CodexCodec.kt

1package com.peachpasswords.codex23import kotlinx.serialization.json.Json4import java.security.MessageDigest5import java.util.Base646import java.util.zip.Deflater7import java.util.zip.Inflater89/**10 * Pure (Android-free) Codex byte pipeline: compress → stripe → Reed-Solomon →11 * envelopes, and the inverse for restore. Kept independent of Aztec/PDF so the12 * full round-trip is unit-testable on the JVM. Port of the browser13 * `generate.ts` / `restore.ts` / `erasure.ts` byte handling.14 */15object CodexCodec {16    val json = Json { encodeDefaults = true; ignoreUnknownKeys = true }1718    private val urlEncoder = Base64.getUrlEncoder().withoutPadding()19    private val urlDecoder = Base64.getUrlDecoder()2021    /**22     * F4: decompression bomb guard — mirrors PeachPayloadCodec's23     * MAX_SECTION_RAW_BYTES (16 MiB). A codex payload is the same vault JSON24     * the PeachScript sections carry, so the cap is identical by definition.25     */26    const val MAX_INFLATED_BYTES = 16L * 1024 * 10242728    fun base64Url(bytes: ByteArray): String = urlEncoder.encodeToString(bytes)29    fun fromBase64Url(s: String): ByteArray = urlDecoder.decode(s)3031    /** Uppercase hex of SHA-256(bytes) — per-chunk hash + vault fingerprint source. */32    fun sha256Hex(bytes: ByteArray): String {33        val digest = MessageDigest.getInstance("SHA-256").digest(bytes)34        val sb = StringBuilder(digest.size * 2)35        for (b in digest) sb.append("%02X".format(b.toInt() and 0xff))36        return sb.toString()37    }3839    fun deflate(data: ByteArray): ByteArray {40        val deflater = Deflater(Deflater.BEST_COMPRESSION).apply { setInput(data); finish() }41        val out = java.io.ByteArrayOutputStream(data.size / 2 + 32)42        val buf = ByteArray(8192)43        while (!deflater.finished()) out.write(buf, 0, deflater.deflate(buf))44        deflater.end()45        return out.toByteArray()46    }4748    fun inflate(data: ByteArray): ByteArray {49        val inflater = Inflater().apply { setInput(data) }50        val out = java.io.ByteArrayOutputStream(data.size * 3 + 32)51        val buf = ByteArray(8192)52        while (!inflater.finished()) {53            val n = inflater.inflate(buf)54            if (n == 0 && inflater.needsInput()) break // trailing zero-pad reached55            out.write(buf, 0, n)56            if (out.size() > MAX_INFLATED_BYTES) {57                inflater.end()58                throw CodexException(59                    CodexErrorCode.PAYLOAD_TOO_LARGE,60                    "Inflated payload exceeds the ${MAX_INFLATED_BYTES / (1024 * 1024)} MiB cap",61                )62            }63        }64        inflater.end()65        return out.toByteArray()66    }6768    // ── Stripe layout ─────────────────────────────────────────────────────6970    fun planStripeShardCounts(totalDataShards: Int): IntArray {71        if (totalDataShards <= 0) return IntArray(0)72        val numStripes = maxOf(1, Math.ceil(totalDataShards.toDouble() / MAX_DATA_SHARDS_PER_STRIPE).toInt())73        val base = totalDataShards / numStripes74        val remainder = totalDataShards % numStripes75        return IntArray(numStripes) { base + if (it < remainder) 1 else 0 }76    }7778    data class Stripe(val dataShards: Int, val parityShards: Int, val shards: Array<ByteArray>) {79        val totalChunks: Int get() = dataShards + parityShards80    }8182    /** Compress + stripe + RS-encode into envelopes for [verifyCode]. */83    fun encodeToEnvelopes(compressed: ByteArray, verifyCode: String): Pair<List<ChunkEnvelope>, CodexLayout> {84        val totalDataShards = maxOf(1, Math.ceil(compressed.size.toDouble() / CHUNK_SIZE).toInt())85        val counts = planStripeShardCounts(totalDataShards)86        val envelopes = ArrayList<ChunkEnvelope>()87        var offset = 088        var totalData = 089        var totalParity = 090        for ((sIdx, dataShards) in counts.withIndex()) {91            val region = dataShards * CHUNK_SIZE92            val parityShards = computeParityShards(dataShards)93            val total = dataShards + parityShards94            val shards = Array(total) { ByteArray(CHUNK_SIZE) }95            for (d in 0 until dataShards) {96                val start = offset + d * CHUNK_SIZE97                val len = (compressed.size - start).coerceIn(0, CHUNK_SIZE)98                if (len > 0) System.arraycopy(compressed, start, shards[d], 0, len)99            }100            ReedSolomon.encode(shards, dataShards, parityShards)101            for (i in 0 until total) {102                envelopes.add(buildEnvelope(shards[i], i, total, i >= dataShards, verifyCode, sIdx, counts.size))103            }104            offset += region105            totalData += dataShards106            totalParity += parityShards107        }108        return envelopes to CodexLayout(counts.size, totalData, totalParity, totalData + totalParity)109    }110111    private fun buildEnvelope(112        chunk: ByteArray, index: Int, total: Int, isParity: Boolean,113        verifyCode: String, stripe: Int, stripeCount: Int,114    ): ChunkEnvelope = ChunkEnvelope(115        v = 3, c = verifyCode, s = stripe, capS = stripeCount,116        i = index, t = total, p = isParity,117        d = base64Url(chunk), h = sha256Hex(chunk).take(32),118    )119120    /**121     * Upper bound on one chunk's decoded bytes, not the generator's exact122     * [CHUNK_SIZE]: the browser's decoder derives the shard size from the123     * first scanned chunk precisely so Codexes printed before the current124     * chunk size (800 bytes) still restore, and it enforces no length of its125     * own. The bound sits above a full-range Aztec code's alphanumeric126     * capacity, so it rejects only what no barcode could have carried.127     */128    private const val MAX_CHUNK_BYTES = 4096129    private val MAX_ENCODED_CHUNK_CHARS = base64Url(ByteArray(MAX_CHUNK_BYTES)).length130131    /**132     * Budget for the whole reconstructed payload. [inflate] refuses anything133     * over [MAX_INFLATED_BYTES], so a payload larger than this can never134     * succeed; the headroom covers deflate framing on a payload at the cap.135     */136    private val MAX_RECONSTRUCTED_BYTES = MAX_INFLATED_BYTES + 1024 * 1024137138    private val VERIFY_CODE = Regex("PX-[0-9A-Z]{4}-[0-9A-Z]{4}")139140    /**141     * Stripe ceiling implied by the inflate cap: the largest legal payload142     * divided by the largest legal stripe, rounded up. The extra stripe covers143     * deflate's own framing overhead, which can push the compressed payload a144     * fraction above the uncompressed cap.145     */146    private val MAX_STRIPES: Int =147        ((MAX_INFLATED_BYTES + MAX_DATA_SHARDS_PER_STRIPE * CHUNK_SIZE - 1) /148            (MAX_DATA_SHARDS_PER_STRIPE * CHUNK_SIZE)).toInt() + 1149150    /**151     * Reject a scanned envelope whose declared geometry no real Codex can152     * produce, before anything is allocated from it (PAI-2026-002).153     *154     * The per-chunk hash covers `d` only — `t`, `S`, `s` and `i` are unauthenticated155     * integers, and reconstruction sized `Array(total)` straight from `t`. A156     * barcode with a valid hash over a tiny payload and t = Int.MAX_VALUE157     * therefore cost a multi-gigabyte allocation before it could be rejected.158     * Every bound here is the encoder's own, except the chunk size, which is a159     * ceiling rather than an exact match so older paper still restores (see160     * [MAX_CHUNK_BYTES]). [buildEnvelope] emits at most161     * [MAX_TOTAL_SHARDS_PER_STRIPE] shards per stripe and at most162     * [MAX_STRIPES] stripes.163     */164    internal fun isValidEnvelope(env: ChunkEnvelope): Boolean =165        env.v == CODEX_VERSION &&166            VERIFY_CODE.matches(env.c) &&167            env.capS in 1..MAX_STRIPES &&168            env.s in 0 until env.capS &&169            env.t in 1..MAX_TOTAL_SHARDS_PER_STRIPE &&170            env.i in 0 until env.t &&171            env.d.length in 1..MAX_ENCODED_CHUNK_CHARS &&172            (env.h.length == 16 || env.h.length == 32)173174    fun verifyChunkIntegrity(env: ChunkEnvelope): Boolean {175        if (env.h.length != 16 && env.h.length != 32) return false176        return sha256Hex(fromBase64Url(env.d)).take(env.h.length).equals(env.h, ignoreCase = true)177    }178179    /**180     * Reconstruct the compressed payload from scanned envelopes, tolerating181     * lost codes via Reed-Solomon. Groups by stripe, reconstructs each, then182     * concatenates the data regions in stripe order.183     */184    fun reconstructCompressed(envelopes: List<ChunkEnvelope>): ByteArray {185        val good = envelopes.filter { isValidEnvelope(it) && verifyChunkIntegrity(it) }186        if (good.isEmpty()) throw CodexException(CodexErrorCode.INSUFFICIENT_CHUNKS, "no valid chunks")187        val verifyCode = good.first().c188        val stripeCount = good.first().capS189        if (good.any { it.c != verifyCode || it.capS != stripeCount }) {190            throw CodexException(CodexErrorCode.INVALID_ENVELOPE, "mixed verify codes or stripe counts")191        }192        val byStripe = good.groupBy { it.s }193        var reconstructedBytes = 0L194        val out = java.io.ByteArrayOutputStream()195        for (sIdx in 0 until stripeCount) {196            val chunks = byStripe[sIdx]197                ?: throw CodexException(CodexErrorCode.INSUFFICIENT_CHUNKS, "stripe $sIdx missing entirely")198            val total = chunks.first().t199            if (chunks.any { it.t != total }) {200                throw CodexException(CodexErrorCode.INVALID_ENVELOPE, "mixed shard totals in stripe $sIdx")201            }202            val (dataShards, parityShards) = deriveShardCounts(total)203            val shards = Array(total) { ByteArray(0) }204            val present = BooleanArray(total)205            for (env in chunks) {206                if (env.i in 0 until total) {207                    val decoded = fromBase64Url(env.d)208                    reconstructedBytes += decoded.size209                    if (reconstructedBytes > MAX_RECONSTRUCTED_BYTES) {210                        throw CodexException(211                            CodexErrorCode.PAYLOAD_TOO_LARGE,212                            "Reconstructed payload exceeds the ${MAX_INFLATED_BYTES / (1024 * 1024)} MiB cap",213                        )214                    }215                    shards[env.i] = decoded216                    present[env.i] = true217                }218            }219            ReedSolomon.reconstruct(shards, dataShards, parityShards, present)220            for (d in 0 until dataShards) out.write(shards[d])221        }222        return out.toByteArray()223    }224}225226data class CodexLayout(val stripeCount: Int, val dataChunks: Int, val parityChunks: Int, val totalChunks: Int)