# -*- coding: utf-8 -*-
"""
编码模板反推 v2 —— 支持零填充与空值占位符。

v1 的 explain() 贪心匹配"非零填充", 把 ULP50040AEK0 的 "040" 拆成 字面量"0" + 值"40",
产生大量假阴性。真实规则是:
  - 数值参数常带前导零 (stroke=40 -> "040")
  - 枚举参数为空时, 编码里用 '_' 占位 (UAGP075_K045WS0)

本脚本对每个参数尝试 {原值, 各前导零位数, '_'占位} 的所有组合, 取能对齐的;
若某系列的"字面量序列"完全一致 → 可模板化。

用法: PYTHONUTF8=1 python infer-sku-template.py
"""
import json, io, collections, sys

CAT = r'D:/开发/OnebotCatalog/release/web_2026.08/catalog.json'
d = json.load(io.open(CAT, encoding='utf-8-sig'))

MAXPAD = 4


def align_all(code, params, values):
    """返回全部可能的字面量序列 (元组)。每个参数尝试: 原值 / 前导零 / '_'占位。"""
    results = []

    def dfs(i, pos, lits):
        if i == len(params):
            results.append(tuple(lits + [code[pos:]]))
            return
        pc = params[i]
        v = str(values.get(pc, ''))
        cands = []
        if v == '':
            cands = ['_', '']            # 空值: 占位符或什么都不写
        else:
            cands = [v] + ['0' * z + v for z in range(1, MAXPAD + 1)]
        seen = set()
        for pat in cands:
            if pat in seen:
                continue
            seen.add(pat)
            if pat == '':
                dfs(i + 1, pos, lits + [''])
                continue
            idx = code.find(pat, pos)
            if idx >= 0:
                dfs(i + 1, idx + len(pat), lits + [code[pos:idx]])
    dfs(0, 0, [])
    return results


def render_template(params, lits):
    """字面量序列 -> 模板串。空值参数渲染为 '_'。"""
    out = ''
    for i, pc in enumerate(params):
        out += lits[i] + '{' + pc + '}'
    out += lits[len(params)]
    return out


good, bad = [], []
for s in d['series']:
    vs = s.get('variants') or []
    if not vs:
        continue
    params = [p['code'] for p in s['parameters']]
    lit_sets = []
    unaligned = 0
    for v in vs:
        alts = align_all(v['modelCode'], params, v['params'])
        if not alts:
            unaligned += 1
            continue
        lit_sets.append(alts)

    if not lit_sets:
        bad.append((s['code'], len(vs), '完全无法对齐', [], 0))
        continue

    # 统计"出现在最多变体里"的字面量序列 (每个变体取任一可对齐解)
    cnt = collections.Counter()
    for alts in lit_sets:
        for a in set(alts):
            cnt[a] += 1
    top, hit = cnt.most_common(1)[0]
    rate = hit / len(vs) * 100

    if rate >= 99.5 and unaligned == 0:
        good.append((s['code'], len(vs), rate, render_template(params, top), len(cnt)))
    else:
        bad.append((s['code'], len(vs), f'{rate:.1f}% ({len(cnt)} 种)', cnt.most_common(3), unaligned))

print('=' * 96)
print(f'可模板化的系列 ({len(good)} 个)')
print('=' * 96)
for code, n, rate, tpl, k in sorted(good):
    print(f'{code:7s} {n:6d} 变体 {rate:6.1f}%  {tpl}')

print()
print('=' * 96)
print(f'不可模板化的系列 ({len(bad)} 个)')
print('=' * 96)
for code, n, why, tops, un in sorted(bad, key=lambda x: -x[1]):
    print(f'{code:7s} {n:6d} 变体 {why}' + (f'  未对齐 {un}' if un else ''))
    for lits, c in tops[:3]:
        print(f'          {c:5d}× {lits}')

print(f'\n合计: 可模板化 {len(good)} / 共 {len(good) + len(bad)} 个系列')