Skip to content

v0.23.0rc4 — wasm 964787ece3a98badb85493839ad77a372295ca7b8ded9f87ac8222a616d3edb5

Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 25 Oct 19:11
· 6 commits to main since this release

Overview

v0.23.0rc4 is the fourth release candidate of v0.23.0: v0.23.0rc3 rebuilt on icpp-pro 7.0.0rc4, which fixes a security bug in the Candid decoder. Every wasm built on icpp-pro up to 7.0.0rc3, v0.23.0rc3 included, has the bug, so upgrade those canisters. For apps nothing else changes since rc3. Releases now have release notes.

Highlights

  • Candid decoder security fix: a message that ends early now traps with "Not enough bytes left" instead of being read past its end (#71)
  • Every release now has notes: a summary, the commits since the previous release, then how to reproduce the wasm (#70)

Upgrade notes

  • Upgrade every canister built from v0.23.0rc3 or earlier; an upgrade from rc3 keeps the uploaded models and prompt caches (#71)
  • Replies are smaller where they repeat a Candid type, e.g. a directory listing: same values, different bytes (#71)

icpp-pro

  • icpp-pro 7.0.0rc4: the decoder fix, shared Candid type tables, and linear decoding of a vec of records (#71)

Scripts and tooling

  • Release notes: docs/release-notes/ holds each release's summary, written from scripts/make-release-summary.txt (#70)
  • scripts/make-release-desc.sh <tag> lists the commits since the previous release; release.yml refuses a tag without notes (#70)
  • For contributors, make test-llm-native runs under icpp run-native, icpp-pro's memory guard; scripts/memory_guard.py is gone (#71)

Changelog since v0.23.0rc3

  • 0b2f7da icpp-pro 7.0.0rc4: Candid decoder security fix, icpp run-native, one golden re-pinned (#71)
  • de3acbe Release notes for every release, modelled on upstream llama.cpp (#70)

Reproducible build

llama_cpp.wasm sha256 964787ece3a98badb85493839ad77a372295ca7b8ded9f87ac8222a616d3edb5
llama_cpp_canister commit 4276f526970b7916cc7105c357ad2f2ed018ed37
llama_cpp_onicai_fork commit 702142008552e33844f99dddefc409550166d175
built with Docker, linux/amd64, icpp-pro 7.0.0rc4
built on 2026-10-25

Reproduce this wasm yourself

git clone https://github.com/onicai/llama_cpp_canister.git
cd llama_cpp_canister
git checkout 4276f526970b7916cc7105c357ad2f2ed018ed37
make docker-build-base
make docker-build-wasm

The last command prints the sha256. It must equal
964787ece3a98badb85493839ad77a372295ca7b8ded9f87ac8222a616d3edb5.

No Mac and no local toolchain are needed: the llama.cpp fork commit,
icpp-pro, wasi-sdk, Rust and Binaryen are all pinned inside the image.

Check what a deployed canister is running

Anyone can read a deployed module hash, with no special rights, from the
public read-state path:

dfx canister --network ic info <canister-id>

It must print Module hash: 0x964787ece3a98badb85493839ad77a372295ca7b8ded9f87ac8222a616d3edb5.

If you control the canister you can instead build and compare in one step:

make docker-verify-wasm VERIFY_CANISTER=<canister-id>

(That target uses icp canister status, a controller-only management
call; it fails with IC0542 for anyone else.)