From 2036ee2c4ffc2074fd0d1dfdb845416a67b76954 Mon Sep 17 00:00:00 2001 From: mayx Date: Mon, 7 Sep 2026 09:23:41 +0000 Subject: Update 5 files - /_tools/blogquine.py - /_data/proxylist.yml - /_layouts/default.html - /deploy.sh - /.readthedocs.yaml --- .readthedocs.yaml | 4 +- _data/proxylist.yml | 1 - _layouts/default.html | 2 +- _tools/blogquine.py | 755 ++++++++++++++++++++++++++++++++++++++++++++++++++ deploy.sh | 4 +- 5 files changed, 760 insertions(+), 6 deletions(-) create mode 100644 _tools/blogquine.py diff --git a/.readthedocs.yaml b/.readthedocs.yaml index 5124702..6ef414a 100644 --- a/.readthedocs.yaml +++ b/.readthedocs.yaml @@ -12,5 +12,5 @@ build: - mkdir Mabbs - curl -L -o Mabbs/README.md https://github.com/Mabbs/Mabbs/raw/main/README.md - bundle exec jekyll build --destination $READTHEDOCS_OUTPUT/html - - tar czvf MayxBlog.tgz -C $READTHEDOCS_OUTPUT html - - mv MayxBlog.tgz $READTHEDOCS_OUTPUT/html \ No newline at end of file + - python3 _tools/blogquine.py $READTHEDOCS_OUTPUT/html MayxBlog.7z + - mv MayxBlog.7z $READTHEDOCS_OUTPUT/html \ No newline at end of file diff --git a/_data/proxylist.yml b/_data/proxylist.yml index cdf3305..e4e85b3 100644 --- a/_data/proxylist.yml +++ b/_data/proxylist.yml @@ -22,7 +22,6 @@ mirrors: - https://mabbs.kinsta.page/ - https://mayx.codeberg.page/ - https://mayx.tildepages.org/ -- https://mayx.pages.lain.la/ - https://mayx.4everland.app/ - https://mayx.readthedocs.io/ - https://mayx.sourceforge.io/ diff --git a/_layouts/default.html b/_layouts/default.html index b3f6133..616f962 100644 --- a/_layouts/default.html +++ b/_layouts/default.html @@ -98,7 +98,7 @@ layout: xslt_container
  • Download TAR Ball
  • View On GitHub
    • {% else %} -
  • Download TGZ File
    • +
  • Download 7-Zip File
    • {% endif %} diff --git a/_tools/blogquine.py b/_tools/blogquine.py new file mode 100644 index 0000000..6fe32d3 --- /dev/null +++ b/_tools/blogquine.py @@ -0,0 +1,755 @@ +#!/usr/bin/env python3 +# -*- coding: utf-8 -*- + +import argparse +import binascii +import lzma +import os +import struct +import sys + +# ============================================================ +# Part 1: 最小 LZMA1 range coder（只做编码） +# +# 设计目标：只实现 quine 构造需要的 token：literal / match / rep0-match / +# end-marker，编码字节与历史内容无关（track=False 时不维护输出历史）， +# 这是"先算结构、后填数据"两阶段装配的基础。 +# 参考：Igor Pavlov 的 LZMA SDK (LzmaEnc.c / LzmaDec.c)。 +# ============================================================ + +kNumBitModelTotalBits = 11 +kBitModelTotal = 1 << kNumBitModelTotalBits # 2048 +kNumMoveBits = 5 +kTopValue = 1 << 24 # 0x1000000 + +kNumStates = 12 +kNumLitStates = 7 +kNumPosBitsMax = 4 +kNumLenToPosStates = 4 +kNumAlignBits = 4 +kEndPosModelIndex = 14 +kNumFullDistances = 1 << (kEndPosModelIndex >> 1) # 128 +kMatchMinLen = 2 +kNumLowLenBits = 3 +kNumMidLenBits = 3 +kNumHighLenBits = 8 +kNumLowLenSymbols = 1 << kNumLowLenBits # 8 +kNumMidLenSymbols = 1 << kNumMidLenBits # 8 +kNumPosSlotBits = 6 + +PROB_INIT = kBitModelTotal >> 1 # 1024 + + +class RangeEncoder: + """LZMA 的区间编码器。low 是 64 位（要容纳进位），range 是 32 位。""" + + def __init__(self): + self.low = 0 + self.range = 0xFFFFFFFF + self.cache = 0 + self.cache_size = 1 # 初值 1 -> 第一个输出字节恒为 0x00 + self.buf = bytearray() + + def _shift_low(self): + if (self.low & 0xFFFFFFFF) < 0xFF000000 or (self.low >> 32) != 0: + temp = self.cache + while True: + self.buf.append((temp + (self.low >> 32)) & 0xFF) + temp = 0xFF + self.cache_size -= 1 + if self.cache_size == 0: + break + self.cache = (self.low >> 24) & 0xFF + self.cache_size += 1 + # 注意：C 里 `(UInt32)low << 8` 会截断回 32 位，必须照做， + # 否则 low 的高位会一直累积，进位位 (low>>32) 就不止 0/1 了。 + self.low = ((self.low & 0xFFFFFFFF) << 8) & 0xFFFFFFFF + + def encode_bit(self, probs, idx, bit): + p = probs[idx] + bound = (self.range >> kNumBitModelTotalBits) * p + if bit == 0: + self.range = bound + probs[idx] = p + ((kBitModelTotal - p) >> kNumMoveBits) + else: + self.low += bound + self.range -= bound + probs[idx] = p - (p >> kNumMoveBits) + while self.range < kTopValue: + self.range = (self.range << 8) & 0xFFFFFFFF + self._shift_low() + + def encode_direct_bits(self, value, num_bits): + for i in range(num_bits - 1, -1, -1): + self.range >>= 1 + if (value >> i) & 1: + self.low += self.range + while self.range < kTopValue: + self.range = (self.range << 8) & 0xFFFFFFFF + self._shift_low() + + def bittree_encode(self, probs, off, num_bits, symbol): + """普通位树：MSB 优先，节点下标从 1 开始。""" + m = 1 + for i in range(num_bits - 1, -1, -1): + bit = (symbol >> i) & 1 + self.encode_bit(probs, off + m, bit) + m = (m << 1) | bit + + def bittree_reverse_encode(self, probs, off, num_bits, symbol): + """反向位树：LSB 优先，用于 SpecPos / Align。""" + m = 1 + for i in range(num_bits): + bit = symbol & 1 + symbol >>= 1 + self.encode_bit(probs, off + m, bit) + m = (m << 1) | bit + + def finish(self): + for _ in range(5): + self._shift_low() + return bytes(self.buf) + + +def _pos_slot_and_bits(d): + """把 0 基距离 d (=真实距离-1) 拆成 (posSlot, 低位数, 低位值)。""" + if d < 4: + return d, 0, 0 + # posSlot >= 4 时：numDirectBits=(slot>>1)-1, base=(2|(slot&1))<> 1) - 1 + base = (2 | (slot & 1)) << n + if d < base + (1 << n): + return slot, n, d - base + raise ValueError("distance too large: %d" % d) + + +class LzmaEncoder: + def __init__(self, lc=3, lp=0, pb=2, track=False): + """ + track=True 时维护输出假历史（literal/match 复制内容），用于一般用途； + quine 构造用 track=False：match 的编码字节只取决于 (dist,len,pos)， + 与历史内容无关，可跳过 37MB 级假历史的逐字节复制。 + """ + self.lc, self.lp, self.pb = lc, lp, pb + self.track = track + self.pos_mask = (1 << pb) - 1 + self.lp_mask = (1 << lp) - 1 + self.rc = RangeEncoder() + self.pos = 0 # 已输出字节数 + self.state = 0 + self.prev_byte = 0 + self.reps = [0, 0, 0, 0] # rep0..rep3，存的是 0 基距离 + self.out = bytearray() + # 概率模型 + n = PROB_INIT + self.p_is_match = [n] * (kNumStates << kNumPosBitsMax) + self.p_is_rep = [n] * kNumStates + self.p_is_rep_g0 = [n] * kNumStates + self.p_is_rep_g1 = [n] * kNumStates + self.p_is_rep_g2 = [n] * kNumStates + self.p_rep0_long = [n] * (kNumStates << kNumPosBitsMax) + self.p_pos_slot = [n] * (kNumLenToPosStates << kNumPosSlotBits) + self.p_spec_pos = [n] * (kNumFullDistances - kEndPosModelIndex) # 114 + self.p_align = [n] * (1 << kNumAlignBits) + self.p_len_choice = [n] * 2 + self.p_len_low = [n] * (16 * kNumLowLenSymbols) + self.p_len_mid = [n] * (16 * kNumMidLenSymbols) + self.p_len_high = [n] * (1 << kNumHighLenBits) + self.p_rep_len_choice = [n] * 2 + self.p_rep_len_low = [n] * (16 * kNumLowLenSymbols) + self.p_rep_len_mid = [n] * (16 * kNumMidLenSymbols) + self.p_rep_len_high = [n] * (1 << kNumHighLenBits) + self.p_lit = [n] * (0x300 << (lc + lp)) + + # ---------- 内部 ---------- + @property + def pos_state(self): + return self.pos & self.pos_mask + + def _encode_len(self, choice, low, mid, high, length): + """choice 是 2 元素数组：choice[0]=choice 位，choice[1]=choice2 位。""" + ps = self.pos_state + l = length - kMatchMinLen + if l < kNumLowLenSymbols: + self.rc.encode_bit(choice, 0, 0) + self.rc.bittree_encode(low, ps << kNumLowLenBits, kNumLowLenBits, l) + else: + self.rc.encode_bit(choice, 0, 1) + l -= kNumLowLenSymbols + if l < kNumMidLenSymbols: + self.rc.encode_bit(choice, 1, 0) + self.rc.bittree_encode(mid, ps << kNumMidLenBits, kNumMidLenBits, l) + else: + self.rc.encode_bit(choice, 1, 1) + self.rc.bittree_encode(high, 0, kNumHighLenBits, l - kNumMidLenSymbols) + + def _copy_out(self, dist, length): + """track=True 时把 match 的输出复制进假历史。""" + src = len(self.out) - dist + for i in range(length): + self.out.append(self.out[src + i]) + self.prev_byte = self.out[-1] + + # ---------- 对外 token ---------- + def literal(self, b): + if self.state >= kNumLitStates: + raise NotImplementedError("matched-literal 未实现（quine 构造不需要）") + # 每个符号先编 isMatch 位：0 表示这是 literal + ps = self.pos_state + self.rc.encode_bit(self.p_is_match, (self.state << kNumPosBitsMax) + ps, 0) + lit_state = ((self.pos & self.lp_mask) << self.lc) + (self.prev_byte >> (8 - self.lc)) + self.rc.bittree_encode(self.p_lit, lit_state * 0x300, 8, b) + self.out.append(b) + self.prev_byte = b + self.pos += 1 + # UpdateState_Literal + if self.state <= 3: + self.state = 0 + elif self.state <= 9: + self.state -= 3 + else: + self.state -= 6 + + def _write_dist(self, dist, lts): + d = dist - 1 + slot, n, low_bits = _pos_slot_and_bits(d) + self.rc.bittree_encode(self.p_pos_slot, lts << kNumPosSlotBits, + kNumPosSlotBits, slot) + if slot >= 4: + if slot < kEndPosModelIndex: + base = (2 | (slot & 1)) << n + off = base - slot - 1 + self.rc.bittree_reverse_encode(self.p_spec_pos, off, n, low_bits) + else: + self.rc.encode_direct_bits(low_bits >> kNumAlignBits, n - kNumAlignBits) + self.rc.bittree_reverse_encode(self.p_align, 0, kNumAlignBits, + low_bits & ((1 << kNumAlignBits) - 1)) + return d + + def match(self, dist, length): + assert kMatchMinLen <= length <= 273, length + ps = self.pos_state + self.rc.encode_bit(self.p_is_match, (self.state << kNumPosBitsMax) + ps, 1) + self.rc.encode_bit(self.p_is_rep, self.state, 0) # 非 rep + lts = min(length - kMatchMinLen, kNumLenToPosStates - 1) + self._encode_len(self.p_len_choice, + self.p_len_low, self.p_len_mid, self.p_len_high, length) + d = self._write_dist(dist, lts) + # 状态与 rep 链更新 + self.state = 7 if self.state < kNumLitStates else 10 + self.reps[3], self.reps[2], self.reps[1], self.reps[0] = \ + self.reps[2], self.reps[1], self.reps[0], d + if self.track: + self._copy_out(dist, length) + self.pos += length + + def rep_match(self, length, rep_idx=0): + assert kMatchMinLen <= length <= 273, length + ps = self.pos_state + self.rc.encode_bit(self.p_is_match, (self.state << kNumPosBitsMax) + ps, 1) + self.rc.encode_bit(self.p_is_rep, self.state, 1) + if rep_idx == 0: + self.rc.encode_bit(self.p_is_rep_g0, self.state, 0) + self.rc.encode_bit(self.p_rep0_long, (self.state << kNumPosBitsMax) + ps, 1) + else: + self.rc.encode_bit(self.p_is_rep_g0, self.state, 1) + if rep_idx == 1: + self.rc.encode_bit(self.p_is_rep_g1, self.state, 0) + else: + self.rc.encode_bit(self.p_is_rep_g1, self.state, 1) + self.rc.encode_bit(self.p_is_rep_g2, self.state, 0 if rep_idx == 2 else 1) + if rep_idx == 3: + self.reps[3] = self.reps[2] + self.reps[2] = self.reps[1] + self.reps[1] = self.reps[0] + self.reps[0] = self.reps[rep_idx] + self._encode_len(self.p_rep_len_choice, + self.p_rep_len_low, self.p_rep_len_mid, self.p_rep_len_high, length) + dist = self.reps[0] + 1 + self.state = 8 if self.state < kNumLitStates else 11 + if self.track: + self._copy_out(dist, length) + self.pos += length + + def end(self): + """LZMA 结束标记：编码一个 0 基距离为 0xFFFFFFFF 的 match。""" + ps = self.pos_state + self.rc.encode_bit(self.p_is_match, (self.state << kNumPosBitsMax) + ps, 1) + self.rc.encode_bit(self.p_is_rep, self.state, 0) + self._encode_len(self.p_len_choice, + self.p_len_low, self.p_len_mid, self.p_len_high, kMatchMinLen) + self._write_dist(0x100000000, 0) + self.state = 7 if self.state < kNumLitStates else 10 + + def finish(self): + return self.rc.finish() + + +# ============================================================ +# Part 2: 7z 格式原语 +# ============================================================ + +SIG = b"7z\xbc\xaf'\x1c" +VER = b"\x00\x04" +MAX_MATCH = 273 # LZMA1 单个 match 长度上限 + + +def crc32(b, v=0): + return binascii.crc32(b, v) & 0xFFFFFFFF + + +def varint(v): + """7z UINT64 变长编码：首字节高位连续 n 个 1 + 值的 (7-n) 个高位，再跟 n 字节 LE。""" + if v < 0x80: + return bytes([v]) + for n in range(1, 9): + if v < (1 << (8 * n + 7 - n)): + break + else: + raise ValueError("varint too big") + first = ((0xFF << (8 - n)) & 0xFF) | ((v >> (8 * n)) & ((1 << (7 - n)) - 1)) + return bytes([first]) + (v & ((1 << (8 * n)) - 1)).to_bytes(n, "little") + + +def store_hdr(payload_len, first=False): + """LZMA2 uncompressed chunk 头（3 字节，大端 size-1）。""" + assert 1 <= payload_len <= 65536 + return bytes([0x01 if first else 0x02]) + (payload_len - 1).to_bytes(2, "big") + + +def matches_for(dist, total): + """dist 固定、总长 total 的 token 序列：首个 match + 后续 rep0。 + 长度按 273 上限切分，并避开"只剩 1 字节"（rep0 最小长度为 2）。""" + assert total >= 2 + toks = [] + first = min(MAX_MATCH, total) + if total - first == 1: + first -= 1 + toks.append(("m", dist, first)) + total -= first + while total > 0: + l = min(MAX_MATCH, total) + if total - l == 1: + l -= 1 + toks.append(("r0", l)) + total -= l + return toks + + +# ============================================================ +# Part 3: 多文件 7z quine 构造器 +# ============================================================ + +PROPS_BYTE = 0x5D # lc=3, lp=0, pb=2 +CHUNK = 65536 # LZMA2 单 chunk 解压上限 +SEED_NOTE = ("\nMayx's Blog!").encode("utf-8") + + +def dict_prop_for(maxdist): + """选最小的 LZMA2 dict prop 使字典 >= maxdist。""" + for p in range(41): + if (2 | (p & 1)) << (p // 2 + 11) >= maxdist: + return p, (2 | (p & 1)) << (p // 2 + 11) + raise ValueError("distance too large") + + +def lzma_chunk(tokens, out_pos): + """编一个 LZMA chunk（0xC0: state+props reset，无 dict reset，无 end marker）。 + match token 的编码字节只取决于 (dist,len,pos)，与历史内容无关（track=False）。""" + enc = LzmaEncoder() + enc.pos = out_pos + total = 0 + for t in tokens: + if t[0] == "m": + enc.match(t[1], t[2]) + total += t[2] + else: + enc.rep_match(t[1], 0) + total += t[1] + data = enc.finish() + assert total - 1 < 65536 and len(data) - 1 < 65536 + hdr = bytes([0xC0]) + (total - 1).to_bytes(2, "big") + \ + (len(data) - 1).to_bytes(2, "big") + bytes([PROPS_BYTE]) + return hdr + data, total + + +class BlogQuine: + def __init__(self, root, quine_name="MayxBlog.7z", seed_name="_quine_seed.bin"): + self.quine_name = quine_name + self.seed_name = seed_name + # walk_entries: walk 顺序（父先子后、同级 dirs 在前）的 (rel, is_dir)。 + # 子流顺序 = 非空条目在此列表中的顺序，必须自始至终保持同一顺序。 + self.walk_entries = [] + self.files = [] # (relpath, data bytes)，顺序 = walk_entries 中文件序 + for dirpath, dirnames, filenames in os.walk(root): + dirnames.sort() + for dn in dirnames: + rel = os.path.relpath(os.path.join(dirpath, dn), root) + self.walk_entries.append((rel.replace(os.sep, "/"), True)) + for fn in sorted(filenames): + p = os.path.join(dirpath, fn) + rel = os.path.relpath(p, root).replace(os.sep, "/") + with open(p, "rb") as fp: + data = fp.read() + self.walk_entries.append((rel.replace(os.sep, "/"), False)) + self.files.append((rel.replace(os.sep, "/"), data)) + self.dirs = [r for r, isd in self.walk_entries if isd] + self.content = b"".join(data for _, data in self.files) + + def entries(self): + ent = [{"name": self.seed_name, "dir": False}] + ent += [{"name": r, "dir": isd} for r, isd in self.walk_entries] + ent.append({"name": self.quine_name, "dir": False}) + return ent + + # ---------- 7z header ---------- + def build_header(self, n, total, d, entries, sub_sizes): + n_sub = len(sub_sizes) + N = len(entries) + h = bytearray() + h += b"\x01\x04" # kHeader, kMainStreamsInfo + h += b"\x06" + varint(0) + varint(1) + b"\x09" + varint(n) + b"\x00" # PackInfo + h += b"\x07" # UnpackInfo + h += b"\x0b" + varint(1) + b"\x00" # kFolder: 1, local + h += varint(1) + b"\x21\x21" + varint(1) + bytes([self.dict_prop]) + h += b"\x0c" + varint(d + total) # kCodersUnpackSize + h += b"\x00" # end UnpackInfo + h += b"\x08" # SubStreamsInfo + h += b"\x0d" + varint(n_sub) # 子流数 + h += b"\x09" + b"".join(varint(s) for s in sub_sizes[:-1]) # 前 n_sub-1 个大小 + h += b"\x0a" + b"\x01" # kCRC: AllDefined + crc_base = len(h) + h += b"\x00" * (4 * n_sub) # CRC 槽（quine 的待定点） + h += b"\x00\x00" # end SSI, end MainStreamsInfo + h += b"\x05" + varint(N) # FilesInfo + bits = bytearray((N + 7) // 8) # kEmptyStream 位域（MSB 优先） + for i, e in enumerate(entries): + if e["dir"]: + bits[i // 8] |= 1 << (7 - i % 8) + h += b"\x0e" + varint(len(bits)) + bytes(bits) + names = bytearray(b"\x00") # kName: external=0 + for e in entries: + names += e["name"].encode("utf-16-le") + b"\x00\x00" + h += b"\x11" + varint(len(names)) + bytes(names) + attrs = b"".join(struct.pack(" CHUNK: + Ls.append(CHUNK) + rem -= CHUNK + Ls.append(rem) + assert Ls[-1] >= 2, "末 chunk 载荷 %d 太小，调整 d_seed" % Ls[-1] + k = len(Ls) + hdrA = store_hdr(CHUNK, first=False) # 0x02 FF FF + hdrB = store_hdr(Ls[-1], first=False) # 末 chunk 头（可能与 A 相同） + self.dict_prop, dict_size = dict_prop_for(maxdist_hint) + # 子流大小（quine 的 total 在迭代后填）：seed, 各文件, quine + file_sizes = [len(data) for _, data in self.files] + + T, h = 100, 100 + for _ in range(30): + chunks = [] + o, f = 0, 32 + S = 0 + for j, L in enumerate(Ls): # C1a..C1k + chunks.append({"kind": "store", "foff": f, "size": 3 + L, + "ooff": o, "olen": L, "first": j == 0, + "c1": True, "S": S}) + o += L + f += 3 + L + S += L + assert o == d + 35 + # repro：复现 file[35 : 32+3k+d+35) = pay0 hdr1 pay1 ... hdr(k-1) pay(k-1) + S = 0 + for j, L in enumerate(Ls): + if j > 0: + # chunk j 的 3 字节头：从 seed 开头样本复制 + srcpos = 0 if (j < k - 1 or L == CHUNK) else 3 + dist_h = (d + 35 + S + 3 * (j - 1)) - srcpos + hb, hu = lzma_chunk([("m", dist_h, 3)], o) + assert hu == 3 + chunks.append({"kind": "lzma", "foff": f, "size": len(hb), + "ooff": o, "olen": hu, "bytes": hb}) + o += hu + f += len(hb) + # chunk j 的载荷：从 W[S:S+L) 复制 + cb, cu = lzma_chunk(matches_for(d + 35 + 3 * j, L), o) + assert cu == L + chunks.append({"kind": "lzma", "foff": f, "size": len(cb), + "ooff": o, "olen": cu, "bytes": cb}) + o += cu + f += len(cb) + S += L + # gears: store(x)+copy(x) 把 slip 压到 <=16 + while f - (o - d) > 16: + s = f - (o - d) + x = min(s + 3, CHUNK) + chunks.append({"kind": "store", "foff": f, "size": 3 + x, + "ooff": o, "olen": x}) + o += x + f += 3 + x + cb, cu = lzma_chunk(matches_for(x, x), o) + chunks.append({"kind": "lzma", "foff": f, "size": len(cb), + "ooff": o, "olen": cu, "bytes": cb}) + o += cu + f += len(cb) + # jump: 单 match 从 W[6:6+y)（seed 里的种植串）复制，使 slip = -3 + s = f - (o - d) + y = s + 17 + for _ in range(10): + jb, ju = lzma_chunk([("m", o - 6, y)], o) + y2 = s + 3 + len(jb) + if y2 == y: + break + y = y2 + assert ju == y and y2 == y, "jump 不动点失败" + assert y + 6 <= d_seed, "种植串 %d+6 超过 seed %d" % (y, d_seed) + chunks.append({"kind": "lzma", "foff": f, "size": len(jb), + "ooff": o, "olen": ju, "bytes": jb, "jump": True}) + o += ju + f += len(jb) + # gadget: 真空 store(T) + match(T,T) + 结束标记 + assert f == (o - d) - 3, "slip=%d 应为 -3" % (f - (o - d)) + chunks.append({"kind": "store", "foff": f, "size": 3 + T, + "ooff": o, "olen": T, "vac": True}) + f += 3 + T + o += T + gb, gu = lzma_chunk(matches_for(T, T), o) + assert gu == T + chunks.append({"kind": "lzma", "foff": f, "size": len(gb), + "ooff": o, "olen": gu, "bytes": gb, "gmatch": True}) + f += len(gb) + o += gu + # LZMA2 流结束标记（0x00）：7z t 需要它确认流完整结束 + chunks.append({"kind": "term", "foff": f, "size": 1, + "ooff": o, "olen": 0, "bytes": b"\x00"}) + f += 1 + n = f - 32 + total = f + h + sub_sizes = [d_seed] + file_sizes + [total] + entries = self.entries() + hdr, crc_base = self.build_header(n, total, d, entries, sub_sizes) + h_new = len(hdr) + T_new = len(gb) + 1 + h_new + if T_new == T and h_new == h: + assert o == d + f + h, "输出终点 %d != d+total %d" % (o, d + f + h) + return {"d": d, "d_seed": d_seed, "k": k, "Ls": Ls, + "chunks": chunks, "header": hdr, "crc_base": crc_base, + "n": n, "total": f + h, "T": T, "h": h, + "hdrA": hdrA, "hdrB": hdrB, "jump_y": y, + "n_sub": len(sub_sizes), "sub_sizes": sub_sizes, + "entries": entries, "dict_size": dict_size} + T, h = T_new, h_new + raise RuntimeError("layout 不收敛") + + # ---------- 装配 ---------- + def assemble(self, lay): + F = bytearray(lay["total"]) + d, T, h = lay["d"], lay["T"], lay["h"] + d_seed = lay["d_seed"] + hoff = lay["total"] - h + # pass A: sig + 所有非 payload 字节 + F[0:6] = SIG + F[6:8] = VER + struct.pack_into("> i) & 1: + coeff |= 1 << j + coeff ^= 1 << i # 对角元 = 1 ^ A[i][i] + eqs.append([coeff, (basevec >> i) & 1]) + for j in range(96): + p = next((i for i in range(j, 96) if (eqs[i][0] >> j) & 1), None) + if p is None: + raise RuntimeError("GF(2) 奇异 @bit%d" % j) + eqs[j], eqs[p] = eqs[p], eqs[j] + for i in range(96): + if i != j and ((eqs[i][0] >> j) & 1): + eqs[i][0] ^= eqs[j][0] + eqs[i][1] ^= eqs[j][1] + x = 0 + for i in range(96): + assert eqs[i][0] == (1 << i) + if eqs[i][1]: + x |= 1 << i + vals = [x & 0xFFFFFFFF, (x >> 32) & 0xFFFFFFFF, (x >> 64) & 0xFFFFFFFF] + for g in range(3): + for p in groups[g]: + F[p:p + 4] = struct.pack("