Files
codebase-memory-mcp/scripts/ci/prepare-release-candidates.sh
T
Martin Vogel 98d7dbab01 feat(ci): add a third release candidate (debug-stripped) as an extra VT draw
Release evidence from run 31744302624 shows the tolerated Microsoft `!ml`
verdict is close to a coin flip per byte image rather than a property of the
code. Across the eight targets the stripped and unstripped candidates of the
SAME linker output disagreed on four, and in both directions:

  linux-amd64    stripped microsoft-ml   unstripped clean
  darwin-arm64   stripped clean          unstripped microsoft-ml
  linux-arm64    stripped clean          unstripped microsoft-ml

If the classifier were keying on something intrinsic to our code the siblings
would agree; they do not. So each variant is close to an independent draw, and
5 of 16 candidates drew the flag.

Two draws is not always enough. On that run linux-amd64-portable came back
microsoft-ml on BOTH candidates, leaving no clean binary to ship for that
target. A third independent draw at a ~31% observed per-candidate hit rate takes
the both/all-flagged case from roughly 1-in-10 per target to roughly 1-in-30.

The third candidate is `--strip-debug` (Apple: `-S`): debug information removed,
symbol table kept. Behaviourally identical to the other two — same linker
output, only metadata differs — but a distinct byte image, which is all
VirusTotal needs to scan it as its own file. Verified on the real v0.10.4
candidates: linux-amd64 gives three distinct hashes (294,634,656 /
294,623,208 / 293,746,096 bytes) and darwin-arm64 likewise, with the ad-hoc
signature verifying after strip.

Selection is unchanged in spirit and now ordered: smallest artifact first
(stripped, debug-stripped, unstripped), take the first CLEAN one, and only if
every candidate drew the tolerated verdict ship the smallest flagged one. A
hard verdict on any candidate still blocks the release before selection.

Cost is 24 objects per release instead of 16.

Derivation enforces that all three hashes differ — identical candidates would be
one draw wearing three hats, and the selector would believe it had alternatives
it does not have. Public claims in README, SECURITY.md and docs/index.html
updated from "both stripped and unstripped" to the three candidates.

All three release contract tests pass, including the native derivation test
which exercises the real strip and codesign path on this host.

Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
2026-08-13 23:52:21 +02:00

272 lines
10 KiB
Bash
Executable File

#!/usr/bin/env bash
# Derive the three final, immutable release candidates for one existing product
# tuple from a single linker output. Strip/signing happens here and nowhere
# after this boundary.
set -euo pipefail
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
usage() {
cat <<'EOF'
Usage: scripts/ci/prepare-release-candidates.sh <goos> <goarch> \
--binary FILE --out-dir DIR
Derive three final executables - unstripped, debug-stripped and stripped - for
exactly one of the eight release product tuples. All three come from the same
linker output
and are finalized before hashing and atomic publication. Runtime testing starts
only after VirusTotal selection, using the selected candidate.
Environment:
STRIP exact strip executable override
CODESIGN exact codesign executable override (Darwin only)
EOF
}
GOOS=""
GOARCH=""
BINARY=""
OUT_DIR=""
expect=""
for arg in "$@"; do
case "$expect" in
binary) BINARY="$arg"; expect=""; continue ;;
out-dir) OUT_DIR="$arg"; expect=""; continue ;;
esac
case "$arg" in
-h | --help) usage; exit 0 ;;
--binary) expect=binary ;;
--binary=*) BINARY="${arg#--binary=}" ;;
--out-dir) expect=out-dir ;;
--out-dir=*) OUT_DIR="${arg#--out-dir=}" ;;
-*) echo "prepare-release-candidates: unknown option '$arg'" >&2; exit 2 ;;
*)
if [ -z "$GOOS" ]; then GOOS="$arg"
elif [ -z "$GOARCH" ]; then GOARCH="$arg"
else echo "prepare-release-candidates: unexpected argument '$arg'" >&2; exit 2
fi
;;
esac
done
[ -z "$expect" ] || { echo "prepare-release-candidates: --$expect needs a value" >&2; exit 2; }
[ -n "$GOOS" ] && [ -n "$GOARCH" ] && [ -n "$BINARY" ] && [ -n "$OUT_DIR" ] || {
usage >&2
exit 2
}
TARGET="$GOOS-$GOARCH"
case "$TARGET" in
linux-amd64 | linux-arm64 | linux-amd64-portable | linux-arm64-portable | \
darwin-amd64 | darwin-arm64 | windows-amd64 | windows-arm64) ;;
*) echo "prepare-release-candidates: unsupported release target '$TARGET'" >&2; exit 2 ;;
esac
[ -f "$BINARY" ] && [ ! -L "$BINARY" ] || {
echo "prepare-release-candidates: input is not a regular non-symlink file: $BINARY" >&2
exit 2
}
command -v python3 >/dev/null 2>&1 || {
echo "prepare-release-candidates: python3 is required" >&2
exit 2
}
BINARY="$(cd "$(dirname "$BINARY")" && pwd -P)/$(basename "$BINARY")"
mkdir -p "$OUT_DIR"
OUT_DIR="$(cd "$OUT_DIR" && pwd -P)"
PUBLISHED="$OUT_DIR/$TARGET"
[ ! -e "$PUBLISHED" ] && [ ! -L "$PUBLISHED" ] || {
echo "prepare-release-candidates: refusing to overwrite $PUBLISHED" >&2
exit 2
}
sha256_file() {
python3 - "$1" <<'PY'
import hashlib
import pathlib
import sys
digest = hashlib.sha256()
with pathlib.Path(sys.argv[1]).open("rb") as stream:
for chunk in iter(lambda: stream.read(1024 * 1024), b""):
digest.update(chunk)
print(digest.hexdigest())
PY
}
one_line() {
LC_ALL=C tr '\t\r\n' ' ' | sed 's/ */ /g; s/^ //; s/ $//'
}
resolve_strip() {
if [ -n "${STRIP:-}" ]; then
[ -x "$STRIP" ] || command -v "$STRIP" >/dev/null 2>&1 || {
echo "prepare-release-candidates: explicit STRIP is not executable: $STRIP" >&2
return 1
}
command -v "$STRIP" 2>/dev/null || printf '%s\n' "$STRIP"
return
fi
local candidate
for candidate in llvm-strip strip; do
if command -v "$candidate" >/dev/null 2>&1; then
command -v "$candidate"
return
fi
done
echo "prepare-release-candidates: no strip tool is available" >&2
return 1
}
tool_version() {
local tool="$1" value=""
value="$({ "$tool" --version || "$tool" -V || true; } 2>&1 | head -n 1 | one_line)"
[ -n "$value" ] || value="$(basename "$tool") (version unavailable)"
printf '%s\n' "$value"
}
STRIP_TOOL="$(resolve_strip)" || exit 2
STRIP_VERSION="$(tool_version "$STRIP_TOOL")"
SOURCE_SHA="$(sha256_file "$BINARY")"
WORK="$(mktemp -d "$OUT_DIR/.${TARGET}.candidates.XXXXXX")"
cleanup() {
local status=$?
trap - EXIT
rm -rf "$WORK"
exit "$status"
}
trap cleanup EXIT
STAGE="$WORK/$TARGET"
BINARY_NAME=codebase-memory-mcp
[ "$GOOS" = windows ] && BINARY_NAME=codebase-memory-mcp.exe
mkdir -p "$STAGE/unstripped" "$STAGE/debug-stripped" "$STAGE/stripped"
UNSTRIPPED="$STAGE/unstripped/$BINARY_NAME"
DEBUG_STRIPPED="$STAGE/debug-stripped/$BINARY_NAME"
STRIPPED="$STAGE/stripped/$BINARY_NAME"
cp "$BINARY" "$UNSTRIPPED"
cp "$BINARY" "$DEBUG_STRIPPED"
cp "$BINARY" "$STRIPPED"
chmod 0755 "$UNSTRIPPED" "$DEBUG_STRIPPED" "$STRIPPED"
STRIP_INVOCATION="$(basename "$STRIP_TOOL") --strip-all"
if ! "$STRIP_TOOL" --strip-all "$STRIPPED" 2>/dev/null; then
# Apple's strip rejects GNU/LLVM --strip-all. Plain Apple strip is the
# measured equivalent; no weaker flag combination is allowed. Restore the
# pristine linker output first because a failed tool may have written.
if [ "$GOOS" != darwin ]; then
echo "prepare-release-candidates: $STRIP_TOOL --strip-all failed" >&2
exit 2
fi
cp "$BINARY" "$STRIPPED"
chmod 0755 "$STRIPPED"
"$STRIP_TOOL" "$STRIPPED"
STRIP_INVOCATION="$(basename "$STRIP_TOOL")"
fi
# Third candidate: debug information removed, symbol table kept. Behaviourally
# identical to the other two — same linker output, nothing but metadata differs
# — but a distinct byte image, so VirusTotal scans it as its own file.
#
# That is the entire point. Release-run evidence shows the single tolerated
# Microsoft `!ml` verdict landing on stripped and unstripped candidates of the
# SAME build essentially at random (they disagreed on 4 of 8 targets, in both
# directions), so each variant is an independent draw. With two candidates one
# target came back flagged on both and had no clean binary to ship; a third
# independent draw makes that outcome substantially rarer.
DEBUG_STRIP_INVOCATION="$(basename "$STRIP_TOOL") --strip-debug"
if ! "$STRIP_TOOL" --strip-debug "$DEBUG_STRIPPED" 2>/dev/null; then
# Apple's strip rejects --strip-debug; -S is its measured equivalent.
if [ "$GOOS" != darwin ]; then
echo "prepare-release-candidates: $STRIP_TOOL --strip-debug failed" >&2
exit 2
fi
cp "$BINARY" "$DEBUG_STRIPPED"
chmod 0755 "$DEBUG_STRIPPED"
"$STRIP_TOOL" -S "$DEBUG_STRIPPED"
DEBUG_STRIP_INVOCATION="$(basename "$STRIP_TOOL") -S"
fi
UNSTRIPPED_PRE_SIGN_SHA="$(sha256_file "$UNSTRIPPED")"
DEBUG_STRIPPED_PRE_SIGN_SHA="$(sha256_file "$DEBUG_STRIPPED")"
STRIPPED_PRE_SIGN_SHA="$(sha256_file "$STRIPPED")"
SIGNATURE=not-applicable
if [ "$GOOS" = darwin ]; then
CODESIGN_TOOL="${CODESIGN:-codesign}"
command -v "$CODESIGN_TOOL" >/dev/null 2>&1 || {
echo "prepare-release-candidates: codesign is required for Darwin candidates" >&2
exit 2
}
CODESIGN_TOOL="$(command -v "$CODESIGN_TOOL")"
for candidate in "$UNSTRIPPED" "$DEBUG_STRIPPED" "$STRIPPED"; do
"$CODESIGN_TOOL" --sign - --force --timestamp=none "$candidate"
"$CODESIGN_TOOL" --verify --strict "$candidate"
done
SIGNATURE=adhoc-verified
fi
UNSTRIPPED_SHA="$(sha256_file "$UNSTRIPPED")"
DEBUG_STRIPPED_SHA="$(sha256_file "$DEBUG_STRIPPED")"
STRIPPED_SHA="$(sha256_file "$STRIPPED")"
# All three must differ: identical candidates would be one draw wearing three
# hats, and the selector would believe it had alternatives it does not have.
if [ "$UNSTRIPPED_SHA" = "$STRIPPED_SHA" ] || \
[ "$UNSTRIPPED_SHA" = "$DEBUG_STRIPPED_SHA" ] || \
[ "$DEBUG_STRIPPED_SHA" = "$STRIPPED_SHA" ]; then
echo "prepare-release-candidates: strip did not produce byte-distinct candidates" >&2
exit 2
fi
for candidate in "$UNSTRIPPED" "$DEBUG_STRIPPED" "$STRIPPED"; do
bash "$ROOT/scripts/ci/check-binary-composition.sh" "$candidate" >/dev/null
done
[ "$(sha256_file "$BINARY")" = "$SOURCE_SHA" ] || {
echo "prepare-release-candidates: linker output changed during derivation" >&2
exit 2
}
[ "$(sha256_file "$UNSTRIPPED")" = "$UNSTRIPPED_SHA" ] && \
[ "$(sha256_file "$DEBUG_STRIPPED")" = "$DEBUG_STRIPPED_SHA" ] && \
[ "$(sha256_file "$STRIPPED")" = "$STRIPPED_SHA" ] || {
echo "prepare-release-candidates: a final candidate changed during validation" >&2
exit 2
}
case "$GOOS" in
linux) FORMAT=elf ;;
darwin) FORMAT=macho ;;
windows) FORMAT=pe ;;
esac
ARCHITECTURE="${GOARCH%-portable}"
case "$TARGET" in
linux-*-portable) LINKAGE=portable ;;
linux-*) LINKAGE=dynamic ;;
*) LINKAGE=native ;;
esac
UNSTRIPPED_SIZE="$(wc -c < "$UNSTRIPPED" | tr -d '[:space:]')"
DEBUG_STRIPPED_SIZE="$(wc -c < "$DEBUG_STRIPPED" | tr -d '[:space:]')"
STRIPPED_SIZE="$(wc -c < "$STRIPPED" | tr -d '[:space:]')"
MANIFEST="$STAGE/candidate-provenance.tsv"
{
printf '# cbm-release-candidate-provenance-v1\n'
printf 'target\tvariant\trelative_path\tsource_sha256\tpre_sign_sha256\tsha256\tsize\tformat\tarchitecture\tlinkage\ttransform\tsignature\tstrip_tool\tstrip_version\tpair_verification\n'
printf '%s\tunstripped\tunstripped/%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\tcopy\t%s\t%s\t%s\tsame-linker-output-v1\n' \
"$TARGET" "$BINARY_NAME" "$SOURCE_SHA" "$UNSTRIPPED_PRE_SIGN_SHA" \
"$UNSTRIPPED_SHA" "$UNSTRIPPED_SIZE" "$FORMAT" "$ARCHITECTURE" \
"$LINKAGE" "$SIGNATURE" "$STRIP_INVOCATION" "$STRIP_VERSION"
printf '%s\tdebug-stripped\tdebug-stripped/%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\tstrip-debug\t%s\t%s\t%s\tsame-linker-output-v1\n' \
"$TARGET" "$BINARY_NAME" "$SOURCE_SHA" "$DEBUG_STRIPPED_PRE_SIGN_SHA" \
"$DEBUG_STRIPPED_SHA" "$DEBUG_STRIPPED_SIZE" "$FORMAT" "$ARCHITECTURE" \
"$LINKAGE" "$SIGNATURE" "$DEBUG_STRIP_INVOCATION" "$STRIP_VERSION"
printf '%s\tstripped\tstripped/%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\tstrip\t%s\t%s\t%s\tsame-linker-output-v1\n' \
"$TARGET" "$BINARY_NAME" "$SOURCE_SHA" "$STRIPPED_PRE_SIGN_SHA" \
"$STRIPPED_SHA" "$STRIPPED_SIZE" "$FORMAT" "$ARCHITECTURE" \
"$LINKAGE" "$SIGNATURE" "$STRIP_INVOCATION" "$STRIP_VERSION"
} > "$MANIFEST"
chmod 0555 "$UNSTRIPPED" "$DEBUG_STRIPPED" "$STRIPPED"
chmod 0444 "$MANIFEST"
mv "$STAGE" "$PUBLISHED"
echo "prepare-release-candidates: published $TARGET unstripped/debug-stripped/stripped candidates"