feat(chat): tier-3 larger messages — protocol + split + cross-tx reassembly core

Foundation for messages too large for a single transaction (tier 2's 16-output
ceiling, ~4000 chars). A long message is now split across SEVERAL transactions
that share a random message id; each carries a slice of the one ciphertext with
GLOBAL chunk indices, and the receiver reassembles across the siblings.

Protocol: the Message header gains an optional "mid" (message id); "n" is reused
as the TOTAL chunk count and its bound raised to kHushChatMaxMessageChunks (64 ⇒
up to ~16 KB / 4 txs, tunable, ≤256 for the 2-hex index). The grouper, when a
tx's header carries "mid", emits a PARTIAL (its global-indexed chunks + message
id + total) instead of reassembling in-place; HushChatTransactionMetadata
carries those partials through. Crypto is untouched — one secretstream session
for the whole message, chunking is transport-only.

Compose: splitOutgoingMessage() returns one OutgoingChatMemos per transaction
(one element = the classic tier-1/2 single tx, byte-identical; several = tier 3).
buildOutgoingMessage (single-tx) is unchanged. Old clients bound "n" to 16 so a
tier-3 header fails to parse and its memos are dropped — graceful degradation.

Unit-tested (testHushChatLargeMessageTier3): a ~9000-char message splits into 3
txs, every output shuffled, slices gathered across siblings, reassembled and
decrypted back to the original; short message stays single-tx (no mid); the
total-chunk cap is enforced. The stateful accumulator (persisted, confirm-gated),
sequential broadcast, and progressive/incomplete UI follow in later commits.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/chat/chat_outgoing.cpp

@@ -5,6 +5,10 @@
 #include "chat_protocol.h"  // kHushChat* constants
 
 #include <nlohmann/json.hpp>
+#include <sodium.h>
+
+#include <algorithm>
+#include <ctime>
 
 namespace dragonx::chat {
 
@@ -19,7 +23,8 @@ std::string buildHeaderMemo(const std::string& replyZaddr,
                             const std::string& publicKeyHex,
                             std::int64_t sentAt,
                             const std::string& op = "",
-                            int parts = 1)
+                            int parts = 1,
+                            const std::string& messageId = "")
 {
     nlohmann::json header;
     header["h"] = 1;                          // header number (>= 1)
@@ -31,6 +36,7 @@ std::string buildHeaderMemo(const std::string& replyZaddr,
     header["p"] = publicKeyHex;               // my 64-hex crypto_kx public key
     if (!op.empty()) header["op"] = op;       // Control only; alphabetically sorts to cid,e,h,op,p,t,v,z
     if (parts > 1) header["n"] = parts;       // multi-part Message only; omitted (==1) keeps 1-part byte-identical
+    if (!messageId.empty()) header["mid"] = messageId;  // tier 3 only: ties this tx's slice to its siblings
     if (sentAt > 0) header["ts"] = sentAt;    // optional sender compose time (Unix s) — receiver shows this
     return header.dump();
 }
@@ -47,6 +53,18 @@ std::string partIndexPrefix(std::size_t i)
 
 bool present(const std::string& value) { return !value.empty(); }
 
+// A random 8-byte message id as 16 hex chars — ties a tier-3 message's sibling transactions together.
+std::string newMessageId()
+{
+    unsigned char buf[8];
+    randombytes_buf(buf, sizeof buf);
+    static const char* kHex = "0123456789abcdef";
+    std::string s;
+    s.reserve(sizeof(buf) * 2);
+    for (unsigned char b : buf) { s += kHex[b >> 4]; s += kHex[b & 0xF]; }
+    return s;
+}
+
 } // namespace
 
 std::vector<ChatSendOutput> chatSendOutputs(const OutgoingChatMemos& memos, bool utf8Prefix)
@@ -113,6 +131,81 @@ ChatComposeStatus buildOutgoingMessage(const ChatKeyPair& mine,
     return ChatComposeStatus::Ok;
 }
 
+ChatComposeStatus splitOutgoingMessage(const ChatKeyPair& mine,
+                                       const std::string& myPublicKeyHex,
+                                       const std::string& myReplyZaddr,
+                                       const std::string& peerPublicKeyHex,
+                                       const std::string& peerZaddr,
+                                       const std::string& conversationId,
+                                       const std::string& plaintext,
+                                       std::vector<OutgoingChatMemos>& out)
+{
+    if (plaintext.empty()) return ChatComposeStatus::EmptyBody;
+    if (!present(myPublicKeyHex) || !present(myReplyZaddr) || !present(peerZaddr) ||
+        !present(conversationId)) {
+        return ChatComposeStatus::MissingField;
+    }
+    if (peerPublicKeyHex.size() != kHushChatPublicKeyHexLength) return ChatComposeStatus::BadPeerKey;
+
+    // Encrypt the whole message ONCE (one secretstream session). Chunking is transport-only, so every
+    // transaction of a tier-3 message carries the same secretstream header "e" and its slice of the one
+    // ciphertext — the crypto is byte-identical to the single-tx path.
+    std::string streamHeaderHex;
+    std::string ciphertextHex;
+    if (encryptOutgoing(mine, peerPublicKeyHex, plaintext, streamHeaderHex, ciphertextHex)
+        != ChatCryptoStatus::Ok) {
+        return ChatComposeStatus::EncryptFailed;
+    }
+
+    const std::int64_t now = static_cast<std::int64_t>(std::time(nullptr));
+
+    // 1 chunk if the ciphertext fits one memo, else ceil(hex / chunkHex) chunks (chunkHex = the 512-byte
+    // memo limit minus the 2-hex global index prefix, rounded to an even hex length).
+    const std::size_t chunkHex = (kHushChatMemoByteLimit - kHushChatPartIndexHexLen) & ~std::size_t(1);
+    const std::size_t nChunks = (ciphertextHex.size() <= kHushChatMemoByteLimit)
+                                    ? 1
+                                    : (ciphertextHex.size() + chunkHex - 1) / chunkHex;
+    if (static_cast<int>(nChunks) > kHushChatMaxMessageChunks) return ChatComposeStatus::TooLong;
+
+    std::vector<OutgoingChatMemos> txs;
+
+    if (nChunks <= 1) {
+        // Tier 1: one classic payload — no index prefix, no "n", no "mid" (byte-identical on the wire).
+        OutgoingChatMemos m;
+        m.recipientZaddr = peerZaddr;
+        m.headerMemo = buildHeaderMemo(myReplyZaddr, conversationId, "Memo", streamHeaderHex,
+                                       myPublicKeyHex, now, "", 1, "");
+        m.payloadMemos = { ciphertextHex };
+        txs.push_back(std::move(m));
+    } else {
+        // Tier 2/3: index each chunk with its GLOBAL index and pack up to kHushChatMaxMessageParts chunks
+        // per transaction. One tx (≤ the per-tx cap) is tier 2 (no "mid"); several is tier 3 (shared "mid",
+        // same "n" = total chunk count in every sibling header).
+        const bool multiTx = static_cast<int>(nChunks) > kHushChatMaxMessageParts;
+        const std::string mid = multiTx ? newMessageId() : std::string();
+        const std::size_t perTx = static_cast<std::size_t>(kHushChatMaxMessageParts);
+        for (std::size_t base = 0; base < nChunks; base += perTx) {
+            OutgoingChatMemos m;
+            m.recipientZaddr = peerZaddr;
+            m.headerMemo = buildHeaderMemo(myReplyZaddr, conversationId, "Memo", streamHeaderHex,
+                                           myPublicKeyHex, now, "", static_cast<int>(nChunks), mid);
+            const std::size_t end = std::min(nChunks, base + perTx);
+            m.payloadMemos.reserve(end - base);
+            for (std::size_t i = base; i < end; ++i)
+                m.payloadMemos.push_back(partIndexPrefix(i) + ciphertextHex.substr(i * chunkHex, chunkHex));
+            txs.push_back(std::move(m));
+        }
+    }
+
+    for (const auto& m : txs) {
+        if (m.headerMemo.size() > kHushChatMemoByteLimit) return ChatComposeStatus::TooLong;
+        for (const auto& p : m.payloadMemos)
+            if (p.size() > kHushChatMemoByteLimit) return ChatComposeStatus::TooLong;
+    }
+    out = std::move(txs);
+    return ChatComposeStatus::Ok;
+}
+
 ChatComposeStatus buildOutgoingContactRequest(const std::string& myPublicKeyHex,
                                               const std::string& myReplyZaddr,
                                               const std::string& peerZaddr,

src/chat/chat_outgoing.h

@@ -58,6 +58,21 @@ ChatComposeStatus buildOutgoingMessage(const ChatKeyPair& mine,
                                        const std::string& plaintext,
                                        OutgoingChatMemos& out);
 
+// Build an ENCRYPTED message, splitting it across as many transactions as its length needs (larger-
+// messages tier 3). Returns one OutgoingChatMemos per transaction to broadcast (in order): a short
+// message yields a single element identical to buildOutgoingMessage (tier 1/2); a long one yields
+// several, each a self-contained chat send carrying a slice of the one ciphertext under a shared
+// message id. The receiver reassembles across the siblings. Refused as TooLong beyond
+// kHushChatMaxMessageChunks total chunks. The caller broadcasts each element as its own chat tx.
+ChatComposeStatus splitOutgoingMessage(const ChatKeyPair& mine,
+                                       const std::string& myPublicKeyHex,
+                                       const std::string& myReplyZaddr,
+                                       const std::string& peerPublicKeyHex,
+                                       const std::string& peerZaddr,
+                                       const std::string& conversationId,
+                                       const std::string& plaintext,
+                                       std::vector<OutgoingChatMemos>& out);
+
 // Build a plaintext contact request — no peer public key needed yet; this is how the peer first
 // learns your public key + reply address. The payload is the (plaintext) request text.
 ChatComposeStatus buildOutgoingContactRequest(const std::string& myPublicKeyHex,

src/chat/chat_protocol.cpp

@@ -163,11 +163,20 @@ HushChatHeaderParseResult parseHushChatHeaderMemo(const std::string& memo)
         }
         // Optional part count ("n"); absent/1 = classic single-payload message. Bound it so a garbage or
         // future-oversized header can't make the grouper try to collect an unbounded number of payloads.
+        // A single-tx (tier 2) message chunks at most kHushChatMaxMessageParts; a multi-tx (tier 3)
+        // message carries the larger TOTAL chunk count here alongside a "mid", capped at the chunk ceiling.
         if (auto it = object.find("n"); it != object.end() && it->is_number_integer()) {
             const int n = it->get<int>();
-            if (n < 1 || n > kHushChatMaxMessageParts) return fail("message part count out of range");
+            if (n < 1 || n > kHushChatMaxMessageChunks) return fail("message part count out of range");
             header.parts = n;
         }
+        // Optional tier-3 message id ("mid"): present only when the message is split across several
+        // transactions, tying this tx's chunks to its siblings. Short hex (the sender stamps 8 bytes).
+        if (auto it = object.find("mid"); it != object.end() && it->is_string()) {
+            header.message_id = it->get<std::string>();
+            if (header.message_id.empty() || header.message_id.size() > 64 || !isHexString(header.message_id))
+                return fail("message id must be short hex");
+        }
     } else if (!header.secretstream_header_hex.empty()) {
         return fail("contact request header must not include a secretstream header");
     }
@@ -233,6 +242,36 @@ HushChatMemoGroupingResult groupHushChatMemoOutputs(const std::vector<HushChatMe
                 if (c >= 'A' && c <= 'F') return c - 'A' + 10;
                 return -1;
             };
+            // --- Tier 3: this transaction carries only a SLICE of a larger message (message_id set) ---
+            // Surface its chunks with their GLOBAL indices as a partial; the chat service accumulates
+            // slices across the sibling txs that share this message_id and reassembles once all `parts`
+            // chunks are present. One tx holds at most kHushChatMaxMessageParts of them.
+            if (!multiHeader->message_id.empty()) {
+                HushChatMemoPair pair;
+                pair.header_position = multiHeaderPos;
+                pair.message_id = multiHeader->message_id;
+                pair.total_parts = want;
+                bool okp = true;
+                std::size_t lowPos = 0; bool haveLow = false;
+                for (const auto* p : payloads) {
+                    const std::string& m = p->memo;
+                    if (m.size() < kHushChatPartIndexHexLen || (m.size() % 2) != 0 || !isHexString(m)) { okp = false; break; }
+                    const int idx = hexVal(m[0]) * 16 + hexVal(m[1]);
+                    if (idx < 0 || idx >= want) { okp = false; break; }
+                    pair.partial_chunks.emplace_back(idx, m.substr(kHushChatPartIndexHexLen));
+                    if (!haveLow || p->position < lowPos) { lowPos = p->position; haveLow = true; }
+                }
+                if (okp && !pair.partial_chunks.empty()) {
+                    pair.header = std::move(*multiHeader);
+                    pair.payload_position = haveLow ? lowPos : 0;
+                    result.pairs.push_back(std::move(pair));
+                    return result;
+                }
+                result.issues.push_back(HushChatMemoGroupingIssueInfo{
+                    HushChatMemoGroupingIssue::MissingPayload, multiHeaderPos,
+                    "tier-3 message slice did not parse"});
+                return result;
+            }
             std::vector<std::string> chunks(static_cast<std::size_t>(want));
             std::vector<bool> seen(static_cast<std::size_t>(want), false);
             bool ok = (static_cast<int>(payloads.size()) == want);
@@ -396,6 +435,9 @@ HushChatTransactionExtractionResult extractHushChatTransactionMetadata(
         metadata.secretstream_header_hex = pair.header.secretstream_header_hex;
         metadata.payload_memo = pair.payload_memo;
         metadata.sent_at = pair.header.sent_at;   // carry the sender's compose time (0 if absent)
+        metadata.message_id = pair.message_id;    // tier 3: set when this tx is a slice of a bigger message
+        metadata.total_parts = pair.total_parts;  //         (payload_memo empty; chunks below)
+        metadata.partial_chunks = pair.partial_chunks;
         result.metadata.push_back(std::move(metadata));
     }
 

src/chat/chat_protocol.h

@@ -1,7 +1,9 @@
 #pragma once
 
 #include <cstddef>
+#include <cstdint>
 #include <string>
+#include <utility>
 #include <vector>
 
 #ifndef DRAGONX_ENABLE_CHAT
@@ -27,8 +29,13 @@ struct HushChatHeader {
     std::string public_key_hex;
     // Message only: how many payload outputs this message's ciphertext is split across (header "n").
     // 1 = the classic single-payload message (field omitted on the wire). >1 means the payloads are
-    // index-prefixed ciphertext chunks reassembled in order (larger-messages tier 2, one transaction).
+    // index-prefixed ciphertext chunks reassembled in order. For tier 2 all N chunks ride ONE tx; for
+    // tier 3 (message_id set) "n" is the TOTAL chunk count spread across several sibling transactions.
     int parts = 1;
+    // Message only, larger-messages tier 3: when a message needs more chunks than one tx can carry, it
+    // is split across several transactions that all share this random id (header "mid"); each tx's
+    // payloads carry their GLOBAL chunk index (0..parts-1). Empty = a single-tx message (tier 1/2).
+    std::string message_id;
     // Optional sender-stamped compose time (header "ts", Unix seconds). 0 = absent (older sender) → the
     // receiver falls back to the tx/receive time. Lets both sides show the SAME (send) time.
     std::int64_t sent_at = 0;
@@ -50,6 +57,12 @@ struct HushChatMemoPair {
     std::size_t header_position = 0;
     std::size_t payload_position = 0;
     std::string payload_memo;
+    // Larger-messages tier 3: a single transaction that carries only PART of a bigger message (the rest
+    // ride sibling txs sharing message_id). payload_memo is then empty; partial_chunks holds this tx's
+    // (global index, ciphertext-hex chunk) pairs and total_parts is the whole message's chunk count.
+    std::string message_id;
+    int total_parts = 0;
+    std::vector<std::pair<int, std::string>> partial_chunks;
 };
 
 enum class HushChatMemoGroupingIssue {
@@ -91,6 +104,13 @@ struct HushChatTransactionMetadata {
     std::string secretstream_header_hex; // header "e": secretstream header (hex; empty for ContactRequest)
     std::string payload_memo;            // ciphertext hex (Message) or plaintext request text (ContactRequest)
     std::int64_t sent_at = 0;            // header "ts": sender compose time (Unix s); 0 = absent → use tx time
+    // Larger-messages tier 3 (a multi-transaction Message): message_id ties this tx's slice to its
+    // siblings, total_parts is the whole message's chunk count, and partial_chunks are this tx's
+    // (global index, ciphertext-hex) chunks. Empty message_id = an ordinary single-tx message (payload
+    // in payload_memo). The chat service accumulates partials by message_id and reassembles when whole.
+    std::string message_id;
+    int total_parts = 0;
+    std::vector<std::pair<int, std::string>> partial_chunks;
 };
 
 struct HushChatTransactionExtractionResult {
@@ -102,10 +122,15 @@ struct HushChatTransactionExtractionResult {
 
 constexpr int kHushChatSupportedVersion = 0;
 constexpr std::size_t kHushChatMemoByteLimit = 512;
-// Larger-messages tier 2: the most payload outputs a single message may be split across (one tx).
-// Keeps the chat tx well under the shielded max_tx_size after Sietch decoy padding + proving cost.
-// A message that would need more parts than this is refused at compose time (tier 3 would lift it).
+// Larger-messages tier 2: the most payload outputs a single transaction may carry (one tx). Keeps the
+// chat tx well under the shielded max_tx_size after Sietch decoy padding + proving cost. A message
+// needing more chunks than this spills into sibling transactions (tier 3), up to kHushChatMaxMessageChunks.
 constexpr int kHushChatMaxMessageParts = 16;
+// Larger-messages tier 3: the most chunks a single message may span IN TOTAL, across all its sibling
+// transactions (ceil(chunks / kHushChatMaxMessageParts) txs, each a separate 0-value chat send + fee).
+// Bounds the on-chain cost / broadcast time of one message; beyond this compose returns TooLong. The
+// 2-hex global part index (below) tops out at 256, so this must stay ≤ 256.
+constexpr int kHushChatMaxMessageChunks = 64;
 // Each multi-part payload chunk is prefixed with a 2-hex-char (1-byte) part index so the receiver can
 // reorder chunks after the daemon shuffles a tx's outputs. 2 hex covers indices 0..255 (>> the cap).
 constexpr std::size_t kHushChatPartIndexHexLen = 2;