soleprint/berth/ctrl/selftest.sh

#!/usr/bin/env bash
# What berth has settled, and what it has withdrawn, written down as assertions.
#
# Two halves:
#   - decisions that hold. Failing one means "you are about to undo this".
#   - every entry in ../STALE.md. Failing one means a withdrawn assumption came
#     back. That is the half that makes STALE.md an audit surface and not an
#     archive — a retraction nobody re-reads is a retraction that decays.
#
# Scope: no cloud, no ssh, no sudo, no network. Cheap enough to actually run.
# `make check` reports on the world and never fails; this exits 1, like rig's.
#
# Usage: make selftest   (or: bash ctrl/selftest.sh)
set -uo pipefail          # NOT -e: one failing check must not abort the rest
cd "$(dirname "$0")"

source ./lib/config.sh

rc=0
passed=0
check() {   # name, expected, actual
    if [ "$2" = "$3" ]; then
        printf '  ok    %s\n' "$1"
        passed=$((passed + 1))
    else
        printf '  FAIL  %s\n        expected: %s\n        got:      %s\n' "$1" "$2" "$3"
        rc=1
    fi
}
note() { printf '\n%s\n' "$1"; }
skip() { printf '  skip  %s (%s)\n' "$1" "$2"; }

# An absence check must not match the files that RECORD the absence. STALE.md
# names every withdrawn thing by definition, and this file names them again to
# assert them — so both are excluded, or every check fails on itself. rig hits
# the same wall and assembles its pattern from fragments for the same reason.
NOSELF="--exclude=selftest.sh --exclude=STALE.md"
absent() { grep -rIl $NOSELF "$@" 2>/dev/null | wc -l; }

# A throwaway estate, for the checks that have to run berth rather than read it.
TMP_ESTATE=_selftest
cleanup() { rm -f "../estate/${TMP_ESTATE}.json"; }
trap cleanup EXIT

note "the withdrawn assumptions — ../STALE.md, one check each"

# B1 — Pulumi. The two surviving mentions are historical fact about ppl/infra
# and live in README.md and the estate, not in anything that runs.
check "B1  no pulumi in the code" "0" "$(absent -i pulumi . ../Makefile)"

# B2 — the ctlptl precedent. Withdrawn; the argument stands on its own now.
check "B2  the withdrawn precedent is cited nowhere" "0" "$(absent -i ctlptl ..)"

# B3 — `wg show <if> dump` leaks the private key in field 1. Stating only
# show-vs-showconf makes the dump form read as safe.
check "B3  all three wg forms are named" "yes" \
    "$(grep -q 'dump' vpn.sh && grep -q 'showconf' vpn.sh && echo yes || echo no)"
check "B3  capture refuses showconf-shaped input" "1" \
    "$(printf '[Interface]\nPrivateKey = x\n' | bash vpn.sh capture >/dev/null 2>&1; echo $?)"
check "B3  capture refuses dump-shaped input" "1" \
    "$(printf 'priv\tpub\t51820\toff\n' | bash vpn.sh capture >/dev/null 2>&1; echo $?)"

# B4 — keepalive belongs to the peer that DIALS, not the one that roams. The
# first version warned on a correctly configured overlay, so the check is run
# against one: hub carries the keepalive, nrft roams.
python3 - <<'PY'
import json, collections
d = json.load(open("../estate/mcrn.json"), object_pairs_hook=collections.OrderedDict)
d["vpn"]["overlays"]["estate"]["peers"]["box"]["keepalive"] = 25
json.dump(d, open("../estate/_selftest.json", "w"), indent=2, ensure_ascii=False)
PY
check "B4  a correct overlay raises no keepalive warning" "0" \
    "$(ESTATE=$TMP_ESTATE bash vpn.sh check 2>/dev/null | grep -ci 'no peer entry carries')"

# B6 — public keys are 44-char base64 too, so shape alone would flag correct
# data. Same fixture, with a real-shaped public key on a peer.
python3 - <<'PY'
import base64, collections, json, os
d = json.load(open("../estate/_selftest.json"), object_pairs_hook=collections.OrderedDict)
d["vpn"]["overlays"]["estate"]["peers"]["box"]["public_key"] = base64.b64encode(os.urandom(32)).decode()
json.dump(d, open("../estate/_selftest.json", "w"), indent=2, ensure_ascii=False)
PY
check "B6  a public key does not trip the secret check" "0" \
    "$(ESTATE=$TMP_ESTATE bash vpn.sh check 2>/dev/null | grep -c 'FAIL.*key')"

cleanup   # the fixture is done with; two estate files would make load_config
          # refuse to guess below, which is right but reads as a config failure

# B5 — the pass-through block must be LAST, or a subcommand that names a real
# target runs that target too. Checked through make, not by reading the file.
note "B5  a subcommand that names a target dispatches once"
for combo in "host ports" "host services" "vpn check" "vpn show estate" "estate show"; do
    check "    make $combo" "1" \
        "$(cd .. && make -n $combo 2>/dev/null | grep -c 'bash ctrl/')"
done

# B7 — the overlay moved out of network.wireguard into a top-level vpn block.
check "B7  nothing reads network.wireguard" "0" "$(absent 'network\.wireguard' .)"

# B8 — peers, not relatives. berth sources nothing from rig.
check "B8  berth sources nothing from rig" "0" "$(absent -E 'rig/ctrl|\.\./rig' .)"

# B9 — langfuse was filed as an exception a template could not express. It was
# the general case. The proof is a live route: render it and diff against the
# hand-written file, normalised for comments and whitespace.
LIVE=/home/mariano/wdir/semester/ppl/gateway/nginx/conf.d/langfuse.conf
if [ -f "$LIVE" ]; then
    norm() { sed -e 's/#.*//' -e 's/[[:space:]]\+/ /g' -e 's/^ //' -e 's/ $//' -e '/^$/d' "$1"; }
    bash services.sh render aws >/dev/null 2>&1
    check "B9  the generated vhost reproduces the live one" "same" \
        "$(diff -q <(norm ./render/out/aws/langfuse.conf) <(norm "$LIVE") >/dev/null 2>&1 \
            && echo same || echo different)"
else
    skip "B9  generated vhost matches the live one" "ppl not on this machine"
fi

note "the safety contract — berth's verbs are not all safe"

check "estate defaults to show"   "show"   "$(cd .. && make -n estate 2>/dev/null | grep -oE 'estate\.sh [a-z]+' | awk '{print $2}')"
check "certs defaults to status"  "status" "$(cd .. && make -n certs  2>/dev/null | grep -oE 'certs\.sh [a-z]+'  | awk '{print $2}')"
check "dns defaults to list"      "list"   "$(cd .. && make -n dns    2>/dev/null | grep -oE 'dns\.sh [a-z]+'    | awk '{print $2}')"
check "vpn defaults to list"      "list"   "$(cd .. && make -n vpn    2>/dev/null | grep -oE 'vpn\.sh [a-z]+'    | awk '{print $2}')"

for verb in apply destroy; do
    check "estate $verb refuses without --yes" "1" \
        "$(bash estate.sh "$verb" >/dev/null 2>&1; echo $?)"
done
for verb in renew push; do
    check "certs $verb refuses"                "1" \
        "$(bash certs.sh "$verb" >/dev/null 2>&1; echo $?)"
done
check "dns add refuses to change live DNS" "1" \
    "$(bash dns.sh add selftest >/dev/null 2>&1; echo $?)"
check "vpn up refuses without --yes" "1" \
    "$(bash vpn.sh up >/dev/null 2>&1; echo $?)"

note "config — the caller's env beats the files"

# Generated from CONFIG_OVERRIDABLE, so a new key enrols itself.
test_value() {
    case "$1" in
        TARGET)  echo "gcp" ;;
        ESTATE)  echo "mcrn" ;;
        *)       echo "selftest-sentinel" ;;
    esac
}
for key in $CONFIG_OVERRIDABLE; do
    want="$(test_value "$key")"
    got="$(export "$key=$want"; load_config >/dev/null 2>&1; echo "${!key}")"
    check "  caller's $key wins" "$want" "$got"
done

note "rig agreement — recomputed, never imported"

# rig pins these same constants in its own selftest. Both arrive at them from
# the same formula with no shared code, which is the coupling rule made testable.
check "derive_port_base rig"     "20310" "$(derive_port_base rig)"
check "derive_port_base foo"     "21690" "$(derive_port_base foo)"
check "derive_port_base my-proj" "21030" "$(derive_port_base my-proj)"

note "containment — berth writes nothing outside berth/"

check "no tracked change outside berth/" "0" \
    "$(cd ../.. && git status --porcelain 2>/dev/null | grep -vc '^.. berth/')"
check "generated output is ignored" "yes" \
    "$(cd .. && git check-ignore -q ctrl/render/out && echo yes || echo no)"
# A trailing-slash pattern matches directories only, so ask about a path
# inside it rather than the (not-yet-existing) directory itself.
check "key material is ignored" "yes" \
    "$(cd .. && git check-ignore -q ctrl/.secrets/vpn/any.key && echo yes || echo no)"

note "every STALE entry has a check here"

# Not "$0": line 15 cd's into this script's directory, so a relative $0 no
# longer resolves. After the cd the file is simply selftest.sh.
# Ids are counted wherever they appear — B5's sits in a note(), not a check name.
entries="$(grep -c '^\*\*✖ B' ../STALE.md)"
checked="$(grep -oE '\bB[1-9][0-9]?\b' selftest.sh | sort -u | wc -l)"
check "STALE.md entries are all covered" "$entries" "$checked"

printf '\n%d passed' "$passed"
[ "$rc" -ne 0 ] && printf ', SOME FAILED'
printf '\n'
exit "$rc"