#!/usr/bin/env python3
"""扫描 BOSS直聘"牛人"候选人列表(推荐/精选/最新)，用 uiautomator dump 读屏幕上已经
渲染出来的候选人卡片文字，落地到本地 SQLite。不调用任何 App 私有网络接口——只是
模拟人手动划列表的操作(adb input swipe) + 读取当前屏幕控件文本(adb uiautomator dump)。

用法:
  python3 scan_geek_list.py                  # 扫描所有已发现的岗位 tab x 推荐/精选/最新
  python3 scan_geek_list.py --job 产品助理 --tab 精选
  python3 scan_geek_list.py --max-swipes 30  # 每个 tab 最多滑动多少次(防止卡死)
"""
import argparse
import hashlib
import re
import sqlite3
import subprocess
import time
import xml.etree.ElementTree as ET
from pathlib import Path

ADB_BIN = "adb"  # 假设已在 PATH 里(见 pull_dbs.sh 里加过 platform-tools)
DUMP_REMOTE = "/sdcard/window_dump.xml"
DUMP_LOCAL = Path(__file__).parent / "data" / "_window_dump.xml"
DB_PATH = Path(__file__).parent / "data" / "geek_list.db"

BOTTOM_NAV_TEXTS = {"牛人", "搜索", "消息", "我的"}
TOPBAR_TEXTS = {"推荐", "精选", "最新", "筛选"}
META_RE = re.compile(r"(应届生|经验|在校|\d+年).*[|｜].*(K|面议|元)")
EDU_RE = re.compile(r".+ · .+")


def adb(*args, timeout=15):
    return subprocess.run([ADB_BIN] + list(args), capture_output=True, text=True, timeout=timeout)


def dump_screen_texts():
    adb("shell", "uiautomator", "dump", DUMP_REMOTE)
    DUMP_LOCAL.parent.mkdir(parents=True, exist_ok=True)
    adb("pull", DUMP_REMOTE, str(DUMP_LOCAL))
    root = ET.parse(str(DUMP_LOCAL)).getroot()
    out = []
    for node in root.iter("node"):
        t = node.attrib.get("text", "").strip()
        if t:
            out.append((t, node.attrib.get("bounds", "")))
    return out


def find_bounds(texts_with_bounds, label):
    for t, b in texts_with_bounds:
        if t == label:
            m = re.match(r"\[(\d+),(\d+)\]\[(\d+),(\d+)\]", b)
            if m:
                x1, y1, x2, y2 = map(int, m.groups())
                return (x1 + x2) // 2, (y1 + y2) // 2
    return None


def tap(x, y):
    adb("shell", "input", "tap", str(x), str(y))
    time.sleep(1)


def swipe_up():
    # 从屏幕中下部往上划一屏，留出顶部 tab 栏和底部导航栏不划到
    adb("shell", "input", "swipe", "1000", "1700", "1000", "700", "400")
    time.sleep(1.3)


def parse_cards(texts):
    """把一屏候选人卡片的文本节点，按已知卡片结构切成一张张候选人记录。"""
    texts = [t for t in texts if t not in BOTTOM_NAV_TEXTS and t not in TOPBAR_TEXTS and t not in ("北京",)]
    cards = []
    i = 0
    n = len(texts)
    while i < n:
        # 找 meta 行(如 "26年应届生  |  本科  |  6-9K")作为卡片锚点
        if META_RE.search(texts[i]):
            name = texts[i - 1] if i - 1 >= 0 else ""
            badge = ""
            # name 前一格可能是徽章("在线"/"刚刚活跃")而不是姓名，姓名再往前一格
            if name in ("在线", "刚刚活跃", "今日活跃") and i - 2 >= 0:
                badge = name
                name = texts[i - 2]
            meta = texts[i]
            j = i + 1
            focus_line = ""
            education = ""
            tags = []
            description = ""
            if j < n and (texts[j].startswith("求职期望") or texts[j].startswith("最近关注")):
                focus_line = texts[j]
                j += 1
            if j < n and EDU_RE.match(texts[j]) and len(texts[j]) < 40:
                education = texts[j]
                j += 1
            # 标签: 连续的短文本(<12 字符)，直到遇到长段描述或下一张卡片的 meta 行
            while j < n and len(texts[j]) < 12 and not META_RE.search(texts[j]):
                tags.append(texts[j])
                j += 1
            if j < n and not META_RE.search(texts[j]):
                description = texts[j]
                j += 1
            cards.append(
                {
                    "name": name,
                    "badge": badge,
                    "meta": meta,
                    "focus_line": focus_line,
                    "education": education,
                    "tags": "、".join(tags),
                    "description": description,
                }
            )
            i = j
        else:
            i += 1
    return cards


def ensure_db():
    DB_PATH.parent.mkdir(parents=True, exist_ok=True)
    con = sqlite3.connect(str(DB_PATH))
    con.execute(
        """CREATE TABLE IF NOT EXISTS candidates (
            id INTEGER PRIMARY KEY AUTOINCREMENT,
            job_tab TEXT,
            sub_tab TEXT,
            name TEXT,
            badge TEXT,
            meta TEXT,
            focus_line TEXT,
            education TEXT,
            tags TEXT,
            description TEXT,
            dedup_key TEXT UNIQUE,
            captured_at INTEGER
        )"""
    )
    con.commit()
    return con


def dedup_key(job_tab, sub_tab, card):
    # 只用姓名+meta做键：同一张卡片在相邻两屏重叠截取时，education/tags 是否被
    # 完整读到可能不一致，用它们做键会把同一个人拆成两条。
    raw = f"{job_tab}|{sub_tab}|{card['name']}|{card['meta']}"
    return hashlib.sha1(raw.encode("utf-8")).hexdigest()


def scan_current_tab(con, job_tab, sub_tab, max_swipes=25):
    seen_keys = set()
    new_total = 0
    stale_rounds = 0
    for _ in range(max_swipes):
        texts = [t for t, _ in dump_screen_texts()]
        cards = parse_cards(texts)
        added_this_round = 0
        for c in cards:
            key = dedup_key(job_tab, sub_tab, c)
            if key in seen_keys:
                continue
            seen_keys.add(key)
            row = con.execute(
                "SELECT id, length(description) FROM candidates WHERE dedup_key = ?", (key,)
            ).fetchone()
            if row is None:
                con.execute(
                    "INSERT INTO candidates (job_tab, sub_tab, name, badge, meta, focus_line, education, tags, description, dedup_key, captured_at) "
                    "VALUES (?,?,?,?,?,?,?,?,?,?,strftime('%s','now'))",
                    (
                        job_tab,
                        sub_tab,
                        c["name"],
                        c["badge"],
                        c["meta"],
                        c["focus_line"],
                        c["education"],
                        c["tags"],
                        c["description"],
                        key,
                    ),
                )
                added_this_round += 1
                new_total += 1
            elif len(c["description"]) > (row[1] or 0):
                # 同一个人前后两屏都拍到了，取信息更完整(描述更长)的那次覆盖
                con.execute(
                    "UPDATE candidates SET badge=?, focus_line=?, education=?, tags=?, description=? WHERE id=?",
                    (c["badge"], c["focus_line"], c["education"], c["tags"], c["description"], row[0]),
                )
        con.commit()
        if added_this_round == 0:
            stale_rounds += 1
            if stale_rounds >= 2:
                break
        else:
            stale_rounds = 0
        swipe_up()
    print(f"  [{job_tab} / {sub_tab}] 新增 {new_total} 条")
    return new_total


def switch_job_and_tab(job_label, sub_tab_label):
    texts = dump_screen_texts()
    pos = find_bounds(texts, job_label)
    if pos:
        tap(*pos)
    texts = dump_screen_texts()
    pos = find_bounds(texts, sub_tab_label)
    if pos:
        tap(*pos)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--job", action="append", help="职位 tab 名(可重复)，不填则自动探测顶部两个职位 tab")
    ap.add_argument("--tab", action="append", default=None, help="子 tab: 推荐/精选/最新，不填则三个都扫")
    ap.add_argument("--max-swipes", type=int, default=25)
    args = ap.parse_args()

    sub_tabs = args.tab or ["推荐", "精选", "最新"]

    con = ensure_db()

    jobs = args.job
    if not jobs:
        # 从当前屏幕顶部读职位 tab 名(比如 "产品助理" / "Prompt Engin...")
        texts = [t for t, _ in dump_screen_texts()]
        jobs = [t for t in texts[:2] if t not in BOTTOM_NAV_TEXTS and t not in TOPBAR_TEXTS]
        print("自动探测到职位 tab:", jobs)

    for job in jobs:
        for sub_tab in sub_tabs:
            print(f"=== 切换到 {job} / {sub_tab} ===")
            switch_job_and_tab(job, sub_tab)
            scan_current_tab(con, job, sub_tab, max_swipes=args.max_swipes)

    total = con.execute("SELECT count(*) FROM candidates").fetchone()[0]
    print(f"完成，库里累计 {total} 条候选人记录 -> {DB_PATH}")
    con.close()


if __name__ == "__main__":
    main()