Repository navigation
Beta13 is a new wallet and a new node (both v2.1.2-beta13). The shielded cryptography no longer uses Rust or libsnark: it is the C23 core in src/c23. Start-up and balance queries are faster, shielded sends use less memory, wallet state survives hard kills and chain reorganisations with correct balances, and syncing from block 0 works again.
No consensus rule, checkpoint, activation height or the IBD proof-skip is changed, by any change listed below either. The only edits to a sealed consensus file since the first beta13 notes are in src/chainparams.cpp: a dead DNS seed is removed and two snapshot server addresses (with their onion addresses) are added. These are network addresses, not consensus parameters. One owner-approved consensus-adjacent change: the 413 historical oversize transactions are admitted by an explicit allowlist (the same table Z23 uses), which is what makes a sync from block 0 possible.
What is new
- No Rust, no libsnark. Sapling, Sprout (Groth16 and PHGR13) and Equihash verification are C23, checked against a manifest.
sprout-proving.key, libgmp and OpenMP are no longer needed. New Sprout PHGR13 proofs can no longer be created (they only existed before Sapling); they are still verified. - Faster. Proof verification about 2x; Equihash checked once per block; start-up of a full mainnet node about 24 s to 8 s;
z_getbalancewith 3,252 notes 1.29 s to 0.012 s; note scanning up to 4x faster with many addresses. - Less memory. Peak for a 20-40 note shielded send about 850 MB to about 390 MB.
- Wallet reliability. A hard kill can no longer leave a spent shielded note counted as unspent. A reorganisation no longer aborts the daemon, and after a reorganisation or
-reindexthe wallet's own unmined sends are offered to the mempool again, so the running wallet shows the same balance as after a restart. Shielded sends check their anchors and witnesses before proving.backupwalletnow copies to a temporary file, syncs, compares SHA-256 and only then renames into place; it works without-exportdir. - Node reliability. Sync from block 0 works.
-reindexno longer silently drops blocks after a stale record header;invalidateblockis reliable across restarts. Peer address selection no longer stalls, and a fresh node holds its outbound peers. Interrupted snapshot downloads resume, with every kept file re-verified. New-mempoolmaxbytes(default 300 MB). - Faster peer connections. A fresh node dials several addresses at once, gives an address it has never reached a shorter first-contact timeout (2 seconds; later attempts use
-timeout), uses the built-in fixed seeds as soon as its address table is thin, and no longer waits on a DNS seed that stopped resolving (dnsseed.zslp.orgis removed). Measured on a Windows home PC, median of 5 fresh starts: 3 peers in 0.75 s (was 8.6 s), 5 peers in 6.4 s (was 35 s or never), first block in 0.69 s (was 2.02 s). Your results depend on your network. - Fast sync from several snapshot servers at once. The snapshot download spreads its streams over every snapshot server it knows, takes identical files from whichever server has them, hashes each file while it downloads, keeps going when a server goes away, and a node killed while the snapshot is being installed now recovers by itself on the next start.
getbootstrapinfogainsserversandservers_in_usewhile several servers are used, and the wallet shows the number of servers next to the download speed. Every byte is still verified against the trusted snapshot manifest, exactly as before; the snapshot protocol is unchanged.- The node now has four built-in snapshot servers (two more than before), each also reachable as a Tor v3 onion service for onion-only nodes.
- Wallet: smaller and safer. Notes that were spent long ago no longer get their witnesses updated or rewritten after every block (a spent note record is about 4.6 KiB instead of about 77 KiB), which makes the wallet file smaller and each block cheaper to process.
z_sendmanytakes an optionalidempotency_key; with it the wallet keeps a durable send journal, so a send you retry after a timeout, a crash or a restart returns the first operation and cannot pay twice.z_listsendjournallists the journal, see doc/send-journal.md.getwalletinfohas ahealthobject (checkpoint, witnesses, pending sends) and the wallet shows a one-line "Wallet needs attention" notice only when something in it needs attention. - Wallet save point after a catch-up. After a snapshot sync or a long time offline, the wallet now saves its progress as soon as the node has caught up instead of up to an hour later, so a crash in that hour no longer repeats the catch-up and the wallet no longer shows "Wallet needs attention" when nothing is wrong. If saving the wallet file ever fails, the wallet says so.
- Wallet start, close and first run. On Windows the wallet starts its node at once instead of first waiting on two refused connection attempts (about 8 seconds on a typical PC). Closing shows "Closing: stopping the ZClassic node safely…" immediately and keeps asking the node to stop until it has heard it. A first launch no longer starts the node twice. A separate-profile folder that does not exist now gets a clear message instead of a silent exit.
- Node speed and limits. A cache of verified Sapling shielded transactions lets a block skip the proof checks it already did when the transaction entered the mempool (
-maxshieldedcachesize, default 4 MiB,0disables); header Equihash solutions are checked on several threads (-headerprecheck, default on); LevelDB open-file limits follow the system file-descriptor limit (override with-dbmaxopenfiles, shown by-help-debug); after a clean shutdown the next start checks only the last 6 blocks, while a crash, a kill or any change to the block or database files runs the usual full check (-cleanshutdownmarker, default on). None of these changes which blocks or transactions are accepted. - Node diagnostics.
getrpcinfolists the RPC calls in flight and per-method counts and times (method names only, never parameters); calls slower than-rpcslowlog=<ms>(default 1000,0disables) are logged as a warning. When the RPC server is overloaded it answers HTTP 503 with a message that names-rpcworkqueueand-rpcthreadsinstead of dropping the call. On Linux and macOS, the error for a data directory that another process already holds now names that process.-helpnow lists-equihashimpl,-finalizationdelay,-finalizationminpeers,-finalizationrequirepeers,-maxreorgdepth,-paramcache,-parkdeepreorgand-uacomment. - Tor (off by default). Optional embedded Tor (
-embeddedtor=1) with fixed onion v3 handling. - ZCL in RPC help. Amounts in RPC help and error text are given in ZCL, and node messages about the zk-SNARK parameter files use the same wording as the rest of the node. A
SECURITY.mdexplains how to report a vulnerability privately. - Tor fail-soft. If the embedded Tor (
-embeddedtor=1) exits while the node runs, the node no longer shuts down: it keeps validating and serving its clearnet peers, and restarts Tor in place after 5, 15, 45, 135 seconds and then every 5 minutes. While Tor is down, the proxies it served point at a fence the node refuses to connect to, so those connections fail closed and nothing goes out in clear (an onion-only node has no peers until Tor is back); the onion address is unchanged after a restart. An in-process Tor is restarted at most three times per process.getnetworkinforeportstorstatus. - Persistent ban list. Bans (
setban, misbehaving peers) are kept inbanlist.datand survive a restart; expired bans are dropped at start-up. A damagedbanlist.datis set aside asbanlist.dat.badand the list starts empty. A damagedpeers.datis no longer discarded whole: it is kept aspeers.dat.corruptand the complete address records it still holds are reused. - Lazy proving parameters, faster start. The node no longer hashes the 692 MiB Sprout parameter file at start-up, and no longer keeps the 51.6 MB of raw Sapling proving bytes in memory: it checks that the Sprout file opens and has the right size, hashes the two Sapling files without keeping them, and verifies the proving files again (against the same digests) when the first proof needs them. The SHA-256 check of the parameter files is still skipped for files the verified-parameter cache (
-paramcache) already vouches for.ZCL_PARAMS_VERIFY=bootrestores hashing every file at start-up. - Lower memory for the block index. The Equihash solutions of block-index entries loaded at start-up stay in the block-index database and are read when a header is served, which saves about 2 GB of memory on a mainnet node (4.26 GB to 2.28 GB measured after the index load); headers served by
getblockheaderand to peers are unchanged. getnodestatus. A new cheap RPC that answers "is this node synced" honestly: after a snapshot installgetblockchaininfocan say synced while the node is still behind its peers, sogetnodestatusalso looks at the tip age and the height the peers reported, and returns the state with reasons, blocks behind, peers, mempool, indexes and free disk space.- Faster signature verification. libsecp256k1 is built with its endomorphism optimisation, which speeds up ECDSA verification and public-key recovery; results are identical.
- ARM CRC32C. On 64-bit ARM (Apple Silicon, Linux aarch64) LevelDB uses the hardware CRC32C instructions, about 6.7 times faster than the portable code in a 64 MiB measurement, with a start-up self-check that falls back to the portable code.
- Updated components. OpenSSL 3.5.9.
- Windows. Non-ASCII data directories work, and the node shuts down cleanly when its console is closed.
debug.logis off by default; use-debuglogfileto enable it.
For developers
The source tree has no Python and no hosted CI: the developer tools are C (zcl-dev, zcl-lint, jsonq, a git-hook agent guard and an MCP server), and release candidates are built and tested on the project's own machines. Contributor and agent rules are in AGENTS.md; the guides for agents and the commands that read them are in doc/agent/.
Upgrading
Close your old wallet normally, keep your wallet and data directory, and open the new one. A wallet backup first is recommended; do not delete or reindex your data directory. Beta13 reads existing wallet.dat files.
Downloads
- Windows x86_64 wallet:
zclwallet-v2.1.2-beta13-win64.exe, a single file with the node embedded. No installer; run it. - Linux x86_64 wallet:
zclwallet-v2.1.2-beta13-linux-x86_64(glibc 2.31+, system X11/XCB/font libraries). - macOS Apple Silicon wallet:
zclwallet-v2.1.2-beta13-macos-arm64.dmg(macOS 14 or later; tested natively on macOS 26). Intel Macs are not supplied. - Command-line node (
zclassicd,zclassic-cli,zclassic-tx):zclassicd-v2.1.2-beta13-win64.zip,zclassicd-v2.1.2-beta13-linux-x86_64.tar.gz,zclassicd-v2.1.2-beta13-macos-arm64.tar.gz.
Verify a download against SHA256SUMS.txt (for example sha256sum -c SHA256SUMS.txt --ignore-missing, or Get-FileHash on Windows). BUILDINFO.txt records the source commit, toolchains and checks for each file.
Unsigned builds
These builds were made on our own machines, not on GitHub Actions, and are not code-signed. Windows SmartScreen may warn: choose "More info", then "Run anyway". The macOS wallet and node are ad-hoc signed, not Apple-notarized: Gatekeeper may refuse the first launch; right-click the app and choose Open, or allow it under System Settings, Privacy & Security.
Known issues
- A pruned node cannot reorganise onto a reconsidered fork that is more than 288 blocks heavier (the undo data is gone).
- New Sprout PHGR13 proofs can no longer be created (see above); existing ones are still verified.
- Windows 7 and Windows 10 older than 1903 are untested; 32-bit targets no longer build.
- Tor is pinned to the
RhettCreighton/torfork with patches carried independs; it stays off by default. - Going back to an older wallet (beta12 or before) after a wallet that has spent many shielded notes ran on this version is safe: balances and notes stay correct, also after importing a key with rescan. One limit: if the first thing the older wallet does is to roll back four or more blocks (an orphaned tip or
invalidateblock) and you then import a key with rescan, it counts old spent notes as unspent until it runs once with-rescanor this version opens the wallet again (which repairs it by itself). Nothing is spent twice. Normal running is faster by the share of the wallet that is spent (those notes are no longer updated per block or rewritten); every unspent note is still written after each block.
Verification
The release files are built from one source commit, recorded in BUILDINFO.txt on the release page and in SOURCE_COMMIT.txt inside each node archive. The work up to PR161 was verified on Linux, macOS arm64 and real Windows with the unit suites, the golden corpus, a real mainnet block replay (3,327,490 blocks, 0 failures), kill-and-verify cycles with no spent-note mismatches, reorganisation and invalidateblock tests, a node synced from block 0 identical to a snapshot-synced node (same best block and UTXO-set hash at height 3,321,763), and a 3-hour mainnet soak with flat memory. The work added after it (PR162: connections, fast sync from several servers, the wallet changes, the node changes and the tooling) was reviewed change by change by a reviewer that is not its author and verified with the unit suites and the shell tests on Linux, macOS arm64 and real Windows; that pull request records the results.
For the release files: each node binary is hardening-checked (PIE/RELRO/NX/canary on Linux, DYNAMIC_BASE/NX_COMPAT/HIGH_ENTROPY_VA on Windows, PIE on macOS); the Windows wallet imports only Windows system DLLs and passes the embedded-daemon extraction check against the node hash; the macOS disk image is verified and mounted read-only with a valid ad-hoc signature; the Linux wallet passes extraction and HTTPS checks. BUILDINFO.txt records the toolchains and the checks run on these exact files. No funded public-mainnet send test and no universal sync-time claim is made.