#!/usr/bin/env python3
"""Aday satırı biçimleri (m2w_filter.py çıktısı üstünde, hızlı CPU işi).

  long : eski biçimin uzun sürümü  -> etiket | role | type | öznitelikler (<=120) | iç metin (<=200), toplam <=400 karakter
  tree : erişilebilirlik-ağacı tarzı -> rol "isim" [durum]

Kullanım:  python m2w_format.py --data m2w_v3 --fmt tree --out m2w_v3_tree
"""
import argparse, gzip, json, os
from collections import Counter

LABEL_ATTRS = ("aria_label", "placeholder", "title", "name", "alt", "value", "id")
ROLE_OK = {"button", "link", "checkbox", "radio", "textbox", "searchbox", "combobox", "menuitem", "tab", "option", "switch",
           "listbox", "slider", "spinbutton", "menuitemcheckbox", "menuitemradio", "treeitem", "heading", "img", "listitem",
           "cell", "row", "columnheader", "menu", "navigation", "dialog"}
INPUT_ROLE = {"checkbox": "checkbox", "radio": "radio", "submit": "button", "button": "button", "reset": "button",
              "image": "button", "search": "searchbox", "range": "slider", "number": "spinbutton"}
TAG_ROLE = {"a": "link", "button": "button", "select": "combobox", "textarea": "textbox", "option": "option",
            "label": "label", "summary": "button", "li": "listitem", "img": "img", "svg": "img", "i": "img",
            "td": "cell", "tr": "row", "th": "columnheader", "h1": "heading", "h2": "heading", "h3": "heading",
            "h4": "heading", "p": "paragraph", "span": "generic", "div": "generic"}


def g(at, *keys):
    for k in keys:
        v = at.get(k)
        if v is not None and str(v).strip():
            return str(v).strip()
    return ""


def fmt_long(c):
    at = c.get("at", {})
    parts = [c["tag"]]
    for k in ("role", "type"):
        if at.get(k):
            parts.append(str(at[k]))
    for k in LABEL_ATTRS:
        v = g(at, k, k.replace("_", "-"))
        if v:
            parts.append(v[:120])
    if c.get("inner"):
        parts.append(c["inner"][:200])
    return " | ".join(parts)[:400]


def role_of(c):
    at, tag = c.get("at", {}), c["tag"]
    r = g(at, "role").lower()
    if r in ROLE_OK:
        return r
    if tag == "input":
        t = g(at, "type").lower()
        return INPUT_ROLE.get(t, "textbox")
    return TAG_ROLE.get(tag, tag)


def is_true(at, *keys):
    for k in keys:
        if k in at:
            return str(at[k]).strip().lower() not in ("false", "0", "no")
    return False


def fmt_tree(c):
    at, role = c.get("at", {}), role_of(c)
    inner = " ".join(c.get("inner", "").split())
    if role in ("textbox", "searchbox", "combobox", "spinbutton"):
        order = [g(at, "aria_label", "aria-label"), g(at, "placeholder"), g(at, "title"), g(at, "name"), inner, g(at, "id")]
    elif role in ("img",):
        order = [g(at, "alt"), g(at, "aria_label", "aria-label"), g(at, "title"), inner, g(at, "name")]
    elif role in ("button", "link", "menuitem", "tab", "option", "checkbox", "radio", "switch", "label", "listitem", "heading",
                  "cell", "row", "columnheader", "paragraph", "generic", "treeitem"):
        order = [g(at, "aria_label", "aria-label"), inner, g(at, "value"), g(at, "title"), g(at, "alt"), g(at, "placeholder"),
                 g(at, "name"), g(at, "id")]
    else:
        order = [g(at, "aria_label", "aria-label"), inner, g(at, "title"), g(at, "alt"), g(at, "value"), g(at, "name"), g(at, "id")]
    name = next((x for x in order if x), "")[:100]
    st = []
    if is_true(at, "checked", "aria_checked", "aria-checked"):
        st.append("checked")
    if is_true(at, "selected", "aria_selected", "aria-selected"):
        st.append("selected")
    if is_true(at, "disabled", "aria_disabled", "aria-disabled"):
        st.append("disabled")
    for k in ("aria_expanded", "aria-expanded"):
        if k in at:
            st.append(f"expanded={str(at[k]).strip().lower()}")
            break
    if is_true(at, "required"):
        st.append("required")
    s = role + (f' "{name}"' if name else "")
    return s + "".join(f" [{x}]" for x in st)


FMT = {"long": fmt_long, "tree": fmt_tree}


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--data", required=True); ap.add_argument("--out", required=True)
    ap.add_argument("--fmt", required=True, choices=list(FMT))
    ap.add_argument("--splits", nargs="+", default=["train", "val", "test_task", "test_website", "test_domain"])
    a = ap.parse_args(); os.makedirs(a.out, exist_ok=True); f = FMT[a.fmt]
    for s in a.splits:
        p = os.path.join(a.data, f"{s}.jsonl.gz")
        if not os.path.exists(p):
            continue
        n = 0; lens = []; roles = Counter()
        with gzip.open(p, "rt", encoding="utf-8") as fi, gzip.open(os.path.join(a.out, f"{s}.jsonl.gz"), "wt", encoding="utf-8") as fo:
            for l in fi:
                ex = json.loads(l)
                for c in ex["cands"] + ex.get("dropped_pos", []):
                    c["text"] = f(c)
                lens += [len(c["text"]) for c in ex["cands"]]
                if n < 300:
                    roles.update(c["text"].split(" ")[0] for c in ex["cands"])
                fo.write(json.dumps(ex, ensure_ascii=False) + "\n"); n += 1
        print(f"  {s}: {n} adım | aday metni uzunluğu (karakter) ortanca {sorted(lens)[len(lens)//2]}, p90 {sorted(lens)[int(len(lens)*.9)]}")
    print("örnek roller (ilk 300 adım):", roles.most_common(12))


if __name__ == "__main__":
    main()