@activate-core/test-forge 2.5.0 → 2.7.0

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.
@@ -51,10 +51,11 @@ export const DEFAULT_EXTERNAL_MODULES = [
51
51
  * clean. Runs AFTER syntax validation.
52
52
  */
53
53
  export function checkTestGates(content, _ext, ctx) {
54
+ const reasons = [];
54
55
  // F3 — runner correctness: a Jest/backend area must never pull in
55
56
  // Playwright (not installed there; a top-level import breaks the suite).
56
57
  if (ctx.framework === 'jest' && /@playwright\/test/.test(content)) {
57
- return "jest area must not import '@playwright/test'";
58
+ reasons.push("jest area must not import '@playwright/test'");
58
59
  }
59
60
  // F4 — goto guard: `page.goto(...)` must take a real route/URL, not a
60
61
  // pasted description and not an unfilled route-file placeholder. Literal
@@ -67,10 +68,10 @@ export function checkTestGates(content, _ext, ctx) {
67
68
  while ((m = gotoRe.exec(content)) !== null) {
68
69
  const arg = m[2];
69
70
  if (/\s/.test(arg)) {
70
- return `goto() argument looks like prose, not a URL/path: "${arg.slice(0, 40)}…"`;
71
+ reasons.push(`goto() argument looks like prose, not a URL/path: "${arg.slice(0, 40)}…"`);
71
72
  }
72
73
  if (/\[[A-Za-z_]|\/:[A-Za-z_]|\{[A-Za-z_]|<[A-Za-z_]/.test(arg)) {
73
- return `goto() has an unresolved route placeholder — fill it with a real value: "${arg.slice(0, 40)}"`;
74
+ reasons.push(`goto() has an unresolved route placeholder — fill it with a real value: "${arg.slice(0, 40)}"`);
74
75
  }
75
76
  }
76
77
  // F13 — a swallowed action must be bounded. `.click({ force: true })`
@@ -85,7 +86,7 @@ export function checkTestGates(content, _ext, ctx) {
85
86
  let sw;
86
87
  while ((sw = swallowed.exec(content)) !== null) {
87
88
  if (!/timeout\s*:/.test(sw[0])) {
88
- return ('an action whose failure is swallowed with .catch() has no timeout — ' +
89
+ reasons.push('an action whose failure is swallowed with .catch() has no timeout — ' +
89
90
  'it will wait until the whole test times out and blame a later line');
90
91
  }
91
92
  }
@@ -96,7 +97,7 @@ export function checkTestGates(content, _ext, ctx) {
96
97
  // that never comes. Hold the promise first, then act, then await it.
97
98
  const lateWait = /\b(?:await\s+)?page\.goto\([^\n]*\);?\s*\n(?:\s*\/\/[^\n]*\n)*\s*await\s+page\.waitFor(?:Response|Request|Event)\s*\(/;
98
99
  if (lateWait.test(content)) {
99
- return ('page.waitForResponse/Request is armed AFTER the navigation that triggers ' +
100
+ reasons.push('page.waitForResponse/Request is armed AFTER the navigation that triggers ' +
100
101
  'it — hold the promise first, then navigate, then await it');
101
102
  }
102
103
  // F2 — no silent false-green: a file with RUNNABLE tests must assert
@@ -108,7 +109,7 @@ export function checkTestGates(content, _ext, ctx) {
108
109
  const assertions = (content.match(/\bexpect\s*\(/g) || []).length +
109
110
  (content.match(/\bassert\b/g) || []).length;
110
111
  if (activeTests > 0 && assertions === 0) {
111
- return 'has runnable test(s) but no assertions — would pass without testing anything';
112
+ reasons.push('has runnable test(s) but no assertions — would pass without testing anything');
112
113
  }
113
114
  // F1 — dependency awareness: reject imports/requires of a module the
114
115
  // project can't resolve (bare specifier not in allowedModules).
@@ -116,7 +117,7 @@ export function checkTestGates(content, _ext, ctx) {
116
117
  for (const mod of importedModules(content)) {
117
118
  const top = topLevelModule(mod);
118
119
  if (!ctx.allowedModules.has(top)) {
119
- return `imports '${mod}' which the project does not have installed`;
120
+ reasons.push(`imports '${mod}' which the project does not have installed`);
120
121
  }
121
122
  }
122
123
  }
@@ -139,7 +140,7 @@ export function checkTestGates(content, _ext, ctx) {
139
140
  content.includes(`jest.mock('${top}'`) ||
140
141
  content.includes(`jest.mock("${top}"`);
141
142
  if (!mocked) {
142
- return `imports external service '${top}' without a jest.mock — the test would hit the live service; mock it at the seam`;
143
+ reasons.push(`imports external service '${top}' without a jest.mock — the test would hit the live service; mock it at the seam`);
143
144
  }
144
145
  }
145
146
  }
@@ -160,7 +161,7 @@ export function checkTestGates(content, _ext, ctx) {
160
161
  const showedWork = /\b(?:port|origin|path\s*prefix|prefix|resource\s*type|header|predicate)\b/i;
161
162
  const m = content.match(sameHost);
162
163
  if (m && !showedWork.test(m[0])) {
163
- return ('a test is skipped because the dependency "shares a host" with the app, ' +
164
+ reasons.push('a test is skipped because the dependency "shares a host" with the app, ' +
164
165
  'without saying what else was checked — a different port is a different ' +
165
166
  'origin, and a path prefix or a request predicate usually separates them ' +
166
167
  'too; scope the stub to one of those, or name the ones you ruled out');
@@ -169,20 +170,20 @@ export function checkTestGates(content, _ext, ctx) {
169
170
  // F16 — no placeholders in a freshly written file: a placeholder never runs,
170
171
  // and a suite full of them reads as coverage it does not have.
171
172
  if (ctx.forbidPlaceholders && pendingDeclarations(content) > 0) {
172
- return 'placeholder tests (test.fixme / test.skip / test.todo declarations) are not written: leave a case with no outcome to check out of the file';
173
+ reasons.push('placeholder tests (test.fixme / test.skip / test.todo declarations) are not written: leave a case with no outcome to check out of the file');
173
174
  }
174
175
  // F23 — need() asks for the name of a thing, anywhere in the file (a constant at
175
176
  // the top too), never for data the test should make itself.
176
177
  if (ctx.forbidPlaceholders) {
177
178
  const data = [...content.matchAll(/\bneed\(\s*(['"`])([^'"`\n]+)\1/g)].map((m) => m[2]).filter((n) => !isThingName(n));
178
179
  if (data.length) {
179
- return `asks need() for ${[...new Set(data)].slice(0, 3).map((n) => `"${n}"`).join(', ')}, which is data, not the name of a thing: the test makes its own ids, payloads and records; need() is only for "… URL", "… command", "… table", "… field", "… label"`;
180
+ reasons.push(`asks need() for ${[...new Set(data)].slice(0, 3).map((n) => `"${n}"`).join(', ')}, which is data, not the name of a thing: the test makes its own ids, payloads and records; need() is only for "… URL", "… command", "… table", "… field", "… label"`);
180
181
  }
181
182
  }
182
183
  // F22 — a fresh file exists to test its rules: one with no test at all says
183
184
  // nothing, and passes. It goes back once, told to stub what it cannot create.
184
185
  if (ctx.forbidPlaceholders && ctx.framework === 'playwright' && !/\b(?:test|it)(?:\.only)?\s*\(\s*['"`]/.test(stripPendingBlocks(content))) {
185
- return 'the file has no test: every rule it was given gets one — a state the test cannot create is stubbed with page.route(), never left out';
186
+ reasons.push('the file has no test: every rule it was given gets one — a state the test cannot create is stubbed with page.route(), never left out');
186
187
  }
187
188
  // F24 — one comment policy: the header, an oracle line above each test,
188
189
  // cannot-test lines, a timeout reason, and a short note on a parametrised
@@ -190,7 +191,7 @@ export function checkTestGates(content, _ext, ctx) {
190
191
  if (ctx.forbidPlaceholders) {
191
192
  const stray = strayComment(content);
192
193
  if (stray) {
193
- return `comment not allowed (line ${stray.line}): "${stray.text.slice(0, 60)}" — the only comments are the // @generated header, one // oracle: line above each test, // cannot-test: lines, a // timeout: reason, and a one-line note on a parametrised row`;
194
+ reasons.push(`comment not allowed (line ${stray.line}): "${stray.text.slice(0, 60)}" — the only comments are the // @generated header, one // oracle: line above each test, // cannot-test: lines, a // timeout: reason, and a one-line note on a parametrised row`);
194
195
  }
195
196
  }
196
197
  // F21 — a run-time skip is for what the ENVIRONMENT lacks (an account, a
@@ -200,27 +201,27 @@ export function checkTestGates(content, _ext, ctx) {
200
201
  if (ctx.forbidPlaceholders) {
201
202
  const why = elementSkips(content);
202
203
  if (why.length) {
203
- return `a test skips itself because an element is not on the page (${why[0]}): a missing element the doc names is a failure — assert it instead. Skip only for an account, a variable or data the environment lacks`;
204
+ reasons.push(`a test skips itself because an element is not on the page (${why[0]}): a missing element the doc names is a failure — assert it instead. Skip only for an account, a variable or data the environment lacks`);
204
205
  }
205
206
  }
206
207
  // F17 — relative imports resolve from where the file is written.
207
208
  if (ctx.filePath) {
208
209
  const missing = unresolvedImports(content, ctx.filePath);
209
210
  if (missing.length) {
210
- return `relative import(s) reach no file from ${ctx.filePath}: ${missing.join(', ')}`;
211
+ reasons.push(`relative import(s) reach no file from ${ctx.filePath}: ${missing.join(', ')}`);
211
212
  }
212
213
  }
213
214
  // F18 — no fixed sleeps. A wait for a set time either wastes it or is too
214
215
  // short on a slower machine; wait for the event or the state instead.
215
216
  if (ctx.framework === 'playwright' && ctx.forbidPlaceholders && /\bwaitForTimeout\s*\(/.test(content)) {
216
- return 'page.waitForTimeout() is not used: wait for the element, response or URL the test needs (expect(...).toBeVisible(), page.waitForResponse(), expect(page).toHaveURL())';
217
+ reasons.push('page.waitForTimeout() is not used: wait for the element, response or URL the test needs (expect(...).toBeVisible(), page.waitForResponse(), expect(page).toHaveURL())');
217
218
  }
218
219
  // F19 — a long test timeout needs a reason. A 240 s budget hides a test that
219
220
  // is stuck; the cap makes it fail early with its real error.
220
221
  if (ctx.maxTestTimeoutMs) {
221
222
  const long = longTimeoutsWithoutReason(content, ctx.maxTestTimeoutMs);
222
223
  if (long.length) {
223
- return `test.setTimeout(${long[0]}) is above ${ctx.maxTestTimeoutMs} ms with no "// timeout: <why>" comment: wait for what the test needs instead, or state why it takes that long`;
224
+ reasons.push(`test.setTimeout(${long[0]}) is above ${ctx.maxTestTimeoutMs} ms with no "// timeout: <why>" comment: wait for what the test needs instead, or state why it takes that long`);
224
225
  }
225
226
  }
226
227
  // F20 — one criterion, a few tests. Thirty inputs against a rule that only
@@ -229,7 +230,7 @@ export function checkTestGates(content, _ext, ctx) {
229
230
  if (ctx.maxTestsPerRule && ctx.forbidPlaceholders) {
230
231
  const over = testsPerCriterion(content).find((c) => c.tests > ctx.maxTestsPerRule);
231
232
  if (over) {
232
- return `${over.tests} tests check the same criterion (${over.criterion.slice(0, 80)}…): keep at most ${ctx.maxTestsPerRule}, one per branch, and put further inputs in one parametrised test`;
233
+ reasons.push(`${over.tests} tests check the same criterion (${over.criterion.slice(0, 80)}…): keep at most ${ctx.maxTestsPerRule}, one per branch, and put further inputs in one parametrised test`);
233
234
  }
234
235
  }
235
236
  // F15 — a pending test says why. A reasonless fixme/skip reaches the report
@@ -238,7 +239,7 @@ export function checkTestGates(content, _ext, ctx) {
238
239
  if (ctx.framework === 'playwright') {
239
240
  const bare = reasonlessPending(content);
240
241
  if (bare.length) {
241
- return (`pending test(s) with no reason: "${bare[0]}"… — declare them as ` +
242
+ reasons.push(`pending test(s) with no reason: "${bare[0]}"… — declare them as ` +
242
243
  `test.fixme(title, { annotation: { type: 'fixme', description: '<why>' } }, async () => {})`);
243
244
  }
244
245
  }
@@ -257,14 +258,14 @@ export function checkTestGates(content, _ext, ctx) {
257
258
  const sel = String.raw `(?:getByTestId\s*\(|\[\s*data-(?:testid|cy|test)\s*=)`;
258
259
  const selNearMarker = new RegExp(`(?:${sel}[^\\n]*?\\b${marker}\\b)|(?:\\b${marker}\\b[^\\n]*?${sel})`, 'i');
259
260
  if (selNearMarker.test(active)) {
260
- return 'a runnable test uses a guessed/placeholder selector — ground it in a real component attribute, or defer the test (test.fixme/skip)';
261
+ reasons.push('a runnable test uses a guessed/placeholder selector — ground it in a real component attribute, or defer the test (test.fixme/skip)');
261
262
  }
262
263
  // (b) When the project's real test-hooks are known, any test-id a
263
264
  // runnable test uses that isn't one of them is an invented selector.
264
265
  if (ctx.knownSelectors && ctx.knownSelectors.size > 0) {
265
266
  for (const id of usedTestIds(active)) {
266
267
  if (!ctx.knownSelectors.has(id)) {
267
- return `references selector '${id}' not found in the project's components — ground it or defer the test`;
268
+ reasons.push(`references selector '${id}' not found in the project's components — ground it or defer the test`);
268
269
  }
269
270
  }
270
271
  }
@@ -272,11 +273,13 @@ export function checkTestGates(content, _ext, ctx) {
272
273
  if (ctx.knownTexts) {
273
274
  const missing = usedLabels(active).filter((l) => !ctx.knownTexts.includes(l.toLowerCase()));
274
275
  if (missing.length) {
275
- return `uses ${missing.map((m) => `"${m}"`).slice(0, 3).join(', ')} as an accessible name, label or placeholder, but that text appears nowhere in the app's source or the journey doc — take every name from the code facts or the rules`;
276
+ reasons.push(`uses ${missing.map((m) => `"${m}"`).slice(0, 3).join(', ')} as an accessible name, label or placeholder, but that text appears nowhere in the app's source or the journey doc — take every name from the code facts or the rules`);
276
277
  }
277
278
  }
278
279
  }
279
- return null;
280
+ // Every reason at once: a retry told one problem fixes it and trips on the next.
281
+ const unique = [...new Set(reasons)];
282
+ return unique.length ? (unique.length === 1 ? unique[0] : `${unique[0]}\nAlso:\n${unique.slice(1, 5).map((r) => `- ${r}`).join('\n')}`) : null;
280
283
  }
281
284
  /**
282
285
  * The run-time skips whose condition looks at the page: test.skip(true, …),
@@ -490,6 +493,7 @@ function looksStructurallyClean(content) {
490
493
  t.startsWith('function ') ||
491
494
  t.startsWith('async ') ||
492
495
  t.startsWith('require(') ||
496
+ /^(?:describe|test|it|beforeAll|beforeEach)\s*\(/.test(t) ||
493
497
  t.startsWith("'use strict'") ||
494
498
  t.startsWith('"use strict"');
495
499
  if (!startsWithCode)
@@ -675,6 +679,29 @@ function testCriteria(content) {
675
679
  return { criterion: (id ?? (quoted ? quoted[1] : m[1])).trim().toLowerCase(), tier };
676
680
  });
677
681
  }
682
+ /**
683
+ * The file with at most `max` tests per criterion: the first ones are kept,
684
+ * the extras removed. A mechanical fix — no AI retry for "too many tests".
685
+ */
686
+ export function trimExtraTests(content, max) {
687
+ const blocks = findTestBlocks(content);
688
+ const crit = testCriteria(content);
689
+ const seen = new Map();
690
+ const extra = blocks.filter((_, i) => {
691
+ const c = crit[i]?.criterion;
692
+ if (!c)
693
+ return false;
694
+ const n = (seen.get(c) ?? 0) + 1;
695
+ seen.set(c, n);
696
+ return n > max;
697
+ });
698
+ if (!extra.length)
699
+ return content;
700
+ let out = content;
701
+ for (const b of [...extra].sort((x, y) => y.start - x.start))
702
+ out = out.slice(0, b.start) + out.slice(b.end).replace(/^;?[ \t]*\n?/, '');
703
+ return out;
704
+ }
678
705
  /** How many tests of a file check each criterion (tests with no oracle are not counted). */
679
706
  export function testsPerCriterion(content) {
680
707
  const counts = new Map();
@@ -12,7 +12,7 @@
12
12
  */
13
13
  import { spawn } from 'node:child_process';
14
14
  import { readFileSync, existsSync } from 'node:fs';
15
- import { loadConfig, splitCommand, withNoTools, withReadOnlyTools, resolveFramework, } from '../project/config.js';
15
+ import { isClaude, loadConfig, splitCommand, withNoTools, withReadOnlyTools, resolveFramework, } from '../project/config.js';
16
16
  import { runAiSync, noteAiCall, traceArgs, traceCall } from './ai-usage.js';
17
17
  import { API, WEB, codeOutput, evidence } from '../prompts/blocks.js';
18
18
  import { stripStrayComments } from '../gates/validate.js';
@@ -339,7 +339,9 @@ export function buildPrompt(journey, map, testPath, focus, harnessPath, transpor
339
339
  ``,
340
340
  `NAMES THE PRODUCT GIVES. The doc says WHAT happens; the product names HOW to`,
341
341
  `reach it — a route, a button's label, a command, a table or a field.`,
342
- ` - When the CODE FACTS show that name, use it as written there.`,
342
+ ` - When the CODE FACTS show that name (in ROUTES, SHAPES or DATA), use it`,
343
+ ` as written there: the request fields a handler validates, a table's`,
344
+ ` columns, a status value. need() is never for a name the facts show.`,
343
345
  ` - When they do not (the code is not built yet, or the facts missed it),`,
344
346
  ` write the test anyway and take the name from need('<plain name>'):`,
345
347
  ` const { need } = require('@activate-core/test-forge/reporters/interface.cjs');`,
@@ -753,6 +755,9 @@ function writerCommand(prompt, memory) {
753
755
  if (!memory)
754
756
  return { ...withReadOnlyTools(cmd), input: prompt };
755
757
  const { system, message } = splitShared(prompt);
758
+ // Only the Claude CLI takes a system prompt; any other AI gets the rules at the top of its input.
759
+ if (system && !isClaude(cmd.program))
760
+ return { ...withNoTools(cmd), input: `${system}\n\n${message}` };
756
761
  return system ? { ...withNoTools(cmd, system), input: message } : { ...withNoTools(cmd), input: prompt };
757
762
  }
758
763
  /**
@@ -843,7 +848,8 @@ export async function writeFilesWithAI(files) {
843
848
  const config = loadConfig();
844
849
  const cmd = splitCommand(config.aiCommand);
845
850
  const { program, args } = system ? withNoTools(cmd, system) : withNoTools(cmd);
846
- return splitBatchAnswer(await runWithStdin(program, args, message, config.aiTimeoutMs), files.map((f) => f.path));
851
+ const input = system && !isClaude(cmd.program) ? `${system}\n\n${message}` : message;
852
+ return splitBatchAnswer(await runWithStdin(program, args, input, config.aiTimeoutMs), files.map((f) => f.path));
847
853
  }
848
854
  /** Each file's code from a batch answer, by its marker line. */
849
855
  export function splitBatchAnswer(out, paths) {
@@ -81,6 +81,7 @@ let traceSeq = 0;
81
81
  const STAGES = [
82
82
  [/^=== REPAIR ===$/m, 'repair'],
83
83
  [/^You are a senior QA engineer\. Generate a complete/, 'write'],
84
+ [/^You are a senior QA engineer writing one .* test file/, 'write'],
84
85
  [/^You are a senior test architect/, 'plan'],
85
86
  [/^You are a senior engineer establishing the MECHANICS/, 'grounding'],
86
87
  [/^You review ONE claimed product defect/, 'defect-review'],
@@ -128,7 +129,9 @@ export function traceCall(prompt, raw, started, error) {
128
129
  }
129
130
  });
130
131
  const result = events.find((e) => e.type === 'result');
131
- const text = events.length ? String(result?.result ?? '') : raw;
132
+ // Only a Claude event stream has typed events; another AI's JSON answer is the answer itself.
133
+ const isStream = events.some((e) => typeof e.type === 'string');
134
+ const text = isStream ? String(result?.result ?? '') : raw;
132
135
  const ru = result?.usage ?? {};
133
136
  const tokens = ['input_tokens', 'output_tokens', 'cache_read_input_tokens', 'cache_creation_input_tokens'].reduce((n, k) => n + (Number(ru[k]) || 0), 0);
134
137
  const cost = Number(result?.total_cost_usd) || 0;
@@ -154,7 +157,7 @@ export function traceCall(prompt, raw, started, error) {
154
157
  try {
155
158
  mkdirSync(TRACE_DIR, { recursive: true });
156
159
  writeFileSync(join(TRACE_DIR, `${name}.prompt.md`), prompt);
157
- writeFileSync(join(TRACE_DIR, `${name}.${events.length ? 'stream.jsonl' : 'out.txt'}`), raw);
160
+ writeFileSync(join(TRACE_DIR, `${name}.${isStream ? 'stream.jsonl' : 'out.txt'}`), raw);
158
161
  const u = result?.usage ?? {};
159
162
  appendFileSync(join(TRACE_DIR, 'index.jsonl'), JSON.stringify({
160
163
  seq, stage, file: file ? basename(file) : undefined,
@@ -27,11 +27,31 @@ export function shortReason(why) {
27
27
  export function withoutNames(content) {
28
28
  return content.replace(/^[ \t]*(?:const|let)\s+([A-Za-z_$][\w$]*)\s*=\s*need\(\s*(['"`])[^'"`\n]*\2\s*\);?[ \t]*\n/gm, (line, name) => new RegExp(`\\b${name}\\b`).test(content.replace(line, '')) ? line : '');
29
29
  }
30
+ /** A file top without the `const X = need('…')` lines whose name is still blank. */
31
+ export function withoutBlankNames(head, isBlank) {
32
+ if (!isBlank)
33
+ return head;
34
+ return head.replace(/^[ \t]*(?:const|let)\s+[A-Za-z_$][\w$]*\s*=\s*need\(\s*(['"`])([^'"`\n]*)\1\s*\);?[ \t]*\n/gm, (line, _q, name) => (isBlank(name) ? '' : line));
35
+ }
30
36
  /** The file without the tests that have a problem. */
31
37
  function withoutBlocks(content, problems) {
32
38
  const blocks = [...new Map(problems.map((p) => [p.block.start, p.block])).values()];
33
39
  return blocks.length ? spliceBlocks(content, blocks.map((block) => ({ block, text: '' }))) : content;
34
40
  }
41
+ /**
42
+ * What an answer adds to the file: its test blocks for the rules asked for.
43
+ * An answer that came back as a whole file (imports, an outer describe) or
44
+ * repeats a test the file already has would otherwise land nested or twice.
45
+ */
46
+ export function onlyNewTests(answer, file, rules) {
47
+ if (!/^\s*(?:import\s|const\s+\w+\s*=\s*require\(|(?:test\.)?describe\s*\()/m.test(answer))
48
+ return answer;
49
+ const have = new Set(findTestBlocks(file).map((b) => b.title));
50
+ return findTestBlocks(answer)
51
+ .filter((b) => !have.has(b.title) && (!b.rules.length || b.rules.some((r) => rules.includes(r))))
52
+ .map((b) => answer.slice(b.start, b.end).replace(/^\s*\n/, ''))
53
+ .join('\n\n');
54
+ }
35
55
  /** Is the file complete as it is? */
36
56
  export function fileIsComplete(content, rules, steps, knownTexts) {
37
57
  const { cov, problems } = review(content, rules, steps, knownTexts);
@@ -46,38 +66,78 @@ export async function completeFile(content, ctx) {
46
66
  const drop = (ps) => { for (const p of ps)
47
67
  if (!dropped.has(p.title))
48
68
  dropped.set(p.title, p.reason); };
49
- if (first.cov.missing.length || first.problems.length) {
50
- out = withoutNames(withoutBlocks(content, first.problems));
51
- drop(first.problems);
69
+ const locked = ctx.locked ?? new Set();
70
+ const before = new Set(findTestBlocks(content).map((b) => b.title));
71
+ // A test that passed is never touched. A kept file's other tests are only
72
+ // reported: a check added later must not quietly shrink a working suite.
73
+ // Only the tests asked for again (redo) go, and so do new tests just written.
74
+ const removable = (p) => !locked.has(p.title) && (!ctx.kept || !before.has(p.title) || !!ctx.redo?.(p.block));
75
+ const redoBlocks = ctx.kept && ctx.redo
76
+ ? findTestBlocks(content).filter((b) => !locked.has(b.title) && ctx.redo(b))
77
+ : [];
78
+ const redo = redoBlocks
79
+ .filter((b) => !first.problems.some((p) => p.block.start === b.start))
80
+ .map((b) => ({ title: b.title, rules: b.rules, reason: `it waits on a name the code now has: write it again from the code with the same title, "${b.title}"`, block: b }));
81
+ const firstDrop = [...first.problems.filter(removable), ...redo];
82
+ let failedRedo = false;
83
+ const refusedRules = [];
84
+ const reported = first.problems.filter((p) => !removable(p));
85
+ if (reported.length) {
86
+ notes.push(`left ${reported.length} earlier test(s) in place that a check flags (${[...new Set(reported.map((p) => shortReason(p.reason)))].slice(0, 3).join('; ')}) — not removed, they were already in the file`);
87
+ }
88
+ if (first.cov.missing.length || firstDrop.length) {
89
+ out = withoutNames(withoutBlocks(content, firstDrop));
90
+ drop(firstDrop);
52
91
  // Only rules the doc has are written again: an id with no text makes the writer invent one.
53
92
  const known = new Set(ctx.memory.rules.filter((r) => r.tag).map((r) => r.id));
54
- const left = ruleCoverage(out, ctx.rules).missing.filter((id) => known.has(id));
93
+ const left = [...new Set([...ruleCoverage(out, ctx.rules).missing, ...redo.flatMap((p) => p.rules)])]
94
+ .filter((id) => known.has(id) && ctx.rules.includes(id) && !(ctx.skip?.has(id) && !redo.some((p) => p.rules.includes(id))));
55
95
  if (left.length) {
56
96
  const blocks = findTestBlocks(out);
57
- const head = out.slice(0, blocks[0] ? blocks[0].start : Math.min(out.length, 4000)).slice(0, 4000);
97
+ // The file's top as the writer sees it: a constant that still waits on a
98
+ // blank name is left out, or the new tests reuse it and wait again.
99
+ const head = withoutBlankNames(out.slice(0, blocks[0] ? blocks[0].start : Math.min(out.length, 4000)).slice(0, 4000), ctx.isBlank);
58
100
  const title = /test\.describe(?:\.\w+)?\(\s*(['"`])((?:(?!\1).)*)\1/.exec(out)?.[2] ?? ctx.journey.name;
59
101
  const slice = journeySlice(ctx.memory, left);
60
102
  calls = 1;
61
103
  try {
62
- const tests = await ctx.write(ctx.journey, ctx.path, head, title, '', ctx.harnessPath, ctx.transport, factsSlice(ctx.facts, slice.controls), { journey: slice.text, helpers: ctx.helpers, rules: left }, describeProblems(first.problems, first.cov));
63
- const candidate = appendTests(out, tests);
104
+ const tests = await ctx.write(ctx.journey, ctx.path, head, title, '', ctx.harnessPath, ctx.transport, factsSlice(ctx.facts, slice.controls), { journey: slice.text, helpers: ctx.helpers, rules: left }, describeProblems(firstDrop, first.cov));
105
+ const candidate = (ctx.fix ?? ((c) => c))(appendTests(out, onlyNewTests(tests, out, left)));
64
106
  const refused = ctx.check(candidate);
65
- if (refused)
107
+ if (refused) {
66
108
  notes.push(`tests written again for ${left.join(', ')} were refused (${refused.split('\n')[0]})`);
109
+ failedRedo = true;
110
+ refusedRules.push(...left);
111
+ }
67
112
  else
68
113
  out = candidate;
69
114
  }
70
115
  catch (e) {
71
116
  notes.push(`the call to write ${left.join(', ')} again failed — ${e.message.split('\n')[0].slice(0, 120)}`);
117
+ failedRedo = true;
72
118
  }
73
119
  // What the second answer still gets wrong is removed, not sent again.
74
- const second = testProblems(out, { steps: ctx.steps, knownTexts: ctx.knownTexts });
120
+ const second = testProblems(out, { steps: ctx.steps, knownTexts: ctx.knownTexts }).filter(removable);
75
121
  if (second.length) {
76
122
  out = withoutNames(withoutBlocks(out, second));
77
123
  drop(second);
78
124
  }
79
125
  }
80
126
  }
127
+ // A kept file whose tests were taken out to be written again, and the new
128
+ // ones did not come: it stays as it was. A waiting test beats no test.
129
+ if (ctx.kept && failedRedo && redo.length + firstDrop.length > 0) {
130
+ notes.push('kept the file as it was: the tests to write again did not come back');
131
+ out = content;
132
+ dropped.clear();
133
+ }
134
+ // The last word, checked mechanically: every test that passed is still here.
135
+ const lost = [...locked].filter((t) => before.has(t) && !findTestBlocks(out).some((b) => b.title === t));
136
+ if (lost.length) {
137
+ notes.push(`kept the file as it was: the change would have lost ${lost.length} test(s) that passed, e.g. "${lost[0].slice(0, 60)}"`);
138
+ out = content;
139
+ dropped.clear();
140
+ }
81
141
  // A cannot-test line that is no longer true (its rule has a test now) or never
82
142
  // was a reason ("the locator is unknown") does not stay in the file.
83
143
  const has = new Set(findTestBlocks(out).flatMap((b) => b.rules));
@@ -95,5 +155,5 @@ export async function completeFile(content, ctx) {
95
155
  }
96
156
  if (cov.missing.length)
97
157
  notes.push(`no test for ${cov.missing.join(', ')}`);
98
- return { content: out, tested, cannotTest: cov.cannotTest, missing: cov.missing, dropped: [...dropped].map(([t, why]) => `${t} — ${why}`), calls, notes };
158
+ return { content: out, tested, cannotTest: cov.cannotTest, missing: cov.missing, dropped: [...dropped].map(([t, why]) => `${t} — ${why}`), calls, notes, refusedRules: [...new Set([...refusedRules, ...(calls ? ruleCoverage(out, ctx.rules).missing.filter((id) => !refusedRules.includes(id) && !failedRedo) : [])])] };
99
159
  }
@@ -246,7 +246,8 @@ export function oracleClaim(t, content) {
246
246
  return `${t.title} ${NOT_CLAIMED}`;
247
247
  }
248
248
  /** A [spec] line that carries the client's words, quoted or pointing at the quote above it ("source as above"). */
249
- export const HAS_SOURCE = /\((?:source(?::\s*(?:[A-Z][A-Z0-9]+-\d+:?\s*)?["“]|:?\s*(?:as|same)\b)|same source\b)/i;
249
+ // A quoted source, a ticket key on its own ("source: SHOP-1") — the ticket is the source — or "same as".
250
+ export const HAS_SOURCE = /\((?:source(?::\s*(?:[A-Z][A-Z0-9]+-\d+:?\s*)?["“]|:\s*[A-Z][A-Z0-9]+-\d+\b|:?\s*(?:as|same)\b)|same source\b)/i;
250
251
  /** Playwright's report of a locator that matched more than one element. */
251
252
  const AMBIGUOUS = /strict mode violation|resolved to \d+ elements/i;
252
253
  /** An application file:line, the only evidence a defect may rest on. */
@@ -40,6 +40,11 @@ export const HARNESS_CONTRACT = [
40
40
  `- resetRateLimits(): Promise<void> (H7)`,
41
41
  `- REQUIRED_PACKAGES: string[] (H6)`,
42
42
  `- a 'run' role alongside the fixture roles (H8)`,
43
+ `- insert(table, row): Promise<id>, one row of scenario data written straight`,
44
+ ` into the test database (H10)`,
45
+ `- DERIVED and AREA (H10)`,
46
+ `- update(table, match, values): Promise<number>, columns set on the rows`,
47
+ ` matching every key of match (H10)`,
43
48
  `Jest areas only`,
44
49
  `- call(method, path, { body?, headers?, role? }):`,
45
50
  ` Promise<{ status, headers, body }>`,
@@ -177,6 +182,35 @@ export const HARNESS_CONTRACT = [
177
182
  ` 'unknown step "<name>" (steps: …)'. If there is no such flow, export`,
178
183
  ` STEPS = {} and have openStep resolve`,
179
184
  ` 'this product has no multi-step flow'.`,
185
+ ``,
186
+ `H10. Scenario data. A scenario describes people and their state (an operator`,
187
+ ` with a location, services, working hours and earlier jobs; a setting`,
188
+ ` switched on). insert() and update() let a test build exactly that:`,
189
+ ` - insert fills every NOT NULL column without a default that row leaves out`,
190
+ ` with a value of its type (read the columns from the database), returns`,
191
+ ` the new id, and remembers the row.`,
192
+ ` - update remembers the previous values of the rows it changes.`,
193
+ ` - Each test starts from a clean slate: the rows a test writes, and what it`,
194
+ ` updates, are undone right after that test (last first), so one test's`,
195
+ ` data never takes part in the next. The testkit does this itself when it`,
196
+ ` is loaded: it notes where the record stands before each test and undoes`,
197
+ ` back to that point after it (Jest: the global beforeEach/afterEach;`,
198
+ ` Playwright: test.beforeEach/afterEach). Tests do nothing for it.`,
199
+ ` - What a beforeAll writes stays for the whole file; disconnect() undoes it`,
200
+ ` and anything left, so every run leaves the database as it found it.`,
201
+ ` - Both refuse a database that is not clearly a test one, like reset() (H5).`,
202
+ ` - Some values a scenario sets are not columns: the product works them out`,
203
+ ` from other records (a standing from past jobs, an average per day, a`,
204
+ ` count). Read in the code how each is worked out. insert() accepts each as`,
205
+ ` a virtual column named _tf_<name> (lower case, underscores) and, instead`,
206
+ ` of writing it, writes the records that give exactly that value (and are`,
207
+ ` undone with the row). Export DERIVED: { _tf_<name>: 'how it is made' }`,
208
+ ` for each, so tests can use them.`,
209
+ ` - Export AREA: { latitude, longitude } (or the project's own key, when`,
210
+ ` the product picks among records by something else): a place no record`,
211
+ ` already in the database reaches, found by querying it. Scenarios where`,
212
+ ` the product picks one record among many are built there, so data that`,
213
+ ` was already in the database never takes part.`,
180
214
  ].join('\n');
181
215
  /**
182
216
  * Stamp written into every generated testkit: the contract it was built
@@ -68,3 +68,15 @@ export function recordInterface(found, path = INTERFACE_PATH) {
68
68
  }
69
69
  return blankNames(Object.keys(found), values);
70
70
  }
71
+ /** Blank these names again (a value the code no longer has): their tests wait instead of failing for the wrong reason. */
72
+ export function clearInterface(names, path = INTERFACE_PATH) {
73
+ if (!names.length)
74
+ return;
75
+ const values = readInterface(path);
76
+ for (const n of names)
77
+ if (n in values)
78
+ values[n] = '';
79
+ mkdirSync(dirname(path), { recursive: true });
80
+ const sorted = Object.fromEntries(Object.entries(values).sort(([a], [b]) => a.localeCompare(b)));
81
+ writeFileSync(path, `${JSON.stringify(sorted, null, 2)}\n`, 'utf-8');
82
+ }
@@ -206,6 +206,10 @@ const RUNTIME_HELPERS = {
206
206
  fillFrameInputs: 'async fillFrameInputs(page, framePrefix, { name: value }) — fills inputs inside an embedded iframe (card fields).',
207
207
  watchDialogs: 'watchDialogs(page) → string[] — accepts browser alerts and records their messages.',
208
208
  };
209
+ const KIT_CONTRACT = {
210
+ insert: "resolves to the new row's id itself (a number or key, not a row object): const id = await insert('users', {...})",
211
+ update: 'resolves to the number of rows changed',
212
+ };
209
213
  export function helperApi(harnessSource) {
210
214
  const names = new Set();
211
215
  const cjs = /module\.exports\s*=\s*\{([\s\S]*?)\}/.exec(harnessSource);
@@ -262,7 +266,9 @@ export function helperApi(harnessSource) {
262
266
  if (text)
263
267
  doc.unshift(text);
264
268
  }
265
- out.push(`- ${sig.length > 200 ? `${sig.slice(0, 197)}…` : sig}${doc.length ? ` — ${doc.join(' ')}` : ''}`);
269
+ // The scenario-data writers' contract (H10) holds whatever comment sits above them.
270
+ const kit = KIT_CONTRACT[name] && /^(?:async\s+)?function/.test(sig) ? KIT_CONTRACT[name] : doc.join(' ');
271
+ out.push(`- ${sig.length > 200 ? `${sig.slice(0, 197)}…` : sig}${kit ? ` — ${kit}` : ''}`);
266
272
  }
267
273
  return out.join('\n');
268
274
  }
@@ -0,0 +1,24 @@
1
+ /**
2
+ * testforge — the tests that have passed, as the run recorded them.
3
+ *
4
+ * A test that passed once checked the product for real: generate never
5
+ * removes or rewrites it again. The reporters add to testforge/passed.json
6
+ * after every run ({ "<file relative to the project>": ["title", …] }); a
7
+ * later failure does not take a test off the list — that is a regression to
8
+ * report, not a test to rewrite. `generate --force` is the only way past it.
9
+ */
10
+ import { readFileSync } from 'node:fs';
11
+ import { join, relative, resolve } from 'node:path';
12
+ import { STATE_DIR } from '../paths.js';
13
+ export const PASSED_PATH = join(STATE_DIR, 'passed.json');
14
+ /** Titles of the tests in `file` that have passed in a recorded run. */
15
+ export function passedTitles(file, path = PASSED_PATH) {
16
+ try {
17
+ const all = JSON.parse(readFileSync(path, 'utf-8'));
18
+ const rel = relative(process.cwd(), resolve(file)).split('\\').join('/');
19
+ return new Set(Array.isArray(all[rel]) ? all[rel] : []);
20
+ }
21
+ catch {
22
+ return new Set();
23
+ }
24
+ }