ci(fuzz): run cargo-fuzz on nightly and keep crash reproducers

rust-toolchain.toml overrides `rustup default nightly`, so the scheduled
job ran cargo-fuzz on stable and its -Z flags failed (B7). Every cargo
call now names +nightly. Also:

- upload fuzz/artifacts/node/ when the run fails;
- drop the false claim that the corpus "grows in place across
  schedules": new inputs live only in a 7-day artifact until someone
  minimises and commits them;
- pin actions by SHA as ci.yml does (checkout v4.4.0, upload-artifact
  v4.6.2), and install cargo-fuzz with --locked.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-22 08:44:09 -07:00
co-authored by Claude Opus 5.5
parent 7a31edd6ad
commit 749d6f54ce
+37 -20
View File
@@ -1,11 +1,12 @@
# Fuzzing lives on `cargo fuzz`, which needs a nightly toolchain and must never
# be part of the stable `cargo clippy --workspace` gate in ci.yml. This workflow
# is therefore SCHEDULED (not a PR push/PR gate) and runs against the corpus
# that is checked into this repo under `fuzz/corpus/node/`.
# be part of the stable `cargo clippy --workspace` gate in ci.yml (ci.sh only
# checks that the fuzz crate still compiles on stable). This workflow is
# therefore SCHEDULED (not a PR gate) and starts from the corpus checked into
# this repo under `fuzz/corpus/node/`.
#
# The 24 CPU-hour budget mentioned in Task 12 is release-time; this scheduled run
# uses a modest per-run budget so it stays cheap while still re-executing every
# known input and continuing to grow the corpus.
# The 24 CPU-hour release budget (spec §7) is spent by hand before a release;
# this scheduled run uses a modest per-run budget so it stays cheap while still
# re-executing every known input.
name: Fuzz
@@ -19,32 +20,48 @@ on:
permissions:
contents: read
env:
CARGO_TERM_COLOR: always
jobs:
fuzz-node:
runs-on: ubuntu-latest
timeout-minutes: 120
steps:
- uses: actions/checkout@v4
with:
# The corpus is committed, so the run continues where it left off and
# re-runs every known crashing/non-crashing input on each schedule.
fetch-depth: 0
- name: Checkout
uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
# rust-toolchain.toml pins stable, which overrides `rustup default`, so
# every cargo call below names `+nightly` explicitly (cargo-fuzz passes
# `-Z` flags that only nightly accepts).
- name: Install nightly toolchain + cargo-fuzz
run: |
rustup toolchain install nightly --profile minimal --component rust-src
rustup default nightly
cargo install cargo-fuzz
cargo +nightly install cargo-fuzz --locked
- name: Run the Node harness against the checked-in corpus
# Re-runs every corpus entry (fast path) and then continues fuzzing for
# the run budget. The corpus directory is part of the repo, so it grows
# in place across schedules; a fresh checkout still has prior inputs.
run: cargo fuzz run node -- -max_total_time=1800
- name: Run the Node harness from the checked-in corpus
# Re-runs every corpus entry first, then keeps fuzzing for the budget.
# Both input modes (basic and adversarial, chosen by the first input
# byte) run in the one `node` target.
run: cargo +nightly fuzz run node -- -max_total_time=1800
- name: Upload corpus artifacts (optional retention)
# A crash leaves its reproducer under fuzz/artifacts/node/. Reproduce
# with `cargo +nightly fuzz run node <file>`.
- name: Upload crash reproducers
if: failure()
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
with:
name: fuzz-artifacts-node
path: fuzz/artifacts/node/
if-no-files-found: ignore
retention-days: 30
# New inputs found by this run exist only on the runner and in this
# 7-day artifact; nothing commits them back. To keep them, download the
# artifact, minimise with `cargo +nightly fuzz cmin node`, and commit.
- name: Upload grown corpus
if: always()
uses: actions/upload-artifact@v4
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
with:
name: fuzz-corpus-node
path: fuzz/corpus/node/