styleproof 3.4.0 → 3.6.0

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/CHANGELOG.md CHANGED
@@ -7,6 +7,112 @@ this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
7
7
 
8
8
  ## [Unreleased]
9
9
 
10
+ ## [3.6.0] - 2026-07-03
11
+
12
+ ### Changed
13
+
14
+ - Crawl candidate collection collapses inherited-cursor subtrees to their
15
+ OUTERMOST clickable. `cursor` is an inherited CSS property, so a clickable card
16
+ makes every descendant compute `cursor: pointer`; each became its own candidate
17
+ even though clicking any of them just bubbles to the card's handler (the same
18
+ surface). Each redundant candidate paid a drive + verified reset — the dominant
19
+ cost of a large crawl. Measured on a 241-class design: base candidates 557 → 22,
20
+ whole coverage crawl ~40min → ~6min (240/241, unchanged). Semantic controls
21
+ nested in a clickable container are still kept.
22
+ - `--until-covered` reaches deep coverage: it now terminates on the queue
23
+ draining (or full coverage), not a fixed "N surfaces without a new class"
24
+ plateau — a plateau cut the crawl off before a productive deep state (an
25
+ automation whose expandable run row is its last candidate) was ever swept.
26
+ Coverage mode also (a) prunes the queue to surfaces that add new render
27
+ vocabulary — a structural repeat like the ninth agent's identical dossier is
28
+ captured once but not re-drilled — and (b) uses the breadth-first queue rather
29
+ than the depth-first in-place descent, so distinct components are reached fast
30
+ and each drilled once via now-reliable resets. Measured on a 241-class design:
31
+ 240/241 (the one gap a post-decision state reachable only with a setup step),
32
+ where a plateau-stopped crawl left the deep automation run-detail unmapped.
33
+
34
+ ### Fixed
35
+
36
+ - Reset-replay to depth >= 2 no longer fails: the crawler's structural
37
+ fingerprint counted StyleProof's OWN injected hover-sink `<div>` (added during
38
+ a capture, so present when a state is captured in place but absent on a fresh
39
+ reset+replay). Every such reset failed its fingerprint verification, so any
40
+ surface reachable ONLY by re-driving from a deep state — a pairwise mode
41
+ combination like a tab's edit view — was silently lost. The fingerprint now
42
+ excludes the sink (and framework route-announcers). Measured: depth-2..5
43
+ resets went from all-fail to all-pass.
44
+
45
+ ### Fixed
46
+
47
+ - Report tables never show an equal-looking Before/After pair for a real diff.
48
+ A colour embedded in a compound value (gradient, shadow) no longer stands in
49
+ for the whole value — `toHex` only converts values that ARE a colour — and
50
+ long value pairs (gradients, data URIs) are excerpted around the differing
51
+ substring with a little shared context on each side.
52
+ - Report values render verbatim: display rounding of decimals is gone (alpha
53
+ `0.18` was shown as `0.2`, and a real `0.18 → 0.2` change could be dropped
54
+ as a no-op after rounding).
55
+
56
+ ### Changed
57
+
58
+ - Crawl reset settle no longer waits on `networkidle`: it polls DOM-growth to
59
+ detect the mounted app (unchanged) and then waits for `document.fonts.ready`
60
+ specifically (fonts are part of the computed style the diff compares, so they
61
+ must be loaded — but that is the deterministic signal, cache-warm on repeat
62
+ loads). networkidle waited a 500ms idle window ON TOP of a cross-origin font
63
+ sheet that lingered ~1s per load with no bearing on readiness; since every
64
+ state reset re-navigates, that dominated crawl time. Measured: settle 911ms →
65
+ 105ms per reset (~8.7x), whole-crawl surface rate ~5/min → ~17/min.
66
+
67
+ - Crawl is now breadth-first (was depth-first): every shallow surface — nav
68
+ tabs, opened panels, the tabs inside a dossier — is exhausted before drilling
69
+ deeper. Depth-first starved breadth: one append-generator branch drilled past
70
+ depth 20 while sibling tabs sat unvisited, so real surfaces (an OAuth card, a
71
+ skills grid, a markdown editor) went uncaptured — measured live as 48 defined
72
+ classes never rendered across ~12 surfaces. Dedup is set-based, so order
73
+ changes only WHICH surface is found first, never the final set.
74
+ - `maxDepth` default 1000 -> 16: exhaustive for real UI (nothing human-navigable
75
+ is 16 clicks from load) while terminating append-generator chains, whose every
76
+ appended node is a fresh tag-path identity. The coverage verifier still names
77
+ any class left unrendered, so a too-low cap fails loudly rather than lying.
78
+
79
+ - The never-click guard now also covers state-mutating verbs (rotate,
80
+ provision, seal, regenerate, renew): mapping must not mutate, and a mutating
81
+ control that persists after its click re-labels its surroundings with fresh
82
+ data on every press — an unbounded mutation farm a crawl must not walk.
83
+ Observed live: credential-rotation cells minted new identities per press,
84
+ inflating a crawl past 470 surfaces at depth 23. Their render states are
85
+ seed-data territory (and anything unreached is named by the verifier).
86
+ - Freshly-opened surfaces are swept IN PLACE (forward-drive), not only queued
87
+ for a later reset+replay. Reaching a surface by a forward click is reliable;
88
+ re-reaching it by reset is slow (a reset per candidate) and, at depth, starved
89
+ — a dossier's many filter combinations flood the queue and bury the deep
90
+ sweep, so an expanded run row inside an expanded job was captured 0 times
91
+ despite being reachable (now 29). The descent drives only fresh controls new
92
+ to the surface (excludes parent-present mode-switchers), so pairwise mode
93
+ coverage is unchanged and the lattice stays pairwise.
94
+
95
+ ### Added
96
+
97
+ - `--until-covered`: stop the crawl as soon as every class the page's
98
+ stylesheets define has rendered (full coverage) or coverage stops improving
99
+ (no new class for a plateau of surfaces). Turns an exhaustive crawl into a
100
+ fast coverage check that stops once it has SEEN everything, instead of
101
+ enumerating every combinatorial surface that adds no new vocabulary. Opt-in;
102
+ exhaustive remains the default.
103
+
104
+ ## [3.5.0] - 2026-07-02
105
+
106
+ ### Added
107
+
108
+ - Parallel crawl: `--workers <n>` (default 4) sweeps queued states concurrently,
109
+ each worker on its own browser context. The surface set is identical to a
110
+ serial crawl — dedup sets are shared, and a state's children only join the
111
+ queue when its sweep completes, so family retry never reads a half-built
112
+ changer registry. Only dup-key suffix attribution can vary with timing; pass
113
+ `--workers 1` for byte-stable keys. Exhaustive runs drop from hours to
114
+ minutes on lattice-heavy designs.
115
+
10
116
  ## [3.4.0] - 2026-07-02
11
117
 
12
118
  ### Fixed
package/README.md CHANGED
@@ -424,7 +424,7 @@ styleproof-diff design .styleproof/maps/current # diff the whole
424
424
 
425
425
  It's **exhaustive by default**: the crawl stops when there is nothing left to drive — every control tried once, every structurally new surface captured — not at a budget. Termination is guaranteed by dedup (controls dedup by selector, surfaces by a structural fingerprint), and the `--max-depth` / `--max-actions` / `--max-states` flags exist only as deliberate throttles. It's deterministic (document order; the same surface reached two ways is captured once) and self-settling — it waits for an async app (React/Vue/Babel that boots after `load`) to mount before reading, so a bare crawl of a client-rendered page still captures the mounted UI.
426
426
 
427
- What makes exhaustive affordable is that the sweep works **in place**: standing in a state, each control is clicked right where the page is, and a cheap DOM fingerprint decides what happened — a no-op click costs nothing, and only a state-changing click pays a reset (fresh navigation + replay of the click-path), which is then **verified by fingerprint** so children are never attributed to the wrong parent. New surfaces are captured at every width the moment they're reached — a deep or animated click-path is never re-driven to capture, so it can't be the thing that drops a surface. Progress streams as it goes, one line per captured surface.
427
+ What makes exhaustive affordable is that the sweep works **in place**: standing in a state, each control is clicked right where the page is, and a cheap DOM fingerprint decides what happened — a no-op click costs nothing, and only a state-changing click pays a reset (fresh navigation + replay of the click-path), which is then **verified by fingerprint** so children are never attributed to the wrong parent. New surfaces are captured at every width the moment they're reached — a deep or animated click-path is never re-driven to capture, so it can't be the thing that drops a surface. Progress streams as it goes, one line per captured surface. And it's **parallel by default** — `--workers <n>` (default 4) sweeps states concurrently on isolated browser contexts with the exact same surface set as a serial crawl (dedup is shared; children only enter the queue when their parent's sweep completes); `--workers 1` if you want byte-stable dup-key attribution.
428
428
 
429
429
  **And it proves nothing was missed.** After the crawl, StyleProof compares every class the page's own stylesheets define (read from the parsed CSSOM) against the classes actually rendered across the captured surfaces, and prints what — if anything — was never seen. `--require-full-coverage` turns any residue into exit code 4, so "the design is fully covered" is a CI-checkable property, not a judgement call. What's left is either dead CSS (delete it) or a state the crawl couldn't reach (drive it with a spec, or file the gap).
430
430
 
@@ -51,6 +51,8 @@ whole surface: --crawl
51
51
  --no-data-states skip the automatic loading/error captures of the entry page
52
52
  (on by default: data requests stalled → loading skeleton;
53
53
  fulfilled with 500 → error render)
54
+ --workers <n> concurrent sweep workers (default 4); same surface set as a
55
+ serial crawl — pass 1 for byte-stable key attribution
54
56
  --max-depth <n> throttle recursion depth (default: unbounded)
55
57
  --max-actions <n> throttle controls tried per state (default: unbounded)
56
58
  --max-states <n> throttle total surfaces (default: unbounded)
@@ -105,6 +107,10 @@ async function runCrawl() {
105
107
  resetStorage: opts.resetStorage,
106
108
  setup: setupSteps,
107
109
  dataStates: opts.dataStates,
110
+ stopWhenCovered: opts.untilCovered,
111
+ workers: opts.workers,
112
+ // each worker page in its OWN context, so storage resets can't interfere
113
+ newPage: async () => (await browser.newContext()).newPage(),
108
114
  // Stream each surface as it is captured, so progress is visible live and an
109
115
  // interrupted run still shows exactly what it mapped.
110
116
  onSurface: (s, ok) =>
@@ -56,6 +56,11 @@ export type CaptureUrlOptions = {
56
56
  /** crawl: exit non-zero unless every class the page's stylesheets define was
57
57
  * rendered in at least one captured surface (default false — report only). */
58
58
  requireFullCoverage: boolean;
59
+ /** crawl: stop as soon as coverage is complete (every defined class seen) or
60
+ * has converged (no new class for a plateau of surfaces). Turns the crawl
61
+ * into a FAST coverage check that stops once it has seen everything, instead
62
+ * of enumerating every combinatorial surface. Default false (exhaustive). */
63
+ untilCovered: boolean;
59
64
  /** crawl: JSON file of deterministic setup steps (login, unlock, seed input)
60
65
  * run after every fresh navigation. See {@link loadSetupSteps}. */
61
66
  setupFile?: string;
@@ -65,6 +70,8 @@ export type CaptureUrlOptions = {
65
70
  /** crawl: also capture automatic `loading`/`error` data states of the entry
66
71
  * page (default true). */
67
72
  dataStates: boolean;
73
+ /** crawl: concurrent sweep workers (default 4). 1 = byte-stable key attribution. */
74
+ workers: number;
68
75
  };
69
76
  /**
70
77
  * Parse `styleproof-capture` argv into options. Pure and throwing so the CLI
@@ -31,7 +31,9 @@ const DEFAULTS = {
31
31
  maxStates: 100000,
32
32
  resetStorage: true,
33
33
  requireFullCoverage: false,
34
+ untilCovered: false,
34
35
  dataStates: true,
36
+ workers: 4,
35
37
  };
36
38
  function positiveNumber(raw, flag) {
37
39
  const n = Number(raw);
@@ -62,6 +64,7 @@ const VALUE_FLAGS = {
62
64
  '--max-actions': (o, v) => (o.maxActionsPerState = positiveNumber(v, '--max-actions')),
63
65
  '--max-states': (o, v) => (o.maxStates = positiveNumber(v, '--max-states')),
64
66
  '--setup': (o, v) => (o.setupFile = v),
67
+ '--workers': (o, v) => (o.workers = positiveNumber(v, '--workers')),
65
68
  };
66
69
  const BOOL_FLAGS = {
67
70
  '--screenshots': (o) => (o.screenshots = true),
@@ -69,6 +72,7 @@ const BOOL_FLAGS = {
69
72
  '--crawl': (o) => (o.crawl = true),
70
73
  '--no-reset-storage': (o) => (o.resetStorage = false),
71
74
  '--require-full-coverage': (o) => (o.requireFullCoverage = true),
75
+ '--until-covered': (o) => (o.untilCovered = true),
72
76
  '--data-states': (o) => (o.dataStates = true),
73
77
  '--no-data-states': (o) => (o.dataStates = false),
74
78
  };
@@ -117,8 +121,10 @@ export function parseCaptureUrlArgs(argv) {
117
121
  maxStates: DEFAULTS.maxStates,
118
122
  resetStorage: DEFAULTS.resetStorage,
119
123
  requireFullCoverage: DEFAULTS.requireFullCoverage,
124
+ untilCovered: DEFAULTS.untilCovered,
120
125
  setupFile: undefined,
121
126
  dataStates: DEFAULTS.dataStates,
127
+ workers: DEFAULTS.workers,
122
128
  };
123
129
  const positional = [];
124
130
  for (let i = 0; i < argv.length; i++)
@@ -90,6 +90,14 @@ export type SurfaceCrawlOptions = {
90
90
  * requests stalled — the skeleton) and `error` (data requests fulfilled with
91
91
  * a 500). Default true; states that render identically to base are skipped. */
92
92
  dataStates?: boolean;
93
+ /** Coverage-oriented termination: stop as soon as every class the page's
94
+ * stylesheets define has been rendered (full coverage) OR coverage stops
95
+ * improving (no new class for a plateau of surfaces — it has converged short
96
+ * of full, and the rest is dead CSS or a genuinely-gated state). Off by
97
+ * default (exhaustive: map every surface). On, the crawl is a fast coverage
98
+ * check — it stops the moment it has SEEN everything, instead of enumerating
99
+ * every combinatorial surface that adds no new vocabulary. */
100
+ stopWhenCovered?: boolean;
93
101
  /** Throttle: recursion depth into opened surfaces (base = 0). Default: unbounded. */
94
102
  maxDepth: number;
95
103
  /** Throttle: fresh controls driven per state. Default: unbounded — try them all. */
@@ -101,6 +109,15 @@ export type SurfaceCrawlOptions = {
101
109
  /** Called as each surface is recorded (captured=false when its full capture failed).
102
110
  * Lets CLIs stream progress instead of reporting only at the end. */
103
111
  onSurface?: (surface: CrawledSurface, captured: boolean) => void;
112
+ /** Concurrent sweep workers (default 4). Each worker gets its own page from
113
+ * `newPage`; without a factory the crawl runs single-page regardless. The
114
+ * surface SET is identical to a serial crawl (same dedup sets); only which
115
+ * path first claims a shape — and so dup-key suffixes — can vary run to run.
116
+ * Use workers: 1 for byte-stable key attribution. */
117
+ workers?: number;
118
+ /** Factory for worker pages — create each in its OWN browser context so
119
+ * storage resets cannot interfere across concurrent sweeps. */
120
+ newPage?: () => Promise<Page>;
104
121
  };
105
122
  export declare const CRAWL_DEFAULTS: {
106
123
  height: number;
@@ -109,14 +126,10 @@ export declare const CRAWL_DEFAULTS: {
109
126
  maxActionsPerState: number;
110
127
  maxStates: number;
111
128
  resetStorage: boolean;
129
+ workers: number;
112
130
  };
113
131
  /** Run the caller's deterministic setup steps (login, unlock, seed input). A
114
132
  * non-optional step that fails throws loudly — a half-established gate must
115
133
  * never silently crawl the ungated page instead. */
116
134
  export declare function runSetup(page: Page, steps: SetupStep[]): Promise<void>;
117
- /**
118
- * Crawl `opts.url` and capture every reachable surface at every width — runs to
119
- * natural termination by default. Returns the surfaces mapped (with the click-path
120
- * that reached each), how many actions were tried/skipped, and captured/failed.
121
- */
122
135
  export declare function crawlAndCapture(page: Page, opts: SurfaceCrawlOptions): Promise<CrawlReport>;
@@ -7,10 +7,18 @@ import { detectViewportWidths } from './breakpoints.js';
7
7
  export const CRAWL_DEFAULTS = {
8
8
  height: 900,
9
9
  screenshots: true,
10
- maxDepth: 1000,
10
+ // ponytail: depth 16 is exhaustive for real UI — no human-navigable surface
11
+ // is 16 clicks from load. Past that you're not finding surfaces, you're
12
+ // riding an append-generator (a composer that adds a row per click) whose
13
+ // every appended node is a fresh tag-path identity, so it recurses forever.
14
+ // The cap terminates those chains; the coverage verifier still NAMES any
15
+ // class left unrendered, so a too-low cap fails loudly rather than lying.
16
+ // Raise with --max-depth if a design genuinely nests deeper.
17
+ maxDepth: 16,
11
18
  maxActionsPerState: 100000,
12
19
  maxStates: 100000,
13
20
  resetStorage: true,
21
+ workers: 4,
14
22
  };
15
23
  function slug(value) {
16
24
  return (value
@@ -40,7 +48,7 @@ function deriveKey(steps, used) {
40
48
  /* c8 ignore start */ // fallow-ignore-next-line complexity
41
49
  function collectClickable() {
42
50
  const SEMANTIC = 'button,summary,[role="button"],[role="tab"],[role="menuitem"],[role="combobox"],select,form';
43
- const DANGER = /\b(delete|remove|destroy|logout|log ?out|sign ?out|publish|deploy|pay|purchase|buy|checkout|archive|disconnect|revoke|reset|wipe|drop)\b/i;
51
+ const DANGER = /\b(delete|remove|destroy|logout|log ?out|sign ?out|publish|deploy|pay|purchase|buy|checkout|archive|disconnect|revoke|reset|wipe|drop|rotate|provision|seal|regenerate|renew)\b/i;
44
52
  const esc = (v) => CSS.escape(v);
45
53
  const quote = (v) => JSON.stringify(v);
46
54
  const visible = (el) => {
@@ -77,6 +85,19 @@ function collectClickable() {
77
85
  .replace(/\s+/g, ' ')
78
86
  .trim()
79
87
  .slice(0, 80) || el.tagName.toLowerCase();
88
+ const identityFor = (el) => {
89
+ // Tag-path only — NO classes and NO indices: classes carry state (body.alt,
90
+ // .on) and would re-contextualize the same control per mode; indices carry
91
+ // position and drift on re-render. Tag ancestry + label is what stays
92
+ // stable across every context the same logical control appears in.
93
+ const parts = [];
94
+ let cur = el;
95
+ while (cur && cur !== document.documentElement) {
96
+ parts.unshift(cur.tagName.toLowerCase());
97
+ cur = cur.parentElement;
98
+ }
99
+ return `${parts.join('>')}|${labelFor(el)}|${el.getAttribute('role') ?? ''}`;
100
+ };
80
101
  // Semantic controls first (stable, meaningful), then anything styled clickable.
81
102
  // `grab` counts too: a draggable card is routinely ALSO a click target (open on
82
103
  // click, drag to move), and we never drag — el.click() fires no drag gesture.
@@ -85,7 +106,28 @@ function collectClickable() {
85
106
  if (pool.has(el))
86
107
  continue;
87
108
  const cursor = getComputedStyle(el).cursor;
88
- if (cursor === 'pointer' || cursor === 'grab')
109
+ if (cursor !== 'pointer' && cursor !== 'grab')
110
+ continue;
111
+ // `cursor` is an INHERITED property: a clickable card makes every descendant
112
+ // compute cursor:pointer, and clicking any descendant just bubbles to the
113
+ // card's own handler — the SAME surface. Left unchecked, a card with N
114
+ // children becomes N+1 candidates, each paying a drive + verified reset to
115
+ // map one surface (the dominant cost of a large crawl). Add only the
116
+ // OUTERMOST clickable in an inherited-cursor subtree: skip when an ancestor
117
+ // is already a candidate. Semantic controls (button, a, [role]) are seeded
118
+ // above and skipped by the guard at the top of the loop, so a real button
119
+ // nested inside a clickable card is never dropped. (querySelectorAll walks
120
+ // document order, so an ancestor is always pooled before its descendants.)
121
+ let anc = el.parentElement;
122
+ let nested = false;
123
+ while (anc && anc !== document.body) {
124
+ if (pool.has(anc)) {
125
+ nested = true;
126
+ break;
127
+ }
128
+ anc = anc.parentElement;
129
+ }
130
+ if (!nested)
89
131
  pool.add(el);
90
132
  }
91
133
  // Neutral text inputs are typed automatically with a deterministic value —
@@ -117,6 +159,7 @@ function collectClickable() {
117
159
  out.push({
118
160
  action: 'fill-input',
119
161
  selector,
162
+ identity: identityFor(el),
120
163
  label: labelFor(el) === el.tagName.toLowerCase() ? (el.getAttribute('placeholder') ?? 'input') : labelFor(el),
121
164
  reason: 'auto-fill',
122
165
  value: AUTO_VALUE[kind] ?? 'sample text',
@@ -139,12 +182,21 @@ function collectClickable() {
139
182
  if (el instanceof HTMLSelectElement) {
140
183
  const next = [...el.options].find((o) => !o.disabled && o.value !== el.value);
141
184
  if (next)
142
- out.push({ action: 'select-option', selector, label, reason: 'select-option', value: next.value, unsafe });
185
+ out.push({
186
+ action: 'select-option',
187
+ selector,
188
+ identity: identityFor(el),
189
+ label,
190
+ reason: 'select-option',
191
+ value: next.value,
192
+ unsafe,
193
+ });
143
194
  }
144
195
  else {
145
196
  out.push({
146
197
  action: 'click',
147
198
  selector,
199
+ identity: identityFor(el),
148
200
  label,
149
201
  reason: el.getAttribute('role') === 'tab' ? 'tab' : 'click',
150
202
  unsafe,
@@ -179,6 +231,17 @@ function domShape() {
179
231
  for (const el of document.body.getElementsByTagName('*')) {
180
232
  if (SKIP.has(el.tagName))
181
233
  continue;
234
+ // Skip non-page artifacts that pollute the fingerprint: StyleProof's own
235
+ // injected hover-sink (added by a capture, so present when a state is
236
+ // captured in place but NOT on a fresh load) and framework route-announcers.
237
+ // Counting them made a state's in-place fingerprint differ from its
238
+ // reset+replay fingerprint, so every reset to depth >= 2 failed verification
239
+ // (and the same pollution split dedup, capturing one surface as two). They
240
+ // are not part of the page.
241
+ if (el.hasAttribute('data-styleproof-hover-sink'))
242
+ continue;
243
+ if (el.tagName === 'NEXT-ROUTE-ANNOUNCER' || el.id === '__next-route-announcer__')
244
+ continue;
182
245
  add(el);
183
246
  }
184
247
  return { shape: parts.join('\n'), elements: parts.length, classes: [...classes] };
@@ -243,15 +306,27 @@ async function settleDom(page, maxMs = 1200) {
243
306
  * non-trivial. Generic — no app-specific selector needed, so a bare crawl of a
244
307
  * Babel/React/Vue page that boots after `load` still captures the mounted UI. */
245
308
  async function waitSettled(page) {
246
- await page.waitForLoadState('networkidle').catch(() => { });
309
+ // DOM-growth settle is the PRIMARY readiness signal — it detects an
310
+ // async-mounted app (the DOM stops growing and is non-trivial) without an
311
+ // app-specific selector, and it also catches fetch-painted content (that
312
+ // grows the DOM). We deliberately do NOT gate on networkidle: it waits a
313
+ // 500ms idle window ON TOP of any lingering request, and a single cross-origin
314
+ // asset — a Google-Fonts stylesheet — keeps the network "busy" ~1s per load
315
+ // with no bearing on readiness. gotoFresh runs once per state (plus per
316
+ // retry), so that dominated crawl time (measured: ~995ms of every ~1531ms
317
+ // reset). Instead we wait for FONTS specifically — they ARE part of the
318
+ // computed style the diff compares, so they must be loaded before capture,
319
+ // but document.fonts.ready is the deterministic signal (and resolves from the
320
+ // page's cache on repeat loads, so it's ~free after the first).
247
321
  let prev = -1;
248
322
  for (let i = 0; i < 40; i++) {
249
323
  const n = await page.evaluate(() => document.body.getElementsByTagName('*').length);
250
324
  if (n > 5 && n === prev)
251
- return;
325
+ break;
252
326
  prev = n;
253
327
  await page.waitForTimeout(100);
254
328
  }
329
+ await page.evaluate(() => document.fonts.ready.then(() => true)).catch(() => { });
255
330
  }
256
331
  /**
257
332
  * Load the URL from a clean slate (storage cleared by the init script armed in
@@ -412,14 +487,43 @@ function stateKey(steps) {
412
487
  }
413
488
  /** Record a newly-found surface, capture it in place (page is already there), and
414
489
  * queue it for its own sweep. Streams progress via onSurface. */
415
- async function record(page, opts, newPath, depth, fp, st, retryOnly = false) {
490
+ async function record(page, opts, newPath, depth, fp, st, sink, retryOnly = false, viaRetry = false) {
416
491
  const key = deriveKey(newPath, st.used);
417
492
  const surface = { key, depth, path: newPath, elements: fp.elements };
418
493
  st.surfaces.push(surface);
419
- st.queue.push({ path: newPath, depth, sig: fp.sig, retryOnly });
494
+ let addsVocab = false;
420
495
  for (const c of fp.classes)
421
- st.classes.add(c);
422
- await captureAndReport(page, opts, surface, st);
496
+ if (!st.classes.has(c)) {
497
+ st.classes.add(c);
498
+ addsVocab = true;
499
+ }
500
+ // Children buffer in the sweep's sink and enter the shared queue only when the
501
+ // parent's sweep completes — family retry reads the parent's changer registry,
502
+ // which is only complete then. (Serial mode passes st.queue directly.)
503
+ //
504
+ // COVERAGE-GUIDED QUEUE PRUNING: in --until-covered mode, a surface that adds
505
+ // NO new render vocabulary is a structural repeat (the ninth agent's dossier,
506
+ // identical to the first) — don't queue it for a sweep, because its subtree is
507
+ // the same repeats and contributes nothing to coverage. This is what turns the
508
+ // exhaustive breadth-first crawl into a FAST coverage check: the queue drains
509
+ // to the few distinct components, each still drilled to depth (its own new
510
+ // vocabulary keeps it queued) via now-reliable resets. Base (depth 0) always
511
+ // seeds. Exhaustive mode (default) queues everything, so the surface-by-surface
512
+ // diff still sees every state.
513
+ const keep = !opts.stopWhenCovered || addsVocab || depth === 0;
514
+ if (keep) {
515
+ sink.push({ path: newPath, depth, sig: fp.sig, retryOnly, viaRetry });
516
+ }
517
+ // Skip the expensive style-map capture (getComputedStyle over every element,
518
+ // per width) for coverage-redundant surfaces: in --until-covered mode a surface
519
+ // that adds no vocabulary needs no map — its classes were already counted from
520
+ // the fingerprint above — so paying the capture is pure waste. This is the
521
+ // difference between a fast coverage check and re-capturing hundreds of
522
+ // structural repeats. Distinct (vocab-adding) surfaces are still captured, so
523
+ // the run leaves a usable map set behind.
524
+ if (keep) {
525
+ await captureAndReport(page, opts, surface, st);
526
+ }
423
527
  }
424
528
  /** Capture the current page as `surface` at every width and report the outcome. */
425
529
  async function captureAndReport(page, opts, surface, st) {
@@ -451,7 +555,7 @@ async function tryInPlace(page, c) {
451
555
  }
452
556
  /** Drive one candidate from where the page stands. Returns whether the page is
453
557
  * still in the swept state (no-op click) and whether the action was a skip. */
454
- async function driveCandidate(page, opts, entry, c, st) {
558
+ async function driveCandidate(page, opts, entry, c, st, sink, viaRetry, presentIds = new Set()) {
455
559
  // (retry-only lineage is inherited: a consumed state's descendants can also
456
560
  // only be mode-switch views, never fresh-candidate exploration.)
457
561
  const outcome = await tryInPlace(page, c);
@@ -468,7 +572,7 @@ async function driveCandidate(page, opts, entry, c, st) {
468
572
  .catch(() => false);
469
573
  const from = stateKey(entry.path);
470
574
  const list = st.changersFrom.get(from) ?? [];
471
- if (!list.some((x) => x.c.selector === c.selector))
575
+ if (!list.some((x) => x.c.identity === c.identity))
472
576
  list.push({ c, persists });
473
577
  st.changersFrom.set(from, list);
474
578
  const fp = await fingerprint(page);
@@ -482,7 +586,41 @@ async function driveCandidate(page, opts, entry, c, st) {
482
586
  reason: c.reason,
483
587
  ...(c.value ? { value: c.value } : {}),
484
588
  };
485
- await record(page, opts, [...entry.path, step], entry.depth + 1, fp, st, entry.retryOnly || !persists);
589
+ const childPath = [...entry.path, step];
590
+ const childRetryOnly = entry.retryOnly || !persists;
591
+ await record(page, opts, childPath, entry.depth + 1, fp, st, sink, childRetryOnly, viaRetry);
592
+ // DESCEND IN PLACE. We are standing in the surface this click just opened,
593
+ // reached by a reliable forward click. Sweep its OWN fresh controls now, while
594
+ // we are here, rather than leaving it for the queued entry to re-reach by
595
+ // reset+replay. In EXHAUSTIVE mode this is the win: it maps a deep branch (an
596
+ // expanded run inside an expanded automation) while the page is already there,
597
+ // instead of via a fragile deep replay. The queued entry still executes for
598
+ // family-retry (pairwise) coverage; global `tried` means its fresh list is
599
+ // already empty here, so this does not double-drive.
600
+ //
601
+ // But descent is DEPTH-FIRST: it drills the first candidate's whole subtree
602
+ // before the base sweep reaches the next candidate, so a page whose distinct
603
+ // vocabulary lives across many components (a roster of dossiers) covers slowly.
604
+ // In --until-covered mode we therefore SKIP the in-place descent and let the
605
+ // breadth-first queue + coverage-guided pruning reach every component fast and
606
+ // drill each once — now that resets are reliable, the queue reaches depth too.
607
+ if (!opts.stopWhenCovered && persists && entry.depth + 1 < opts.maxDepth) {
608
+ const child = {
609
+ path: childPath,
610
+ depth: entry.depth + 1,
611
+ sig: fp.sig,
612
+ retryOnly: childRetryOnly,
613
+ viaRetry,
614
+ };
615
+ // Exclude controls that were ALSO present in the parent (mode-switchers,
616
+ // shared chrome): those are breadth, owned by the queued family-retry.
617
+ // Descend only into controls genuinely NEW to this surface — a run row that
618
+ // exists only after the job expanded — which is exactly the deep nested UI
619
+ // the reset-replay path starves.
620
+ await sweepCandidatesHere(page, opts, child, st, sink, /* freshOnly */ true, presentIds).catch(() => {
621
+ /* fail-soft: the surface was already captured; deeper descent is best-effort */
622
+ });
623
+ }
486
624
  return { inState: false, skipped: false };
487
625
  }
488
626
  /**
@@ -494,32 +632,61 @@ async function driveCandidate(page, opts, entry, c, st) {
494
632
  /** The family-retry list for a state: its parent's persistent mode-switchers that
495
633
  * are visible right now — minus the step that created this state itself. */
496
634
  function familyRetries(entry, all, st) {
635
+ // RETRIES DO NOT COMPOUND: a state reached via a family retry still explores
636
+ // its genuinely-new UI (fresh selectors), but never re-retries mode-switchers.
637
+ // Every PAIRWISE mode combination is captured; N-way products of independent
638
+ // toggles are not walked — they multiply states without new render vocabulary,
639
+ // and anything class-visible only at 3-way depth is still NAMED by the
640
+ // coverage verifier.
641
+ if (entry.viaRetry)
642
+ return [];
497
643
  if (entry.path.length === 0)
498
644
  return [];
499
645
  const parentKey = stateKey(entry.path.slice(0, -1));
500
646
  const ownSelector = entry.path[entry.path.length - 1].selector;
501
- const visibleNow = new Set(all.map((c) => c.selector));
647
+ // Match registered changers to THIS state's candidates by semantic identity —
648
+ // the mode switch re-rendered the subtree, so positional selectors drifted;
649
+ // the current candidate carries the right selector for this state.
650
+ const byIdentity = new Map(all.map((c) => [c.identity, c]));
502
651
  return (st.changersFrom.get(parentKey) ?? [])
503
- .filter((x) => x.persists && x.c.selector !== ownSelector && visibleNow.has(x.c.selector))
504
- .map((x) => x.c);
652
+ .filter((x) => x.persists && x.c.selector !== ownSelector)
653
+ .map((x) => byIdentity.get(x.c.identity))
654
+ .filter((c) => Boolean(c));
505
655
  }
506
656
  /** The work list for one state's sweep: fresh controls first (already-driven
507
657
  * global chrome would otherwise starve a deep surface's own controls; the
508
658
  * throttle applies to fresh ones), then the parent's persistent mode-switchers
509
659
  * re-tried in THIS sibling mode. A state reached through a consuming action
510
- * collects NO fresh candidates — see QueueEntry.retryOnly. */
511
- function sweepWorkList(entry, all, opts, st) {
512
- const fresh = entry.retryOnly ? [] : all.filter((c) => !st.tried.has(c.selector)).slice(0, opts.maxActionsPerState);
660
+ * collects NO fresh candidates — see QueueEntry.retryOnly.
661
+ *
662
+ * `freshOnly` drops the family-retries: the in-place descent of a
663
+ * freshly-opened surface explores its own new UI (DEPTH), but must NOT re-apply
664
+ * the parent's mode-switchers — that is the pairwise BREADTH, owned by the
665
+ * queued sweep. Re-applying them while descending would compound modes into
666
+ * N-way products (the very thing "retries don't compound" prevents). */
667
+ function sweepWorkList(entry, all, opts, st, freshOnly = false, excludeIds = new Set()) {
668
+ const fresh = entry.retryOnly
669
+ ? []
670
+ : all.filter((c) => !st.tried.has(c.identity) && !excludeIds.has(c.identity)).slice(0, opts.maxActionsPerState);
513
671
  return [
514
672
  ...fresh.map((c) => ({ c, retry: false })),
515
- ...familyRetries(entry, all, st).map((c) => ({ c, retry: true })),
673
+ ...(freshOnly ? [] : familyRetries(entry, all, st).map((c) => ({ c, retry: true }))),
516
674
  ];
517
675
  }
518
- async function sweepState(page, opts, entry, st) {
519
- if (!(await resetToState(page, opts, entry.path, entry.sig)))
520
- return { tried: 0, skipped: 0 };
676
+ /** Drive one state's work list from where the page ALREADY stands (no reset). A
677
+ * no-op click leaves the page in the state and the loop continues; a
678
+ * state-changing click drives the child in place (see driveCandidate) and then
679
+ * a verified reset returns here before the next candidate. Split out from
680
+ * sweepState so driveCandidate can call it to DESCEND a freshly-opened surface
681
+ * in place — reaching that surface via a forward click is reliable, so its deep
682
+ * descendants are captured on the first visit instead of via a later reset. */
683
+ async function sweepCandidatesHere(page, opts, entry, st, sink, freshOnly = false, excludeIds = new Set()) {
521
684
  const all = await page.evaluate(collectClickable).catch(() => []);
522
- const work = sweepWorkList(entry, all, opts, st);
685
+ const work = sweepWorkList(entry, all, opts, st, freshOnly, excludeIds);
686
+ // Controls present HERE — passed to each child's in-place descent as its
687
+ // exclude set, so the descent skips this surface's mode-switchers/chrome and
688
+ // only drills genuinely new nested UI.
689
+ const presentIds = new Set(all.map((c) => c.identity));
523
690
  let tried = 0;
524
691
  let skipped = 0;
525
692
  let inState = true;
@@ -532,19 +699,24 @@ async function sweepState(page, opts, entry, st) {
532
699
  inState = true;
533
700
  }
534
701
  if (!retry)
535
- st.tried.add(c.selector);
702
+ st.tried.add(c.identity);
536
703
  if (c.unsafe) {
537
704
  skipped++;
538
705
  continue;
539
706
  }
540
707
  tried++;
541
- const r = await driveCandidate(page, opts, entry, c, st);
708
+ const r = await driveCandidate(page, opts, entry, c, st, sink, retry, presentIds);
542
709
  inState = r.inState;
543
710
  if (r.skipped)
544
711
  skipped++;
545
712
  }
546
713
  return { tried, skipped };
547
714
  }
715
+ async function sweepState(page, opts, entry, st, sink) {
716
+ if (!(await resetToState(page, opts, entry.path, entry.sig)))
717
+ return { tried: 0, skipped: 0 };
718
+ return sweepCandidatesHere(page, opts, entry, st, sink);
719
+ }
548
720
  /** Capture one synthetic data state of the entry page (its data requests stalled
549
721
  * or failed) in place, deduped and coverage-counted like any surface — but never
550
722
  * queued: a stalled app is not a state to crawl deeper from. */
@@ -578,6 +750,81 @@ async function recordDataState(page, opts, mode, st) {
578
750
  await page.unroute('**/*');
579
751
  }
580
752
  }
753
+ /**
754
+ * Sweep the queue with N concurrent workers, each on its own page. LIFO keeps
755
+ * the depth-first bias; a worker's discovered children enter the shared queue
756
+ * only when its sweep completes (see record). The surface SET matches a serial
757
+ * crawl — dedup sets are shared and mutated synchronously — only dup-key
758
+ * suffix attribution can vary with timing.
759
+ */
760
+ async function runPool(primary, opts, st, counters, defined = []) {
761
+ const target = Math.max(1, opts.workers ?? 1);
762
+ const pages = [primary];
763
+ while (opts.newPage && pages.length < target) {
764
+ const extra = await opts.newPage();
765
+ if (opts.resetStorage)
766
+ await armResetStorage(extra);
767
+ pages.push(extra);
768
+ }
769
+ // Coverage-oriented early stop (opt-in): the moment every defined class has
770
+ // been SEEN, stop — the rest of the queue is redundant for a coverage check.
771
+ // We do NOT plateau-stop on "N surfaces without a new class": a productive
772
+ // deep state (an automation whose expandable run row is its last candidate)
773
+ // adds its vocabulary only after many no-new siblings, so a plateau cuts the
774
+ // crawl off before that state's sweep even starts. Termination instead comes
775
+ // from the queue draining — and it drains fast because coverage-guided pruning
776
+ // (see record) never queues a structural repeat. `cover.every` runs only when
777
+ // the class count actually grows, so this is cheap.
778
+ const cover = opts.stopWhenCovered && defined.length > 0 ? defined : null;
779
+ let prevSize = st.classes.size;
780
+ let covered = false;
781
+ const converged = () => {
782
+ if (!cover)
783
+ return false;
784
+ if (!covered && st.classes.size > prevSize) {
785
+ prevSize = st.classes.size;
786
+ if (cover.every((c) => st.classes.has(c)))
787
+ covered = true;
788
+ }
789
+ return covered;
790
+ };
791
+ let active = 0;
792
+ await new Promise((resolve) => {
793
+ const pump = () => {
794
+ while (pages.length > 0 && st.queue.length > 0 && st.surfaces.length < opts.maxStates && !converged()) {
795
+ const entry = st.queue.shift(); // FIFO → breadth-first: exhaust every
796
+ // shallow surface (nav tabs, opened panels — where distinct UI lives)
797
+ // before drilling. Depth-first starves breadth: one append-generator
798
+ // branch drills to depth 20+ while sibling tabs sit unpopped, so real
799
+ // surfaces (an OAuth card, a skills grid) go uncaptured. Dedup is
800
+ // set-based, so order never changes WHAT is found — only that shallow
801
+ // is found first, which is what full coverage needs.
802
+ if (entry.depth >= opts.maxDepth)
803
+ continue;
804
+ const worker = pages.pop();
805
+ active++;
806
+ const sink = [];
807
+ sweepState(worker, opts, entry, st, sink)
808
+ .then((r) => {
809
+ counters.tried += r.tried;
810
+ counters.skipped += r.skipped;
811
+ })
812
+ .catch(() => {
813
+ /* fail-soft: the state's surface was already captured in place */
814
+ })
815
+ .finally(() => {
816
+ st.queue.push(...sink);
817
+ pages.push(worker);
818
+ active--;
819
+ pump();
820
+ });
821
+ }
822
+ if (active === 0 && (st.queue.length === 0 || st.surfaces.length >= opts.maxStates || converged()))
823
+ resolve();
824
+ };
825
+ pump();
826
+ });
827
+ }
581
828
  /** Depth-first discovery + in-place capture of every reachable surface. Depth-first
582
829
  * so a surface's OWN sub-states (a modal's tab → its toggles) are mapped while the
583
830
  * branch is fresh; with no budget, order affects time-to-depth, not coverage. */
@@ -617,7 +864,7 @@ async function discover(page, opts) {
617
864
  captured: 0,
618
865
  failed: [],
619
866
  };
620
- await record(page, opts, [], 0, fp, st, false);
867
+ await record(page, opts, [], 0, fp, st, st.queue, false, false);
621
868
  // Automatic data states of the entry page — the loading skeleton and the
622
869
  // error render exist in every data-driven app but almost never in a click
623
870
  // path. Captured out of the box; identical-to-base renders dedup away.
@@ -625,21 +872,13 @@ async function discover(page, opts) {
625
872
  await recordDataState(page, opts, 'loading', st);
626
873
  await recordDataState(page, opts, 'error', st);
627
874
  }
628
- let actionsTried = 0;
629
- let skipped = 0;
630
- while (st.queue.length > 0 && st.surfaces.length < opts.maxStates) {
631
- const entry = st.queue.pop(); // LIFO → depth-first
632
- if (entry.depth >= opts.maxDepth)
633
- continue;
634
- const r = await sweepState(page, opts, entry, st);
635
- actionsTried += r.tried;
636
- skipped += r.skipped;
637
- }
875
+ const counters = { tried: 0, skipped: 0 };
876
+ await runPool(page, opts, st, counters, defined);
638
877
  const missing = defined.filter((c) => !st.classes.has(c)).sort();
639
878
  return {
640
879
  surfaces: st.surfaces,
641
- actionsTried,
642
- skipped,
880
+ actionsTried: counters.tried,
881
+ skipped: counters.skipped,
643
882
  captured: st.captured,
644
883
  failed: st.failed,
645
884
  coverage: { defined: defined.length, rendered: defined.length - missing.length, missing },
@@ -650,19 +889,21 @@ async function discover(page, opts) {
650
889
  * natural termination by default. Returns the surfaces mapped (with the click-path
651
890
  * that reached each), how many actions were tried/skipped, and captured/failed.
652
891
  */
892
+ /** Clear storage before the app's code runs on EVERY load, so each gotoFresh is
893
+ * a clean slate in one navigation (no clear-then-reload round trip). */
894
+ async function armResetStorage(page) {
895
+ await page.addInitScript(() => {
896
+ try {
897
+ localStorage.clear();
898
+ sessionStorage.clear();
899
+ }
900
+ catch {
901
+ /* storage unavailable (e.g. file://) — ignore */
902
+ }
903
+ });
904
+ }
653
905
  export async function crawlAndCapture(page, opts) {
654
- if (opts.resetStorage) {
655
- // Clear storage before the app's code runs on EVERY load, so each gotoFresh is
656
- // a clean slate in one navigation (no clear-then-reload round trip).
657
- await page.addInitScript(() => {
658
- try {
659
- localStorage.clear();
660
- sessionStorage.clear();
661
- }
662
- catch {
663
- /* storage unavailable (e.g. file://) — ignore */
664
- }
665
- });
666
- }
906
+ if (opts.resetStorage)
907
+ await armResetStorage(page);
667
908
  return discover(page, opts);
668
909
  }
package/dist/describe.js CHANGED
@@ -18,7 +18,10 @@ const PALETTE = [
18
18
  ['pink', [236, 64, 122]],
19
19
  ];
20
20
  function parseColor(v) {
21
- const m = v.match(/rgba?\(\s*(\d+)[,\s]+(\d+)[,\s]+(\d+)(?:[,/\s]+([\d.]+))?\s*\)/i);
21
+ // Anchored: only a value that IS a colour parses. An embedded colour inside a
22
+ // gradient/shadow/url must not stand in for the whole value — that once made a
23
+ // report show the same "representative" rgba on both sides of a real diff.
24
+ const m = v.match(/^rgba?\(\s*(\d+)[,\s]+(\d+)[,\s]+(\d+)(?:[,/\s]+([\d.]+))?\s*\)$/i);
22
25
  if (!m)
23
26
  return null;
24
27
  return [Number(m[1]), Number(m[2]), Number(m[3]), m[4] === undefined ? 1 : Number(m[4])];
package/dist/report.d.ts CHANGED
@@ -71,4 +71,5 @@ export type ReportResult = {
71
71
  export declare function summarizeProps(props: PropChange[]): PropChange[];
72
72
  /** `div.who-grid`, `a.nav-cta`, `h3` — the semantic marker class, else the tag. */
73
73
  export declare function prettyLabel(p: string, cls: string): string;
74
+ export declare function excerptPair(before: string, after: string): [string, string];
74
75
  export declare function generateStyleMapReport(opts: ReportOptions): ReportResult;
package/dist/report.js CHANGED
@@ -292,7 +292,9 @@ const orderIdx = (p) => {
292
292
  return i === -1 ? PROP_ORDER.length : i;
293
293
  };
294
294
  function cleanVal(v) {
295
- let s = v.replace(/-?\d+\.\d+/g, (m) => String(Math.round(parseFloat(m) * 10) / 10));
295
+ // Verbatim by design: no rounding. Rounding once showed alpha 0.18 as 0.2
296
+ // and could erase a real 0.18→0.2 diff entirely via the no-op filter below.
297
+ let s = v;
296
298
  if (!s.includes('(')) {
297
299
  const toks = s.split(' ');
298
300
  if (toks.length > 1 && new Set(toks).size === 1)
@@ -512,11 +514,48 @@ function regionHeading(regionPaths, findings) {
512
514
  // A "no value here" marker renders as an em dash; colours render as `#hex` so the
513
515
  // table cell shows GitHub's live swatch.
514
516
  const cell = (v) => (isNonValue(v) ? '—' : `\`${toHex(v)}\``);
517
+ // Long values (gradients, data URIs) would swamp the table, but truncating each
518
+ // side independently can show two IDENTICAL cells for a real diff: both
519
+ // sides of a gradient rendered as the same rgba while the actual change — a
520
+ // dropped `0px` stop — was elsewhere in the string. Instead, trim the shared
521
+ // prefix/suffix and show each side's differing substring with a little context.
522
+ const EXCERPT_AT = 64; // both sides at or under this → show whole values
523
+ const EXCERPT_CTX = 12; // chars of shared context kept around the diff
524
+ const EXCERPT_MAX = 96; // hard cap per excerpt; the diff itself may be huge
525
+ export function excerptPair(before, after) {
526
+ if (before.length <= EXCERPT_AT && after.length <= EXCERPT_AT)
527
+ return [before, after];
528
+ let p = 0;
529
+ while (p < before.length && p < after.length && before[p] === after[p])
530
+ p++;
531
+ let s = 0;
532
+ const maxS = Math.min(before.length, after.length) - p;
533
+ while (s < maxS && before[before.length - 1 - s] === after[after.length - 1 - s])
534
+ s++;
535
+ const cut = (v) => {
536
+ const start = Math.max(0, p - EXCERPT_CTX);
537
+ let end = Math.min(v.length, v.length - s + EXCERPT_CTX);
538
+ if (end - start > EXCERPT_MAX)
539
+ end = start + EXCERPT_MAX;
540
+ return (start > 0 ? '…' : '') + v.slice(start, end) + (end < v.length ? '…' : '');
541
+ };
542
+ return [cut(before), cut(after)];
543
+ }
544
+ /** Before/After cells as a pair, so long values excerpt around their actual diff. */
545
+ function cellPair(before, after) {
546
+ if (isNonValue(before) || isNonValue(after))
547
+ return [cell(before), cell(after)];
548
+ const [b, a] = excerptPair(before, after);
549
+ return [`\`${toHex(b)}\``, `\`${toHex(a)}\``];
550
+ }
515
551
  function beforeAfterTable(rows) {
516
552
  return [
517
553
  '| Property | Before | After |',
518
554
  '| --- | --- | --- |',
519
- ...rows.map((r) => `| \`${r.prop}\` | ${cell(r.before)} | ${cell(r.after)} |`),
555
+ ...rows.map((r) => {
556
+ const [b, a] = cellPair(r.before, r.after);
557
+ return `| \`${r.prop}\` | ${b} | ${a} |`;
558
+ }),
520
559
  ];
521
560
  }
522
561
  // A brand-new element has no meaningful "before", so its resting style renders
@@ -546,10 +585,12 @@ function styleSection(styles, added) {
546
585
  function statesSection(states, added) {
547
586
  const rows = [];
548
587
  for (const st of states)
549
- for (const c of summarizeProps(st.props))
588
+ for (const c of summarizeProps(st.props)) {
589
+ const [b, a] = cellPair(c.before, c.after);
550
590
  rows.push(added
551
591
  ? `| \`:${st.state}\` | \`${c.prop}\` | ${cell(c.after)} |`
552
- : `| \`:${st.state}\` | \`${c.prop}\` | ${cell(c.before)} → ${cell(c.after)} |`);
592
+ : `| \`:${st.state}\` | \`${c.prop}\` | ${b} → ${a} |`);
593
+ }
553
594
  if (!rows.length)
554
595
  return [];
555
596
  return [
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "styleproof",
3
- "version": "3.4.0",
3
+ "version": "3.6.0",
4
4
  "description": "Catch every CSS change before it ships — review PRs and certify refactors by the browser's computed styles, not pixels. Works with any styling system.",
5
5
  "keywords": [
6
6
  "playwright",