"""Central configuration: env, story styles, difficulty levels, file filters, palette.

StoryCode reads a non-coder's project and explains it as a story. The values here
are deliberately plain-language: labels a non-technical person sees should never
contain jargon unless we explain it.
"""
from __future__ import annotations

import os
from dataclasses import dataclass

from dotenv import load_dotenv

load_dotenv()

# --- Runtime / services -----------------------------------------------------
# The model runs on Modal (vLLM, OpenAI-compatible API). The Space is CPU-only.
MODAL_ENDPOINT_URL = os.getenv("MODAL_ENDPOINT_URL", "").rstrip("/")
MODAL_API_KEY = os.getenv("MODAL_API_KEY", "")
MODEL_ID = os.getenv("MODEL_ID", "openbmb/MiniCPM4.1-8B")
DB_PATH = os.getenv("STORYCODE_DB_PATH", "data/storycode.db")
USE_ZEROGPU_FALLBACK = os.getenv("USE_ZEROGPU_FALLBACK", "0") == "1"

# Generation limits (keep the Space responsive; map-reduce stays well under 32k ctx).
LLM_TIMEOUT_S = int(os.getenv("LLM_TIMEOUT_S", "90"))
MAX_FILES = int(os.getenv("STORYCODE_MAX_FILES", "60"))      # analysed per project
MAX_FILE_BYTES = int(os.getenv("STORYCODE_MAX_FILE_BYTES", str(200_000)))
MAX_TOTAL_BYTES = int(os.getenv("STORYCODE_MAX_TOTAL_BYTES", str(5_000_000)))

# --- Languages we understand (stated honestly in the UI) --------------------
# Extension -> language key. Anything else is listed but not deeply parsed.
LANG_BY_EXT: dict[str, str] = {
    ".py": "python",
    ".js": "javascript", ".jsx": "javascript",
    ".ts": "typescript", ".tsx": "typescript", ".mjs": "javascript", ".cjs": "javascript",
    ".html": "html", ".htm": "html",
    ".css": "css", ".scss": "css",
    ".json": "json",
    ".yaml": "yaml", ".yml": "yaml",
    ".md": "markdown",
}

# Directories and files we never analyse (noise / not the user's own code).
IGNORE_DIRS: frozenset[str] = frozenset({
    "node_modules", ".git", ".svn", "__pycache__", ".venv", "venv", "env",
    "dist", "build", ".next", ".nuxt", "out", "target", ".idea", ".vscode",
    "coverage", ".pytest_cache", ".mypy_cache", "site-packages", ".cache",
    "vendor", "bower_components", ".gradio",
})
IGNORE_FILES: frozenset[str] = frozenset({
    "package-lock.json", "yarn.lock", "pnpm-lock.yaml", "poetry.lock",
    ".ds_store", "thumbs.db",
})
# Binary-ish extensions to skip outright.
BINARY_EXTS: frozenset[str] = frozenset({
    ".png", ".jpg", ".jpeg", ".gif", ".webp", ".ico", ".svg", ".pdf", ".zip",
    ".gz", ".tar", ".mp3", ".mp4", ".wav", ".woff", ".woff2", ".ttf", ".otf",
    ".pyc", ".so", ".dll", ".exe", ".bin", ".db", ".sqlite", ".lock",
})

# Manifest files the dependency explainer reads.
DEP_MANIFESTS: frozenset[str] = frozenset({
    "requirements.txt", "package.json", "pyproject.toml", "pipfile",
})


# --- Roles -------------------------------------------------------------------
# A file's "role" drives the architecture-map grouping and the safe-to-edit
# verdict. These are plain-English buckets, not technical layers.
@dataclass(frozen=True)
class Role:
    key: str
    label: str
    color: str  # used in the diagram + chips


ROLES: tuple[Role, ...] = (
    Role("entry", "Front door", "#4b5bd6"),       # what runs first / users hit
    Role("frontend", "What you see", "#3a7ca5"),  # UI / pages / styles
    Role("backend", "The engine", "#6b4fb0"),     # logic / services / API calls
    Role("data", "Memory", "#3f9b6b"),            # db / storage / models / embeddings
    Role("config", "Settings", "#8a7a66"),        # config / env / manifests
    Role("test", "Safety checks", "#9aa0a6"),     # tests
    Role("util", "Helpers", "#b0793a"),           # shared utilities
    Role("other", "Other", "#9aa0a6"),
)
ROLE_KEYS: tuple[str, ...] = tuple(r.key for r in ROLES)
ROLE_LABEL = {r.key: r.label for r in ROLES}
ROLE_COLOR = {r.key: r.color for r in ROLES}

# Filename / path hints used by analyzer.graph.classify_role (first match wins).
ENTRY_NAMES: frozenset[str] = frozenset({
    "app.py", "main.py", "manage.py", "server.py", "run.py", "wsgi.py", "asgi.py",
    "index.js", "index.ts", "index.jsx", "index.tsx", "main.js", "main.ts",
    "index.html", "app.js", "app.ts",
    "train.py", "play.py", "demo.py", "infer.py", "evaluate.py", "test.py",
})
CONFIG_HINTS: tuple[str, ...] = (
    "config", "settings", ".env", "requirements", "package.json", "pyproject",
    "dockerfile", "makefile", ".yaml", ".yml", ".toml", ".ini",
)
DATA_HINTS: tuple[str, ...] = (
    "db", "database", "model", "schema", "store", "embed", "vector",
    "index", "ingest", "migration", "seed",
)
TEST_HINTS: tuple[str, ...] = ("test", "spec", "__tests__")
UTIL_HINTS: tuple[str, ...] = ("util", "helper", "common", "lib", "shared", "tools")


# --- Story styles ------------------------------------------------------------
# Simple Walkthrough is the DEFAULT (per the brief: creative styles can feel
# patronising; they're opt-in). Each carries a short voice instruction handed to
# the model, but the model only ever narrates facts we extracted.
@dataclass(frozen=True)
class Style:
    key: str
    label: str
    emoji: str
    voice: str  # instruction injected into the narration prompt


STYLES: tuple[Style, ...] = (
    Style("plain", "Simple Walkthrough", "📋",
          "Explain plainly and warmly, step by step, no metaphors, no jargon. "
          "Like a patient friend who read the whole project for them."),
    Style("kids", "Kids Book", "🧸",
          "Tell it like a children's picture book. Each file is a friendly "
          "character with a job. Keep sentences short and gentle."),
    Style("thriller", "Thriller", "🕵️",
          "Tell it like a suspense thriller — the request arrives, tension "
          "builds as it moves through the code — but every claim must stay true."),
    Style("news", "News Report", "📰",
          "Report it like a breaking-news anchor: who did what, in what order, "
          "with punchy headlines. Accurate, never sensational about facts."),
    Style("recipe", "Recipe", "🍳",
          "Explain it like a cooking recipe: ingredients (the pieces) and steps "
          "(the flow), in order."),
)
STYLE_KEYS: tuple[str, ...] = tuple(s.key for s in STYLES)
STYLE_BY_KEY = {s.key: s for s in STYLES}
DEFAULT_STYLE = "plain"


@dataclass(frozen=True)
class Difficulty:
    key: str
    label: str
    voice: str


DIFFICULTIES: tuple[Difficulty, ...] = (
    Difficulty("kid", "Explain like I'm 5",
               "Use no technical terms at all. Everyday words a child knows."),
    Difficulty("teen", "Teenager",
               "Light technical terms are fine if you explain each one in the "
               "same sentence."),
    Difficulty("adult", "Adult / curious beginner",
               "Full clarity. Technical terms are okay but define them the first "
               "time they appear."),
)
DIFFICULTY_KEYS: tuple[str, ...] = tuple(d.key for d in DIFFICULTIES)
DIFFICULTY_BY_KEY = {d.key: d for d in DIFFICULTIES}
DEFAULT_DIFFICULTY = "teen"


# --- Where did your code come from? (onboarding) ----------------------------
SOURCE_HINTS: tuple[str, ...] = (
    "Claude / Claude Code", "Cursor", "ChatGPT", "Bolt / v0 / Lovable",
    "A ZIP someone sent me", "Not sure",
)


# --- Visual palette (mirrors ui/styles.css; kept here for diagram + chips) ---
PALETTE = {
    "paper": "#f7f4ef",
    "ink": "#1d2430",
    "accent": "#4b5bd6",
    "safe": "#3f9b6b",
    "careful": "#d8932f",
    "danger": "#c2503f",
    "muted": "#6b7280",
    "line": "#e7e3da",
}

# Safe-to-edit verdicts (computed in analyzer.graph, never by the model).
SAFE = "safe"
CAREFUL = "careful"
DANGER = "danger"
SAFETY_COLOR = {SAFE: PALETTE["safe"], CAREFUL: PALETTE["careful"], DANGER: PALETTE["danger"]}
SAFETY_EMOJI = {SAFE: "🟢", CAREFUL: "🟠", DANGER: "🔴"}
SAFETY_LABEL = {SAFE: "Safe to change", CAREFUL: "Change carefully", DANGER: "Don't touch"}