create-cmp-cli 0.11.0 → 0.13.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +11 -9
- package/bin/create-cmp.mjs +3 -0
- package/package.json +1 -1
- package/src/commands/upgrade.mjs +287 -0
- package/src/lib/harness-upgrade.mjs +364 -0
- package/src/lib/package-name.mjs +72 -0
- package/src/scaffold.mjs +7 -2
- package/template/.claude/settings.json +30 -0
- package/template/CLAUDE.md +51 -6
- package/template/README.md +4 -0
- package/template/composeApp/build.gradle.kts +44 -0
- package/template/composeApp/src/androidDebug/AndroidManifest.xml +9 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/PlatformBehaviorSeamTest.kt +277 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/RuntimeStateSeamTest.kt +308 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/AlarmAsserts.kt +152 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/ConfigControl.kt +124 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/DozeControl.kt +113 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/NetworkControl.kt +137 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/NotificationAsserts.kt +163 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/PermissionControl.kt +132 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/ProcessControl.kt +217 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/Shell.kt +79 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/SystemState.kt +113 -0
- package/template/composeApp/src/androidInstrumentedTest/kotlin/com/example/app/testing/TimeWarp.kt +114 -0
- package/template/composeApp/src/commonMain/kotlin/com/example/app/di/AppModule.kt +5 -2
- package/template/composeApp/src/commonMain/kotlin/com/example/app/presentation/components/AppBottomBar.kt +1 -1
- package/template/composeApp/src/commonMain/kotlin/com/example/app/presentation/components/AppButton.kt +1 -1
- package/template/composeApp/src/commonMain/kotlin/com/example/app/presentation/components/AppIconButton.kt +1 -1
- package/template/composeApp/src/desktopTest/kotlin/com/example/app/conformance/ArchitectureConformanceTest.kt +58 -0
- package/template/docs/ARCHITECTURE.md +41 -2
- package/template/docs/TESTING.md +165 -0
- package/template/gradle/libs.versions.toml +15 -0
- package/template/manifest.json +1 -0
- package/template/qa/evidence/schema.json +20 -2
- package/template/qa/lib/affected-tests.mjs +147 -0
- package/template/qa/lib/audit-cadence.mjs +290 -0
- package/template/qa/lib/determinism.mjs +179 -0
- package/template/qa/lib/device-lease.mjs +249 -0
- package/template/qa/lib/evidence-badge.mjs +158 -0
- package/template/qa/lib/evidence-level.mjs +117 -0
- package/template/qa/lib/flight-recorder.mjs +332 -0
- package/template/qa/lib/inputs-hash.mjs +16 -1
- package/template/qa/lib/spec-coverage.mjs +54 -3
- package/template/qa/lib/step-cache.mjs +221 -0
- package/template/qa/receipt-check.mjs +22 -2
- package/template/qa/record-audit.mjs +83 -0
- package/template/qa/retrospective.mjs +51 -0
- package/template/qa/scaffold-feature.mjs +20 -1
- package/template/qa/verify.mjs +934 -57
- package/template/qa/watch.mjs +622 -0
- package/template/specs/app-base.spec.md +11 -0
|
@@ -0,0 +1,249 @@
|
|
|
1
|
+
// The machine-global device lease — mutual exclusion for the ONE Android
|
|
2
|
+
// device/emulator a machine typically has.
|
|
3
|
+
//
|
|
4
|
+
// WHY THIS EXISTS: the lane marker (composeApp/build/.cmp-lane-in-progress) is
|
|
5
|
+
// per-PROJECT, but the device is machine-GLOBAL. A scratch app stamped in /tmp
|
|
6
|
+
// and the real app are different roots sharing one emulator: each stamps its
|
|
7
|
+
// own marker, neither sees the other, and two concurrent device drivers produce
|
|
8
|
+
// exactly the observed failure class — wedged adbd, `device offline` while
|
|
9
|
+
// `adb devices` looks fine, crossed app state between sessions, false-red runs.
|
|
10
|
+
// The lease is keyed by the DEVICE (its adb serial), not the project, so the
|
|
11
|
+
// primitive finally matches the scarce resource it protects.
|
|
12
|
+
//
|
|
13
|
+
// ── ON-DISK CONTRACT ────────────────────────────────────────────────────────
|
|
14
|
+
// This exact contract is implemented independently by the create-cmp
|
|
15
|
+
// inspector MCP (inspector/mcp/src/lib/device-lease.mjs in the create-cmp
|
|
16
|
+
// repo — a check-only reader for connect_live / navigate_and_inspect). The two
|
|
17
|
+
// codebases ship separately and cannot import each other, so the contract
|
|
18
|
+
// lives verbatim in BOTH file headers, each pointing at the other. Changing
|
|
19
|
+
// anything below means changing it there too.
|
|
20
|
+
//
|
|
21
|
+
// Location <os.tmpdir()>/create-cmp/device-leases/<sanitized-serial>.json
|
|
22
|
+
// tmpdir on purpose: a lease must never survive a reboot.
|
|
23
|
+
// Sanitizing serial chars outside [A-Za-z0-9._-] become "_"
|
|
24
|
+
// ("emulator-5554" → emulator-5554.json,
|
|
25
|
+
// "192.168.1.5:5555" → 192.168.1.5_5555.json).
|
|
26
|
+
// Shape { "pid": number, "holder": string, "root": string,
|
|
27
|
+
// "serial": string, "acquiredAt": ISO-8601 string }
|
|
28
|
+
// `holder` is a human/agent-readable label naming WHO is driving
|
|
29
|
+
// ("verify lane e2eSmoke", "connect_live", "fleet-check scratch
|
|
30
|
+
// lane"); `root` is the holder's project root.
|
|
31
|
+
// Staleness a lease is DEAD (silently reclaimable) when EITHER
|
|
32
|
+
// - its pid is not alive — process.kill(pid, 0) throws ESRCH.
|
|
33
|
+
// EPERM means the process EXISTS under another user: ALIVE.
|
|
34
|
+
// - OR acquiredAt is older than MAX_LEASE_AGE_MS.
|
|
35
|
+
// An unparseable lease file (torn write from a crashed holder)
|
|
36
|
+
// counts as dead. Readers treat dead as free; only acquirers
|
|
37
|
+
// delete/overwrite.
|
|
38
|
+
// Writes atomic — temp file in the same directory + rename. After the
|
|
39
|
+
// rename the acquirer re-reads and confirms its OWN pid is in
|
|
40
|
+
// the file: two simultaneous renames resolve last-writer-wins,
|
|
41
|
+
// and the loser reports contention instead of believing it holds
|
|
42
|
+
// the device.
|
|
43
|
+
// ────────────────────────────────────────────────────────────────────────────
|
|
44
|
+
//
|
|
45
|
+
// Dependency-free Node. Every function takes an optional { dir } override so
|
|
46
|
+
// tests exercise real files in a temp dir without touching the machine's
|
|
47
|
+
// actual leases, and an optional { killImpl } so the EPERM-means-alive branch
|
|
48
|
+
// is testable without a foreign-user process.
|
|
49
|
+
|
|
50
|
+
import crypto from "node:crypto";
|
|
51
|
+
import fs from "node:fs";
|
|
52
|
+
import os from "node:os";
|
|
53
|
+
import path from "node:path";
|
|
54
|
+
|
|
55
|
+
// 30 minutes: the longest legitimate single holder is a full release-profile
|
|
56
|
+
// device phase on a cold emulator (installDebug + Maestro smoke +
|
|
57
|
+
// connectedDebugAndroidTest + installRelease + release smoke), observed in the
|
|
58
|
+
// low tens of minutes — a live holder is protected by the pid check anyway, so
|
|
59
|
+
// this cap only decides how long a crashed holder whose pid was RECYCLED by an
|
|
60
|
+
// unrelated long-lived process can wedge the device. 30 min bounds that to
|
|
61
|
+
// roughly one lane-length: long enough never to reclaim under a healthy run,
|
|
62
|
+
// short enough that the machine heals itself within the hour.
|
|
63
|
+
export const MAX_LEASE_AGE_MS = 30 * 60 * 1000;
|
|
64
|
+
|
|
65
|
+
/** Serial → safe file stem: anything outside [A-Za-z0-9._-] becomes "_". */
|
|
66
|
+
export function sanitizeSerial(serial) {
|
|
67
|
+
return String(serial).replace(/[^A-Za-z0-9._-]/g, "_");
|
|
68
|
+
}
|
|
69
|
+
|
|
70
|
+
/** The machine-global lease directory (override with { dir } in tests only). */
|
|
71
|
+
export function leaseDir(dir) {
|
|
72
|
+
return dir || path.join(os.tmpdir(), "create-cmp", "device-leases");
|
|
73
|
+
}
|
|
74
|
+
|
|
75
|
+
/** Absolute path of the lease file for one serial. */
|
|
76
|
+
export function leasePath(serial, { dir } = {}) {
|
|
77
|
+
return path.join(leaseDir(dir), `${sanitizeSerial(serial)}.json`);
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
/**
|
|
81
|
+
* Is a pid alive? ESRCH = no such process → dead. EPERM = the process exists
|
|
82
|
+
* but belongs to another user → ALIVE (killing rights are not liveness).
|
|
83
|
+
* Any other error is treated as alive — when in doubt, never steal a lease.
|
|
84
|
+
*/
|
|
85
|
+
export function pidAlive(pid, { killImpl = process.kill.bind(process) } = {}) {
|
|
86
|
+
if (!Number.isInteger(pid) || pid <= 0) return false;
|
|
87
|
+
try {
|
|
88
|
+
killImpl(pid, 0);
|
|
89
|
+
return true;
|
|
90
|
+
} catch (err) {
|
|
91
|
+
return !(err && err.code === "ESRCH");
|
|
92
|
+
}
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
function readLeaseFile(file) {
|
|
96
|
+
try {
|
|
97
|
+
return JSON.parse(fs.readFileSync(file, "utf8"));
|
|
98
|
+
} catch {
|
|
99
|
+
return null; // missing OR unparseable — both mean "no live lease here"
|
|
100
|
+
}
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
function describeLease(lease, nowMs) {
|
|
104
|
+
const acquiredMs = Date.parse(lease.acquiredAt);
|
|
105
|
+
return {
|
|
106
|
+
holder: lease.holder ?? "unknown",
|
|
107
|
+
pid: lease.pid ?? null,
|
|
108
|
+
root: lease.root ?? null,
|
|
109
|
+
acquiredAt: lease.acquiredAt ?? null,
|
|
110
|
+
ageMs: Number.isFinite(acquiredMs) ? Math.max(0, nowMs - acquiredMs) : Number.POSITIVE_INFINITY,
|
|
111
|
+
};
|
|
112
|
+
}
|
|
113
|
+
|
|
114
|
+
function leaseDead(lease, nowMs, { killImpl } = {}) {
|
|
115
|
+
if (!pidAlive(lease.pid, { ...(killImpl ? { killImpl } : {}) })) return true;
|
|
116
|
+
const acquiredMs = Date.parse(lease.acquiredAt);
|
|
117
|
+
if (!Number.isFinite(acquiredMs)) return true; // no readable birth time — unverifiable, dead
|
|
118
|
+
return nowMs - acquiredMs >= MAX_LEASE_AGE_MS;
|
|
119
|
+
}
|
|
120
|
+
|
|
121
|
+
/** `"verify lane e2eSmoke" (pid 4711, /tmp/scratch-x, 2m ago)` — for reasons/errors. */
|
|
122
|
+
export function formatHolder(heldBy) {
|
|
123
|
+
if (!heldBy) return "an unknown holder";
|
|
124
|
+
const age =
|
|
125
|
+
!Number.isFinite(heldBy.ageMs) ? "age unknown"
|
|
126
|
+
: heldBy.ageMs < 60_000 ? `${Math.max(1, Math.round(heldBy.ageMs / 1000))}s ago`
|
|
127
|
+
: `${Math.round(heldBy.ageMs / 60_000)}m ago`;
|
|
128
|
+
return `"${heldBy.holder}" (pid ${heldBy.pid ?? "?"}, ${heldBy.root ?? "unknown root"}, ${age})`;
|
|
129
|
+
}
|
|
130
|
+
|
|
131
|
+
/**
|
|
132
|
+
* Read the live lease on a serial, if any. Dead/stale leases read as null
|
|
133
|
+
* (free) — reading never deletes; reclaim-by-overwrite is the acquirer's job.
|
|
134
|
+
*
|
|
135
|
+
* @returns {{holder,pid,root,acquiredAt,ageMs}|null}
|
|
136
|
+
*/
|
|
137
|
+
export function readDeviceLease(serial, { dir, killImpl, now = Date.now } = {}) {
|
|
138
|
+
const lease = readLeaseFile(leasePath(serial, { dir }));
|
|
139
|
+
if (!lease) return null;
|
|
140
|
+
const nowMs = now();
|
|
141
|
+
if (leaseDead(lease, nowMs, { killImpl })) return null;
|
|
142
|
+
return describeLease(lease, nowMs);
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
/**
|
|
146
|
+
* Acquire the machine-global lease on one device serial.
|
|
147
|
+
*
|
|
148
|
+
* @param {{serial: string, holder: string, root: string, dir?: string,
|
|
149
|
+
* killImpl?: Function, now?: () => number}} opts
|
|
150
|
+
* @returns {{ok: true, handle: {file, serial, pid, acquiredAt}, reclaimed: object|null}
|
|
151
|
+
* | {ok: false, heldBy: {holder, pid, root, acquiredAt, ageMs}}}
|
|
152
|
+
* `reclaimed` names the dead lease this acquire silently replaced (a crashed
|
|
153
|
+
* run must never wedge the machine forever) so the acquiring run can note it
|
|
154
|
+
* in its own output.
|
|
155
|
+
*/
|
|
156
|
+
export function acquireDeviceLease({ serial, holder, root, dir, killImpl, now = Date.now } = {}) {
|
|
157
|
+
if (!serial) throw new Error("acquireDeviceLease: serial is required");
|
|
158
|
+
const d = leaseDir(dir);
|
|
159
|
+
fs.mkdirSync(d, { recursive: true });
|
|
160
|
+
const file = leasePath(serial, { dir });
|
|
161
|
+
const nowMs = now();
|
|
162
|
+
|
|
163
|
+
let reclaimed = null;
|
|
164
|
+
const existing = readLeaseFile(file);
|
|
165
|
+
if (existing) {
|
|
166
|
+
if (!leaseDead(existing, nowMs, { killImpl })) {
|
|
167
|
+
return { ok: false, heldBy: describeLease(existing, nowMs) };
|
|
168
|
+
}
|
|
169
|
+
reclaimed = describeLease(existing, nowMs); // dead — reclaim silently
|
|
170
|
+
} else if (fs.existsSync(file)) {
|
|
171
|
+
reclaimed = { holder: "unreadable lease (torn write)", pid: null, root: null, acquiredAt: null, ageMs: Number.POSITIVE_INFINITY };
|
|
172
|
+
}
|
|
173
|
+
|
|
174
|
+
const lease = {
|
|
175
|
+
pid: process.pid,
|
|
176
|
+
holder: String(holder || "unknown"),
|
|
177
|
+
root: String(root || process.cwd()),
|
|
178
|
+
serial: String(serial),
|
|
179
|
+
acquiredAt: new Date(nowMs).toISOString(),
|
|
180
|
+
};
|
|
181
|
+
// Atomic claim: temp file + rename means no reader ever sees a half-written
|
|
182
|
+
// lease, and two simultaneous acquirers cannot interleave bytes.
|
|
183
|
+
const tmp = path.join(d, `.${sanitizeSerial(serial)}.${process.pid}.${crypto.randomBytes(4).toString("hex")}.tmp`);
|
|
184
|
+
fs.writeFileSync(tmp, `${JSON.stringify(lease, null, 2)}\n`);
|
|
185
|
+
fs.renameSync(tmp, file);
|
|
186
|
+
|
|
187
|
+
// Last-writer-wins detection: both racers reached the rename; whoever's bytes
|
|
188
|
+
// survived owns the device. Re-read and confirm it is US — the loser reports
|
|
189
|
+
// contention instead of driving a device someone else holds.
|
|
190
|
+
const confirm = readLeaseFile(file);
|
|
191
|
+
if (!confirm || confirm.pid !== lease.pid || confirm.acquiredAt !== lease.acquiredAt || confirm.holder !== lease.holder) {
|
|
192
|
+
return {
|
|
193
|
+
ok: false,
|
|
194
|
+
heldBy: confirm
|
|
195
|
+
? describeLease(confirm, now())
|
|
196
|
+
: { holder: "unknown (lease vanished mid-acquire)", pid: null, root: null, acquiredAt: null, ageMs: Number.POSITIVE_INFINITY },
|
|
197
|
+
};
|
|
198
|
+
}
|
|
199
|
+
return { ok: true, handle: { file, serial: lease.serial, pid: lease.pid, acquiredAt: lease.acquiredAt }, reclaimed };
|
|
200
|
+
}
|
|
201
|
+
|
|
202
|
+
/**
|
|
203
|
+
* Release a held lease. Idempotent and safe: missing file is fine, and a file
|
|
204
|
+
* that no longer carries OUR pid+acquiredAt belongs to a newer holder (they
|
|
205
|
+
* reclaimed us as stale, or won a race) — another holder's lease is NEVER
|
|
206
|
+
* deleted.
|
|
207
|
+
*/
|
|
208
|
+
export function releaseDeviceLease(handle) {
|
|
209
|
+
if (!handle || !handle.file) return;
|
|
210
|
+
const current = readLeaseFile(handle.file);
|
|
211
|
+
if (!current) return; // already gone (or unreadable — not provably ours, leave it)
|
|
212
|
+
if (current.pid !== handle.pid || current.acquiredAt !== handle.acquiredAt) return; // someone else's now
|
|
213
|
+
try {
|
|
214
|
+
fs.rmSync(handle.file, { force: true });
|
|
215
|
+
} catch {
|
|
216
|
+
/* releasing is best-effort; staleness reclaim is the backstop */
|
|
217
|
+
}
|
|
218
|
+
}
|
|
219
|
+
|
|
220
|
+
/**
|
|
221
|
+
* Run `fn(handle)` under the lease, releasing in a finally. Sync or async fn.
|
|
222
|
+
* A refused acquire is returned as-is ({ok:false, heldBy}) — the caller decides
|
|
223
|
+
* what contention means (the verify lane turns it into a SKIP, never a FAIL).
|
|
224
|
+
*/
|
|
225
|
+
export function withDeviceLease(opts, fn) {
|
|
226
|
+
const res = acquireDeviceLease(opts);
|
|
227
|
+
if (!res.ok) return res;
|
|
228
|
+
let out;
|
|
229
|
+
try {
|
|
230
|
+
out = fn(res.handle);
|
|
231
|
+
} catch (err) {
|
|
232
|
+
releaseDeviceLease(res.handle);
|
|
233
|
+
throw err;
|
|
234
|
+
}
|
|
235
|
+
if (out && typeof out.then === "function") {
|
|
236
|
+
return out.then(
|
|
237
|
+
(value) => {
|
|
238
|
+
releaseDeviceLease(res.handle);
|
|
239
|
+
return { ok: true, result: value, reclaimed: res.reclaimed };
|
|
240
|
+
},
|
|
241
|
+
(err) => {
|
|
242
|
+
releaseDeviceLease(res.handle);
|
|
243
|
+
throw err;
|
|
244
|
+
},
|
|
245
|
+
);
|
|
246
|
+
}
|
|
247
|
+
releaseDeviceLease(res.handle);
|
|
248
|
+
return { ok: true, result: out, reclaimed: res.reclaimed };
|
|
249
|
+
}
|
|
@@ -0,0 +1,158 @@
|
|
|
1
|
+
// The README's evidence badge — the evidence ladder, rendered where a human
|
|
2
|
+
// actually looks (roadmap §10 item 2: "render the rung in the console and
|
|
3
|
+
// README badge").
|
|
4
|
+
//
|
|
5
|
+
// The console already shows the rung (rail foot, Evidence section, receipt
|
|
6
|
+
// timeline). The README is the surface a human meets FIRST, and the one that
|
|
7
|
+
// travels — into a GitHub repo page, a PR, a screenshot in a deck. That makes
|
|
8
|
+
// it the surface where an overclaim does the most damage, so the badge obeys
|
|
9
|
+
// one rule above all others:
|
|
10
|
+
//
|
|
11
|
+
// **The badge is a statement about a specific commit, never about "now".**
|
|
12
|
+
//
|
|
13
|
+
// A badge that says "L2 device" says nothing about whether the code has moved
|
|
14
|
+
// since. So it never renders a bare rung: it renders the rung AND the commit
|
|
15
|
+
// it was attested against AND the date. That sentence stays true forever — a
|
|
16
|
+
// reader can see at a glance whether the sha still matches what they are
|
|
17
|
+
// looking at. Everything else follows from the same rule: no receipt says so,
|
|
18
|
+
// a FAIL says so, and a --fast run (which the ladder deliberately grants no
|
|
19
|
+
// rung) says so rather than borrowing the last good one.
|
|
20
|
+
//
|
|
21
|
+
// Written by the lane AFTER the receipt (it is an output derived from the
|
|
22
|
+
// receipt, never a gate), and committed alongside it.
|
|
23
|
+
|
|
24
|
+
import fs from "node:fs";
|
|
25
|
+
import path from "node:path";
|
|
26
|
+
|
|
27
|
+
export const README_REL_PATH = "README.md";
|
|
28
|
+
export const BADGE_SECTION_ID = "evidence";
|
|
29
|
+
|
|
30
|
+
const MARKER_RE = new RegExp(
|
|
31
|
+
`<!-- cmp:generated ${BADGE_SECTION_ID} -->\\n([\\s\\S]*?)<!-- /cmp:generated -->`
|
|
32
|
+
);
|
|
33
|
+
|
|
34
|
+
/** Shields.io colours, one per rung — the ladder read at a glance. */
|
|
35
|
+
const RUNG_COLOR = {
|
|
36
|
+
L0: "9E9E9E", // scaffold — grey: a green build, nothing proven about behavior
|
|
37
|
+
L1: "42A5F5", // desktop — blue
|
|
38
|
+
L2: "26A69A", // device — teal
|
|
39
|
+
L3: "43A047", // release — green: the strongest rung this harness can attest
|
|
40
|
+
};
|
|
41
|
+
|
|
42
|
+
/** shields.io escaping: `-` → `--`, `_` → `__`, space → `_`. */
|
|
43
|
+
function shieldEscape(s) {
|
|
44
|
+
return String(s).replace(/-/g, "--").replace(/_/g, "__").replace(/ /g, "_");
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
/**
|
|
48
|
+
* The badge body for a receipt — Markdown, no trailing newline handling (the
|
|
49
|
+
* caller frames it). Pure: every degraded state has its own honest rendering
|
|
50
|
+
* and NONE of them fall back to a rung.
|
|
51
|
+
*
|
|
52
|
+
* @param {object|null} receipt parsed qa/evidence/latest.json, or null
|
|
53
|
+
* @returns {string} Markdown
|
|
54
|
+
*/
|
|
55
|
+
export function renderEvidenceBadge(receipt) {
|
|
56
|
+
const link = "https://github.com/kvdm-co-pilot/create-cmp";
|
|
57
|
+
const badge = (label, message, color, title) =>
|
|
58
|
+
`[}-${shieldEscape(message)}-${color})](${link})`;
|
|
59
|
+
|
|
60
|
+
if (!receipt || typeof receipt !== "object") {
|
|
61
|
+
return `${badge("evidence", "none yet", "9E9E9E", "No evidence receipt")} — no verify receipt yet. Run \`node qa/verify.mjs\`.`;
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
const verdict = typeof receipt.verdict === "string" ? receipt.verdict : "?";
|
|
65
|
+
const mode = receipt.mode === "fast" ? "fast" : "full";
|
|
66
|
+
const sha = typeof receipt.commit?.sha === "string" ? receipt.commit.sha.slice(0, 7) : null;
|
|
67
|
+
const dirty = Array.isArray(receipt.commit?.dirty) ? receipt.commit.dirty.length : 0;
|
|
68
|
+
const when = typeof receipt.generatedAt === "string" ? receipt.generatedAt.slice(0, 10) : null;
|
|
69
|
+
|
|
70
|
+
// Provenance is not decoration — it is what keeps the sentence true later.
|
|
71
|
+
const at = sha ? ` at \`${sha}\`` : "";
|
|
72
|
+
const on = when ? ` on ${when}` : "";
|
|
73
|
+
const uncommitted =
|
|
74
|
+
dirty > 0
|
|
75
|
+
? ` The tree had ${dirty} uncommitted file${dirty === 1 ? "" : "s"} at attestation, so this describes that run, not that commit.`
|
|
76
|
+
: "";
|
|
77
|
+
|
|
78
|
+
if (verdict !== "PASS") {
|
|
79
|
+
return `${badge("evidence", `lane ${verdict}`, "E53935", `Verify lane ${verdict}`)} — the last lane run${at}${on} did not pass. No rung is earned by a failed lane.${uncommitted}`;
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
if (mode === "fast") {
|
|
83
|
+
// The inner loop is a signal, never evidence. Borrowing the previous
|
|
84
|
+
// full run's rung here is exactly the lie the ladder exists to prevent.
|
|
85
|
+
return `${badge("evidence", "fast run, no rung", "9E9E9E", "Fast run — no evidence rung")} — the last run${at}${on} was \`--fast\`: the device and release tiers were skipped, so it earns no rung.${uncommitted} Run \`node qa/verify.mjs\` for evidence.`;
|
|
86
|
+
}
|
|
87
|
+
|
|
88
|
+
const level = receipt.evidenceLevel;
|
|
89
|
+
if (!level || typeof level.rung !== "string" || typeof level.name !== "string") {
|
|
90
|
+
return `${badge("evidence", `PASS, rung unrecorded`, "9E9E9E", "Lane PASS, no rung recorded")} — the lane passed${at}${on} but the receipt records no evidence rung.`;
|
|
91
|
+
}
|
|
92
|
+
|
|
93
|
+
const color = RUNG_COLOR[level.rung] || "9E9E9E";
|
|
94
|
+
const satisfied = Array.isArray(level.satisfiedBy) && level.satisfiedBy.length
|
|
95
|
+
? ` Earned by: ${level.satisfiedBy.map((s) => `\`${s}\``).join(", ")}.`
|
|
96
|
+
: "";
|
|
97
|
+
return (
|
|
98
|
+
`${badge("evidence", `${level.rung} ${level.name}`, color, `Evidence ${level.rung} — ${level.name}`)}` +
|
|
99
|
+
` — the verify lane passed${at}${on} at rung **${level.rung} · ${level.name}**.` +
|
|
100
|
+
`${satisfied}${uncommitted}` +
|
|
101
|
+
` The rung describes that run; it says nothing about changes made since.`
|
|
102
|
+
);
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
/**
|
|
106
|
+
* Read the project's receipt and rewrite README.md's `cmp:generated evidence`
|
|
107
|
+
* block from it. Never creates the marker — a project that removed the block
|
|
108
|
+
* has opted out, and that is honoured silently.
|
|
109
|
+
*
|
|
110
|
+
* NOTE the asymmetry with renderEvidenceBadge above: the RENDERER is total —
|
|
111
|
+
* every receipt, including a `--fast` one, has an honest rendering. The WRITER
|
|
112
|
+
* is selective: a fast receipt is not written to the README at all. Two
|
|
113
|
+
* reasons, and the second is the load-bearing one:
|
|
114
|
+
* 1. The badge reports EVIDENCE. A fast run produces none, so it has nothing
|
|
115
|
+
* to say — and overwriting a true statement about a real full-lane run
|
|
116
|
+
* with "no rung" loses information rather than adding honesty.
|
|
117
|
+
* 2. `qa/watch.mjs` runs the fast lane on every save. A writer that fired
|
|
118
|
+
* there would rewrite README.md on every keystroke-to-save cycle, putting
|
|
119
|
+
* a permanently-dirty file in the inner loop. A recorder must not disturb
|
|
120
|
+
* what it records.
|
|
121
|
+
* The badge therefore always describes the last run that could BEAR evidence,
|
|
122
|
+
* and says so by naming that run's commit and date.
|
|
123
|
+
*
|
|
124
|
+
* @param {string} root project root
|
|
125
|
+
* @returns {{changed: boolean, reason?: string}}
|
|
126
|
+
*/
|
|
127
|
+
export function updateReadmeBadge(root) {
|
|
128
|
+
const readmePath = path.join(root, README_REL_PATH);
|
|
129
|
+
let readme;
|
|
130
|
+
try {
|
|
131
|
+
readme = fs.readFileSync(readmePath, "utf8");
|
|
132
|
+
} catch {
|
|
133
|
+
return { changed: false, reason: `${README_REL_PATH} not found` };
|
|
134
|
+
}
|
|
135
|
+
if (!MARKER_RE.test(readme)) {
|
|
136
|
+
return { changed: false, reason: `${README_REL_PATH} has no cmp:generated ${BADGE_SECTION_ID} block` };
|
|
137
|
+
}
|
|
138
|
+
|
|
139
|
+
let receipt = null;
|
|
140
|
+
try {
|
|
141
|
+
receipt = JSON.parse(fs.readFileSync(path.join(root, "qa", "evidence", "latest.json"), "utf8"));
|
|
142
|
+
} catch {
|
|
143
|
+
receipt = null; // no receipt / unreadable → the "none yet" rendering, never a guess
|
|
144
|
+
}
|
|
145
|
+
|
|
146
|
+
if (receipt && receipt.mode === "fast") {
|
|
147
|
+
return { changed: false, reason: "fast run — the inner loop bears no evidence, so the badge is left as it stands" };
|
|
148
|
+
}
|
|
149
|
+
|
|
150
|
+
const body = `${renderEvidenceBadge(receipt)}\n`;
|
|
151
|
+
const next = readme.replace(
|
|
152
|
+
MARKER_RE,
|
|
153
|
+
() => `<!-- cmp:generated ${BADGE_SECTION_ID} -->\n${body}<!-- /cmp:generated -->`
|
|
154
|
+
);
|
|
155
|
+
if (next === readme) return { changed: false };
|
|
156
|
+
fs.writeFileSync(readmePath, next);
|
|
157
|
+
return { changed: true };
|
|
158
|
+
}
|
|
@@ -0,0 +1,117 @@
|
|
|
1
|
+
// The evidence ladder — the receipt's COARSE grade, derived, never declared.
|
|
2
|
+
//
|
|
3
|
+
// Receipts already grade themselves in fine print ("PASS (desktop-only)",
|
|
4
|
+
// "PASS (on-device: e2eSmoke+androidChecks)"). This module names the rungs so
|
|
5
|
+
// every surface that shows a receipt can say the same thing in one word:
|
|
6
|
+
//
|
|
7
|
+
// L0 "scaffold" — the scaffold profile's checks passed (stamp-time green
|
|
8
|
+
// build: build + unit tests + the pure-Node gates).
|
|
9
|
+
// L1 "desktop" — full static + JVM evidence: everything L0 proves PLUS
|
|
10
|
+
// conformance, golden trees, a11y, and the release COMPILE
|
|
11
|
+
// (releaseBuild) — a green lane with no on-device step run.
|
|
12
|
+
// L2 "device" — L1 plus at least one on-device EXECUTION step PASSed
|
|
13
|
+
// (e2eSmoke, androidChecks, or the live tokenDrift tier).
|
|
14
|
+
// L3 "release" — L2 plus releaseSmoke PASSed (the release APK installed
|
|
15
|
+
// and driven on a device).
|
|
16
|
+
//
|
|
17
|
+
// HONESTY RULES — the rung must be honest to a fault, it is the vocabulary
|
|
18
|
+
// evidence is sold in:
|
|
19
|
+
// - A rung is DERIVED from which steps actually ran and PASSED. It is never
|
|
20
|
+
// declared: the `profile` argument is deliberately NOT part of the
|
|
21
|
+
// derivation — a requested profile can never buy a rung its steps did not
|
|
22
|
+
// earn (it is accepted so callers state what was asked for vs. earned).
|
|
23
|
+
// - A SKIP never upgrades. A SKIPped device step does not count toward L2;
|
|
24
|
+
// a SKIPped releaseSmoke (e.g. unsigned keystore) is NOT L3. The label
|
|
25
|
+
// can never overclaim.
|
|
26
|
+
// - A FAILED lane has no rung: the rung is only computed for a PASS
|
|
27
|
+
// verdict; the receipt of a FAIL records evidenceLevel null.
|
|
28
|
+
// - A FAST-MODE lane has no rung either — not even L0. `verify --fast` is
|
|
29
|
+
// the inner loop, a signal rather than evidence, so a fast receipt must
|
|
30
|
+
// never be silently reused as if it were a full-lane result: pass the
|
|
31
|
+
// run's mode and "fast" derives null, always.
|
|
32
|
+
// - The rung is COARSE by design. The per-step list (and the existing
|
|
33
|
+
// strength string) stays the fine print alongside it — steps that may
|
|
34
|
+
// SKIP for honest configuration absence (approvals unreviewed, no
|
|
35
|
+
// exported schemas) are visible there; only the always-run steps gate
|
|
36
|
+
// the desktop rungs, and only executed PASSes gate the device rungs.
|
|
37
|
+
|
|
38
|
+
/** The scaffold profile's step set (verify.mjs stepsForProfile.scaffold). */
|
|
39
|
+
const SCAFFOLD_CORE = [
|
|
40
|
+
"specCoverage",
|
|
41
|
+
"approvals",
|
|
42
|
+
"componentStories",
|
|
43
|
+
"reachability",
|
|
44
|
+
"archDoc",
|
|
45
|
+
"schemaHistory",
|
|
46
|
+
"build",
|
|
47
|
+
"unitTests",
|
|
48
|
+
];
|
|
49
|
+
|
|
50
|
+
/** Steps every PASS must carry to claim even L0 — they run in every profile and never SKIP. */
|
|
51
|
+
const L0_REQUIRED = ["build", "unitTests"];
|
|
52
|
+
|
|
53
|
+
/**
|
|
54
|
+
* The steps that distinguish full desktop evidence (L1) from the scaffold
|
|
55
|
+
* checks. None of these can SKIP — they PASS or FAIL — so "PASSed" is exactly
|
|
56
|
+
* "ran green".
|
|
57
|
+
*/
|
|
58
|
+
const L1_REQUIRED = ["releaseBuild", "conformance", "goldenTrees", "a11y"];
|
|
59
|
+
|
|
60
|
+
/** On-device EXECUTION steps — the only steps that can earn L2. */
|
|
61
|
+
const DEVICE_EXECUTION = ["e2eSmoke", "tokenDrift", "androidChecks"];
|
|
62
|
+
|
|
63
|
+
/** The one step that can lift L2 to L3. */
|
|
64
|
+
const RELEASE_EXECUTION = "releaseSmoke";
|
|
65
|
+
|
|
66
|
+
const RUNG_NAMES = { L0: "scaffold", L1: "desktop", L2: "device", L3: "release" };
|
|
67
|
+
|
|
68
|
+
/**
|
|
69
|
+
* Derive the receipt's evidence rung from the lane's step results.
|
|
70
|
+
*
|
|
71
|
+
* @param {Array<{name: string, verdict: string}>} stepResults the lane's steps
|
|
72
|
+
* as recorded on the receipt (verdict PASS | FAIL | SKIP per step)
|
|
73
|
+
* @param {string} [profile] the profile that was REQUESTED — recorded context
|
|
74
|
+
* only, never part of the derivation (see honesty rules above)
|
|
75
|
+
* @param {{mode?: string}} [opts] the run's mode ("full" | "fast"). "fast"
|
|
76
|
+
* derives null unconditionally — the inner loop earns no rung (see honesty
|
|
77
|
+
* rules above). Absent/other values mean full.
|
|
78
|
+
* @returns {{rung: "L0"|"L1"|"L2"|"L3", name: string, satisfiedBy: string[]}|null}
|
|
79
|
+
* null when any step FAILed (a failed lane has no rung), when the run was
|
|
80
|
+
* fast-mode (the inner loop is never evidence), or when even the L0 floor
|
|
81
|
+
* was not earned. `satisfiedBy` lists the PASSed steps the rung counts as
|
|
82
|
+
* its evidence, in lane order.
|
|
83
|
+
*/
|
|
84
|
+
export function evidenceLevel(stepResults, profile, { mode } = {}) { // eslint-disable-line no-unused-vars
|
|
85
|
+
if (mode === "fast") return null; // the inner loop derives no rung — ever
|
|
86
|
+
const steps = Array.isArray(stepResults) ? stepResults.filter((s) => s && typeof s.name === "string") : [];
|
|
87
|
+
if (steps.some((s) => s.verdict === "FAIL")) return null; // a failed lane has no rung
|
|
88
|
+
const passed = new Set(steps.filter((s) => s.verdict === "PASS").map((s) => s.name));
|
|
89
|
+
|
|
90
|
+
if (!L0_REQUIRED.every((name) => passed.has(name))) return null; // not even a stamp-time green build
|
|
91
|
+
|
|
92
|
+
const inLaneOrder = (names) => steps.filter((s) => names.has(s.name) && passed.has(s.name)).map((s) => s.name);
|
|
93
|
+
|
|
94
|
+
let rung = "L0";
|
|
95
|
+
const counted = new Set(SCAFFOLD_CORE);
|
|
96
|
+
|
|
97
|
+
if (L1_REQUIRED.every((name) => passed.has(name))) {
|
|
98
|
+
rung = "L1";
|
|
99
|
+
for (const name of L1_REQUIRED) counted.add(name);
|
|
100
|
+
|
|
101
|
+
// Only an EXECUTED (PASSed) device step lifts to L2 — a SKIP never does.
|
|
102
|
+
const deviceRan = DEVICE_EXECUTION.some((name) => passed.has(name));
|
|
103
|
+
if (deviceRan) {
|
|
104
|
+
rung = "L2";
|
|
105
|
+
for (const name of DEVICE_EXECUTION) counted.add(name);
|
|
106
|
+
|
|
107
|
+
// Only a PASSed releaseSmoke lifts to L3 — a SKIP (unsigned keystore,
|
|
108
|
+
// no device) never does.
|
|
109
|
+
if (passed.has(RELEASE_EXECUTION)) {
|
|
110
|
+
rung = "L3";
|
|
111
|
+
counted.add(RELEASE_EXECUTION);
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
|
|
116
|
+
return { rung, name: RUNG_NAMES[rung], satisfiedBy: inLaneOrder(counted) };
|
|
117
|
+
}
|