Skip to content

Common Foundry v1.0.11

Choose a tag to compare

@JustAResearcher JustAResearcher released this 06 Oct 14:08

Common Foundry v1.0.11

Storage update for node operators: opt-in proof pruning. Nothing changes unless you turn it on. Network, launch plan, beacon round, wallets, consensus and monetary policy are unchanged; keep your chain data and wallet.

What changes

Proof pruning (opt-in)

  • cmfd-node run --prune-keep-blocks <N> (N of at least 288) keeps full blocks only for the newest N blocks. Older blocks keep their header, coinbase and transactions but drop their roughly 12 MB proof, so each takes a few KB.
  • Balances, wallet history, the explorer, deposits and verbose getblock are unchanged. Raw getblock <hash> 0 on a pruned block returns block_proof_pruned; getblockchaininfo reports pruned and pruneheight.
  • A pruned node does not serve old blocks to peers and cannot follow a reorganization deeper than its window. The first prune happens about N blocks after the flag is first turned on, and pruning cannot be undone.
  • Back up prune-anchor-<height>.bin together with blocks.log; a pruned node does not start without it. Details: docs/proof-pruning.md.

Packages

Package File
Windows runtime (node + wallet) commonfoundry-mainnet-runtime-windows-x86_64-v1.0.11.zip
Windows miner commonfoundry-mainnet-miner-windows-x86_64-v1.0.11.zip
Linux runtime (node + wallet) commonfoundry-mainnet-runtime-linux-x86_64-v1.0.11.tar.gz
Linux miner commonfoundry-mainnet-miner-linux-x86_64-v1.0.11.tar.gz
HiveOS commonfoundry-mainnet-hiveos-1.0.11.tar.gz

HiveOS flight sheet custom URL: https://github.com/JustAResearcher/CommonFoundry-Binaries/releases/download/v1.0.11/commonfoundry-mainnet-hiveos-1.0.11.tar.gz

Upgrading: extract the new archive to a new folder. Desktop wallet data lives in %LOCALAPPDATA%\org.commonfoundry.wallet.mainnet and is picked up automatically; headless START-NODE.bat users keep their data-mainnet folder. Copy the old folder's production-v4 directory into the new one to skip re-downloading the model files.

Pool operators: if you upgrade, update expected_node_sha256, expected_launch_sha256 and expected_proof_worker_sha256 in your pool config to the values in the new package's MAINNET-PACKAGE.json. The dashboard, replay worker and CUDA runtime are unchanged.

Verification

SHA256SUMS.txt is signed by the same release key as v1.0.0–v1.0.10 (commonfoundry-mainnet-owner, SHA256:cA1Tsf8hL/pxDV5WpOE3iPW3b4uDQosu1dOh//4a/fk), policy MAINNET-RELEASE.allowed_signers SHA-256 9c5be92681092801d89687823837823d0966055c03a84e64551e26387da179d2. Follow docs/mainnet-release-verification.md.

All five archives were built twice from clean and are byte-identical, as are every native binary and the proof worker. The owner approved the unchanged mainnet plan for this source (approval manifest efc5a504…).

Qualification limits: CI run 37471625690 on the final commit was in progress at signing and was not used as a gate (explorer completed/failure; pool-dashboard completed/failure; rust-base in_progress/; rust-proof in_progress/; rust-proof in_progress/; wallet completed/failure). The new proof worker was checked on an RTX 4090 against the preserved RCNet-1 fixture: replay traces identical, one-shot proof with the worker's complete self-verification, and server-mode proofs with and without the preload (17.108 s and 11.22 s request-to-done); all three proofs passed the independent verifier and agree up to the grinding offset. The RTX 50-series path was built but not separately re-qualified. The console-interrupt shutdown tests cannot run in the build harness; they passed in CI's node-shutdown jobs on Windows and Ubuntu. The explorer, pool-dashboard and wallet CI jobs failed only at npm audit, which now flags a new advisory for source-map-js (GHSA-68fv-2mgg-jv7q); it is a build-time development dependency, unchanged since v1.0.10, and the shipped dashboard files are identical to v1.0.10. Pruning was tested on Windows and Linux and end to end on a development network with the real node (startup and background pruning under concurrent reads, following a full peer, refusing a fresh peer, offline storage inspection, restarts, and turning pruning off); the first mainnet prune happens on the relays after this release.