Files
codebase-memory-mcp/.github/workflows/codeql.yml
T
Martin Vogel a545a0aa8c ci: fail apt-get update on any fetch error
apt-get update exits 0 when index fetches fail: it only prints
"W: Some index files failed to download. They have been ignored, or
old ones used instead." and the job breaks later at install time with
a misleading "Unable to locate package", or installs from stale lists.
Add --error-on=any to every apt-get update this branch hardens, placed
after the -o options as in test-infrastructure/Dockerfile.msan, so the
update step itself fails when an index cannot be fetched.

Proof in ubuntu:24.04 (apt 2.8.3) and ubuntu:22.04 (apt 2.4.14)
containers with a runner-shaped mirror+file:/etc/apt/apt-mirrors.txt
source and the same -o options CI passes:
- every mirror unreachable: exit 100 with the flag (E: lines), exit 0
  without it (W: lines, zero Packages lists), on both apt versions
- first mirror refused or stalled (30 s connect timeout) and the real
  archive second: the Ign: failover lines do not trip the flag, exit 0
  with every index fetched (measured with Acquire::Retries=0)
- the image sources with the flag: exit 0 and no Ign/W/E lines, so the
  option is understood by apt 2.4 and 2.8
- green hosted-runner logs (ubuntu-24.04 and ubuntu-22.04 legs) show no
  Ign/Err/W/E lines in the update step

With Acquire::Retries=3 a mirror failover leaves apt idle after the
InRelease failover, with or without the flag and also for a plain
apt-get update; that apt behaviour predates this change and is
unchanged by it.

Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
2026-11-01 10:08:28 +01:00

196 lines
10 KiB
YAML

name: CodeQL SAST
# One analysis per language: c-cpp, javascript-typescript (graph-ui, pkg/npm),
# python, actions. Scope: .github/codeql/codeql-config.yml (first-party source).
#
# FULL scan -- push to main, weekly, dry run, release: every language; c-cpp
# traces the whole production build, vendored C included.
# PR scan -- pr.yml's `codeql` job (light: true), needed by ci-ok. One leg per
# language whose scope the PR changes (codeql-* lanes of
# scripts/ci/select-lanes.sh; an undeterminable diff or a change to the
# workflows, the build or the CodeQL machinery selects every language).
# Each leg analyses its language's WHOLE first-party scope, so cross-file
# data flow is kept; c-cpp compiles the vendored objects before tracing
# starts, except vendored files the PR changes. Diff-informed analysis then
# restricts the REPORTED alerts to the changed lines; what is extracted and
# analysed is never narrowed by the diff.
# There is no pull_request trigger and no job anywhere that polls for a CodeQL
# run: the former PR gate waited up to 150 min per PR for a full analysis and,
# with ~30 PRs pushed at once, starved PR CI for hours.
on:
push:
branches: [main]
schedule:
- cron: "23 5 * * 1"
workflow_dispatch:
inputs:
light:
description: "PR-style c-cpp build (vendored objects compiled untraced) -- to time it"
type: boolean
default: false
workflow_call:
inputs:
light:
description: "PR scan: vendored C compiled untraced except changed files; diff-informed results"
type: boolean
default: false
matrix:
description: 'The legs to run, {"include":[...]} (pr.yml: the selected codeql-* lanes); empty = every language'
type: string
default: ""
permissions:
contents: read
jobs:
analyze:
runs-on: ubuntu-latest
strategy:
fail-fast: false
# The same legs (and keys) everywhere, so a PR's analysis and main's
# share one code-scanning configuration per language.
matrix: ${{ fromJSON(inputs.matrix || '{"include":[{"language":"c-cpp","build-mode":"manual","lane":"codeql-cpp"},{"language":"javascript-typescript","build-mode":"none","lane":"codeql-js"},{"language":"python","build-mode":"none","lane":"codeql-py"},{"language":"actions","build-mode":"none","lane":"codeql-actions"}]}') }}
# Superseded runs are always cancelled: a newer push to main cancels the
# analysis of an older main commit (batched merges -> only the newest main
# commit is scanned), a newer push to a PR cancels that PR's scan. Every
# other trigger gets a group of its own and is never cancelled: a dry run
# or a release tests ONE commit, and two dispatched together must not
# cancel each other's gate. Per language, so legs never cancel each other.
# In a called workflow `github` is the CALLER's context: event_name is
# pull_request for pr.yml, github.workflow names the dry run or release.
concurrency:
group: codeql-${{ matrix.language }}-${{ github.event_name == 'push' && github.ref || github.event_name == 'pull_request' && format('pr-{0}', github.event.pull_request.number) || format('{0}-{1}', github.workflow, github.run_id) }}
cancel-in-progress: ${{ github.event_name == 'push' || github.event_name == 'pull_request' }}
# CodeQL needs security-events: write to upload results (and read for the
# alert check); scoped to this job so the workflow's top-level token stays
# read-only (Scorecard TokenPermissions).
permissions:
security-events: write
actions: read
contents: read
env:
# One code-scanning configuration for every trigger. The action derives
# the key from the RUN's workflow file, which for a called workflow is
# the caller: pr.yml / dry-run.yml / release.yml would each open a
# configuration of their own, and on main an alert fixed by a later push
# would stay open in a stale one. The value is the key the push-to-main
# analysis has always used (the action documents this variable as
# settable for non-standard setups).
CODEQL_ACTION_ANALYSIS_KEY: .github/workflows/codeql.yml:analyze
# Diff-informed analysis: codeql-action v4.38.1 (CLI 2.27.0) reports a
# pull request's alerts only on its changed lines when the
# diff_informed_queries feature is on (default on; this variable forces
# it, src/feature-flags.ts). It applies only when the event payload is a
# pull request -- pr.yml's call -- and is inert on main, dry run, release.
CODEQL_ACTION_DIFF_INFORMED_QUERIES: ${{ inputs.light && 'true' || '' }}
# No compiler cache: a cache hit skips the compiler, and CodeQL extracts
# only what a traced compiler invocation actually compiles.
CBM_NO_CCACHE: "1"
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
# The PR scan reads the merge commit's first parent (the base) to
# find the vendored files the PR changes.
fetch-depth: ${{ inputs.light && 2 || 1 }}
- name: Install build dependencies
if: ${{ matrix.language == 'c-cpp' }}
timeout-minutes: 10
run: sudo apt-get -o Acquire::Retries=3 -o Acquire::http::Timeout=30 -o Acquire::https::Timeout=30 --error-on=any update && sudo apt-get -o Acquire::Retries=3 -o Acquire::http::Timeout=30 -o Acquire::https::Timeout=30 install -y zlib1g-dev
# PR c-cpp: compile the third-party objects (tree-sitter grammars,
# sqlite, mimalloc, lz4, zstd, the tree-sitter runtime: ~1.3 GB of
# vendored C against ~45 MB first-party) BEFORE tracing starts, so the
# traced build below compiles first-party code only ...
- name: Build third-party objects (PR scan, untraced)
if: ${{ matrix.language == 'c-cpp' && inputs.light }}
run: scripts/build.sh --vendored-only
# ... except the vendored files this PR changes: those are rebuilt
# under tracing (an undeterminable diff leaves the list empty; the scan
# still covers all first-party C).
- name: Vendored files this PR changes (traced too)
if: ${{ matrix.language == 'c-cpp' && inputs.light }}
run: scripts/ci/codeql-retrace-vendored.sh "$RUNNER_TEMP/codeql-retrace.txt"
- name: Initialize CodeQL
uses: github/codeql-action/init@1c5b675653bb5c22dbe9b12b556ec555138e09fd # v4.38.1
with:
languages: ${{ matrix.language }}
build-mode: ${{ matrix.build-mode }}
config-file: ./.github/codeql/codeql-config.yml
- name: Build for CodeQL analysis (full)
if: ${{ matrix.language == 'c-cpp' && !inputs.light }}
run: scripts/build.sh
- name: Build for CodeQL analysis (PR scan, first-party + changed vendored)
if: ${{ matrix.language == 'c-cpp' && inputs.light }}
run: scripts/build.sh --reuse-vendored --retrace-from "$RUNNER_TEMP/codeql-retrace.txt"
- name: Perform CodeQL Analysis
uses: github/codeql-action/analyze@1c5b675653bb5c22dbe9b12b556ec555138e09fd # v4.38.1
with:
category: "/language:${{ matrix.language }}"
# Alert policy: the default branch carries ZERO open code-scanning
# alerts. Any open alert on the analysed ref fails this leg: on main the
# commit turns red and GitHub notifies whoever triggered the run; the
# dry run and the release cannot pass; a PR (refs/pull/<n>/merge: what
# its changed lines introduce) fails ci-ok. With a zero baseline "any
# open alert" is exactly "a new alert"; a dismissal (with a reason, in
# the Security tab) holds on every ref. Every leg checks after its own
# upload, so each language's alerts are seen by at least its own leg.
# Moved from _security.yml's former PR-CI codeql-gate job.
- name: Check for open code scanning alerts
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
ALERT_REF: ${{ github.ref }}
run: |
# FAIL CLOSED. The previous form was `gh api ... 2>/dev/null || echo "0"`,
# which looks defensive and is the opposite. gh writes the API error
# BODY to stdout, so a 403 made ALERTS the string
# {"message":"Resource not accessible by integration",...,"status":"403"}0
# every `[` comparison then failed as non-integer, the `if` took its
# false branch, and the step printed "CodeQL gate passed (0 alerts)"
# and exited 0. That is exactly how a security gate should NEVER fail:
# silently, in the direction of "everything is fine". An unreadable
# alert list is now an error, not a zero.
# NOTE: this runs in a command substitution, so it must `return`, not
# `exit` -- an `exit` here would only end the subshell and let the
# caller continue with an empty count, recreating the very bug this
# replaces. The callers below check the status explicitly.
count_open_alerts() {
local body status
body=$(gh api "repos/${{ github.repository }}/code-scanning/alerts?state=open&ref=$ALERT_REF" \
--jq 'length' 2>/dev/null)
status=$?
if [ $status -ne 0 ]; then
echo "BLOCKED: cannot read code scanning alerts (gh exit $status)." >&2
echo " Refusing to treat an unreadable alert list as zero." >&2
echo " Most likely cause: the job lacks security-events:read." >&2
return 1
fi
case "$body" in
''|*[!0-9]*)
echo "BLOCKED: alert count was not a number: ${body:-<empty>}" >&2
return 1
;;
esac
printf '%s' "$body"
}
echo "Waiting 60s for alert API to settle..."
sleep 60
ALERTS=$(count_open_alerts) || exit 1
sleep 15
ALERTS2=$(count_open_alerts) || exit 1
[ "$ALERTS" -lt "$ALERTS2" ] && ALERTS=$ALERTS2
if [ "$ALERTS" -gt 0 ]; then
echo "::error title=Open code scanning alerts::$ALERTS open alert(s) on $ALERT_REF -- fix, or dismiss with a reason in the Security tab"
echo "BLOCKED: $ALERTS open alert(s)"
gh api "repos/${{ github.repository }}/code-scanning/alerts?state=open&ref=$ALERT_REF" \
--jq '.[] | " #\(.number) [\(.rule.security_severity_level // .rule.severity)] \(.rule.id) — \(.most_recent_instance.location.path):\(.most_recent_instance.location.start_line)"' 2>/dev/null || true
exit 1
fi
echo "=== CodeQL gate passed (0 alerts) ==="