
        print(f"  Removed : {len(removed)} stale files")
        print(f"  Output  : {self.output_dir}")
        print("─" * 68)
        if self.show_tree and not self.dry_run:
            self._print_tree(self.output_dir)

    @staticmethod
    def _print_tree(root: Path) -> None:
        print(f"\n{root.name}/")

        def walk(directory: Path, prefix: str = "") -> None:
            items = sorted(directory.iterdir(), key=lambda item: (item.is_file(), item.name))
            for index, item in enumerate(items):
                last = index == len(items) - 1
                print(prefix + ("└── " if last else "├── ") + item.name)
                if item.is_dir():
                    walk(item, prefix + ("    " if last else "│   "))

        walk(root)


def build_arg_parser() -> argparse.ArgumentParser:
    parser = argparse.ArgumentParser(
        description="Split a monolithic Python application into a source-complete executable package.",
        formatter_class=argparse.RawDescriptionHelpFormatter,
        epilog=textwrap.dedent(
            """\
            Examples:
              python split_coder.py Clouds_Coder.py
              python split_coder.py Clouds_Coder.py -o Code_Structure --self-check
              python split_coder.py Clouds_Coder.py --dry-run --dump-layout
              python split_coder.py Clouds_Coder.py --auto-layout
            """
        ),
    )
    parser.add_argument("source", help="Monolithic Python source file")
    parser.add_argument("--output-dir", "-o", help="Output package directory")
    parser.add_argument("--dry-run", "-n", action="store_true", help="Analyze and preview without writing")
    parser.add_argument("--update", "-u", action="store_true", help="Regenerate only files whose content changed")
    parser.add_argument("--show-tree", action="store_true", help="Print the generated tree")
    parser.add_argument("--dump-layout", action="store_true", help="Print routed statement names as JSON")
    parser.add_argument("--auto-layout", action="store_true", help="Use generic architecture heuristics")
    parser.add_argument("--layout-file", help="Custom JSON module-to-symbol layout")
    parser.add_argument("--report-name", default="FRAMEWORK.md", help="Framework report filename")
    parser.add_argument("--no-report", action="store_true", help="Do not generate the framework report")
    parser.add_argument("--self-check", action="store_true", help="Compile and import-test the generated package")
    return parser


def load_layout(args: argparse.Namespace, source_path: Path) -> dict[str, list[str]]:
    if args.layout_file:
        path = Path(args.layout_file).expanduser().resolve()
        if not path.is_file():
            raise SystemExit(f"Layout file not found: {path}")
        data = json.loads(path.read_text(encoding="utf-8"))
        if not isinstance(data, dict):
            raise SystemExit("Layout JSON must be an object mapping module paths to symbol lists")
        return {str(key): [str(value) for value in values] for key, values in data.items()}
    if args.auto_layout:
        analyzer = ArchitectureAnalyzer(source_path)
        nodes = analyzer.analyze()
        return AutoLayoutGenerator().generate(nodes)
    return DEFAULT_LAYOUT


def main() -> int:
    parser = build_arg_parser()
    args = parser.parse_args()
    source_path = Path(args.source).expanduser().resolve()
    if not source_path.is_file():
        print(f"Error: source file not found: {source_path}", file=sys.stderr)
        return 1
    output_dir = (
        Path(args.output_dir).expanduser().resolve()
        if args.output_dir
        else source_path.parent / "Code_Structure"
    )
    if args.self_check and args.dry_run:
        print("Error: --self-check cannot be combined with --dry-run", file=sys.stderr)
        return 2
    if output_dir == source_path.parent:
        print("Error: output directory must not be the source directory", file=sys.stderr)
        return 2
    layout = load_layout(args, source_path)
    splitter = Splitter(
        source_path,
        output_dir,
        layout,
        dry_run=args.dry_run,
        update_mode=args.update,
        show_tree=args.show_tree,
        dump_layout=args.dump_layout,
        report_name=str(args.report_name or "FRAMEWORK.md"),
        generate_report=not args.no_report,
    )
    splitter.run()
    if args.self_check:
        return 0 if run_self_check(
            source_path,
            output_dir,
            expect_report=not args.no_report,
            report_name=args.report_name,
        ) else 2
    return 0


if __name__ == "__main__":
    raise SystemExit(main())