test(#351): close the mark-attr coverage hole + reclassify table spans (review round 1)

F1 [WARNING] The 'no invisible coverage hole' guard enumerated only
schema.nodes, so MARK attributes silently escaped the value-fuzz completeness
check — link.internal/target/rel/class are never fuzzed and nothing flagged it,
and a new attributed mark would slip through. Added allSchemaMarkAttrKeys() plus a
MARK_ATTR_FUZZED / MARK_ATTR_ALLOWLIST registry and two tests: every schema mark
attr must be in exactly one set (a new one turns it red), and neither set may hold
a stale row.

F2 [WARNING] The ACCEPTED annotation misclassified table colspan/rowspan as
having 'no md representation'. They DO round-trip — a spanned cell makes the
converter emit the whole table as a raw <table> with colspan/rowspan, which the
tiptap parser reads back. They are frozen only because generating a
geometrically-valid spanned table is deferred PR-2 structural work (the flat
generator hardcodes span = 1), not a markdown limit. Reclassified them as
DEFERRED-BUG (distinct from ACCEPTED) so a maintainer does not read them as an
inherent limitation; colwidth / backgroundColor(Name) stay ACCEPTED (the
raw-<table> fallback drops them).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/prosemirror-markdown/test/generative/attr-arbitraries.ts

@@ -31,12 +31,11 @@
  *
  *       (a) ACCEPTED LIMITATION — the attribute has NO markdown representation,
  *           so the loss is inherent to targeting markdown, not a converter
- *           defect. GFM/CommonMark simply cannot encode it. These: `paragraph`/
- *           `heading` `indent`, `callout.icon`, `orderedList.type` (a/A/i
- *           markers), table `colspan`/`rowspan`/`colwidth`/`backgroundColor(Name)`
- *           (GFM tables are span-less/style-less). Each is tagged
- *           `// ACCEPTED:` inline. Freezing them is correct — there is nothing to
- *           preserve in the target format.
+ *           defect. These: `paragraph`/`heading` `indent`, `callout.icon`,
+ *           `orderedList.type` (a/A/i markers), table `colwidth` /
+ *           `backgroundColor(Name)` (dropped by the raw-<table> fallback). Each is
+ *           tagged `// ACCEPTED:` inline. Freezing them is correct — there is
+ *           nothing to preserve in the target format.
  *
  *       (b) PINNED BUG — the attribute IS representable in markdown but the
  *           converter drops it anyway (a real defect). These are NOT silently
@@ -46,6 +45,13 @@
  *           on accept-vs-fix (the epic guardrail reserves that call). These:
  *           `column.width` (parseFloat drops `%`), `orderedList.start` (non-1
  *           start renders as `1.`). Tagged `// PINNED-BUG:` inline.
+ *
+ *       (c) DEFERRED-BUG — representable AND round-trips, frozen only because the
+ *           flat generator can't yet build a valid instance. Table
+ *           `colspan`/`rowspan` round-trip via the raw-<table> fallback, but a
+ *           geometrically-valid spanned table is PR-2 structural work; the flat
+ *           generator hardcodes span = 1. Tagged `// DEFERRED-BUG:` inline so a
+ *           maintainer does not read them as an inherent limitation.
  *       - Several non-null-default attrs are MATERIALIZED on import but are not
  *         in canonicalize's KNOWN_DEFAULTS (`callout.type`, `status.color`,
  *         table `colspan`/`rowspan`, `columns.layout`/`widthMode`,
@@ -160,18 +166,26 @@ const OVERRIDES: Record<string, AttrPolicy> = {
   'callout.icon': { frozen: true }, // ACCEPTED: no md representation (dropped on export)
   'status.text': { noDefault: true, arb: phraseArb, degen: '' },
   'status.color': { always: true, arb: str('green', 'orange', 'red', 'blue', 'yellow', 'purple') },
-  // ── table cells — ACCEPTED: GFM tables cannot express spans / bg / colwidth ─
+  // ── table cells ────────────────────────────────────────────────────────────
+  // DEFERRED-BUG (not ACCEPTED): colspan/rowspan ARE representable and round-trip
+  // — a spanned cell makes the converter emit the whole table as a raw <table>
+  // with colspan/rowspan attrs (markdown-converter.ts tableToHtml), which the
+  // tiptap parser reads back. They are frozen only because generating a
+  // geometrically-valid spanned table is deferred STRUCTURAL work (the flat
+  // generator hardcodes colspan/rowspan = 1), NOT a markdown limitation.
   'tableCell.colspan': { always: true, frozen: true },
   'tableCell.rowspan': { always: true, frozen: true },
+  // ACCEPTED: colwidth / backgroundColor(Name) have no representation — the
+  // raw-<table> fallback (tableToHtml) drops them, so there is nothing to preserve.
   'tableCell.colwidth': { frozen: true },
   'tableCell.backgroundColor': { frozen: true },
   'tableCell.backgroundColorName': { frozen: true },
   'tableCell.align': { arb: str('left', 'center', 'right') },
-  'tableHeader.colspan': { always: true, frozen: true },
-  'tableHeader.rowspan': { always: true, frozen: true },
-  'tableHeader.colwidth': { frozen: true },
-  'tableHeader.backgroundColor': { frozen: true },
-  'tableHeader.backgroundColorName': { frozen: true },
+  'tableHeader.colspan': { always: true, frozen: true }, // DEFERRED-BUG (see tableCell.colspan)
+  'tableHeader.rowspan': { always: true, frozen: true }, // DEFERRED-BUG (see tableCell.rowspan)
+  'tableHeader.colwidth': { frozen: true }, // ACCEPTED: no representation
+  'tableHeader.backgroundColor': { frozen: true }, // ACCEPTED: no representation
+  'tableHeader.backgroundColorName': { frozen: true }, // ACCEPTED: no representation
   'tableHeader.align': { arb: str('left', 'center', 'right') },
   // ── details ──────────────────────────────────────────────────────────────
   'details.open': { always: true, arb: fc.constant(true) }, // boolean, default false
@@ -230,7 +244,7 @@ export function attrIsValueFuzzed(type: string, attr: string): boolean {
   return !!policyFor(type, attr, def).arb;
 }
 
-/** Every `type.attr` in the schema (excluding the auto `id`), sorted. */
+/** Every node `type.attr` in the schema (excluding the auto `id`), sorted. */
 export function allSchemaAttrKeys(): string[] {
   const keys: string[] = [];
   for (const type of Object.keys(schema.nodes)) {
@@ -242,6 +256,23 @@ export function allSchemaAttrKeys(): string[] {
   return keys.sort();
 }
 
+/**
+ * Every MARK attribute in the schema, keyed `mark:<name>.<attr>`, sorted. Marks
+ * are not driven by the node OVERRIDES table (they are fuzzed by the text
+ * generator, text-arbitraries.ts), so their value-fuzz coverage is tracked with a
+ * separate snapshot (see flat-roundtrip.property.test.ts) — without this the
+ * "no invisible coverage hole" guarantee would hold for node attrs only, letting a
+ * new mark attr slip through unfuzzed and unallowlisted.
+ */
+export function allSchemaMarkAttrKeys(): string[] {
+  const keys: string[] = [];
+  for (const [name, mark] of Object.entries(schema.marks)) {
+    const attrs = (mark.spec?.attrs ?? {}) as Record<string, unknown>;
+    for (const attr of Object.keys(attrs)) keys.push(`mark:${name}.${attr}`);
+  }
+  return keys.sort();
+}
+
 export type AttrMode = 'p1' | 'fuzz';
 
 /**

packages/prosemirror-markdown/test/generative/flat-roundtrip.property.test.ts

@@ -10,6 +10,7 @@ import { firstDivergence } from '../roundtrip-helpers.js';
 import {
   schema,
   allSchemaAttrKeys,
+  allSchemaMarkAttrKeys,
   attrIsValueFuzzed,
 } from './attr-arbitraries.js';
 import {
@@ -57,6 +58,33 @@ const ATTR_VALUE_FUZZ_ALLOWLIST = new Set<string>([
   'youtube.align', 'youtube.height', 'youtube.width',
 ]);
 
+// ── MARK attribute-value coverage ───────────────────────────────────────────
+// Marks are fuzzed by the text generator (text-arbitraries.ts markedTextRunArb),
+// not the node OVERRIDES table, so their value-fuzz coverage is tracked with this
+// separate registry — otherwise the "no invisible coverage hole" guarantee would
+// hold for node attrs only, and a new mark attr (or a new attributed mark) would
+// silently escape the fuzz set. Every schema mark attr must be in exactly one of:
+//   MARK_ATTR_FUZZED    — actually driven at a non-default value by the generator;
+//   MARK_ATTR_ALLOWLIST — deliberately not value-fuzzed, with a reason.
+const MARK_ATTR_FUZZED = new Set<string>([
+  'mark:link.href', // markedTextRunArb sets a random webUrl href
+  'mark:link.title', // ...and an optional letter-bearing title
+  'mark:highlight.color', // highlight mark carries a generated color
+  'mark:textStyle.color', // textStyle mark carries a generated color
+  'mark:comment.commentId', // comment anchor id (alphanumeric token)
+  'mark:comment.resolved', // comment resolved flag (rides only when true)
+]);
+const MARK_ATTR_ALLOWLIST = new Set<string>([
+  // link presentational/routing attrs: not part of the markdown link surface the
+  // converter emits (it round-trips href + title only), so there is no
+  // non-default value to assert here — a deferred concern for a link-specific
+  // fixture, not the flat generative pass.
+  'mark:link.internal',
+  'mark:link.target',
+  'mark:link.rel',
+  'mark:link.class',
+]);
+
 // Each run does a real convert + marked + jsdom parse (~ms). Give ample headroom
 // so the suite is deterministic regardless of parallel worker load (like the
 // sibling property file).
@@ -176,6 +204,31 @@ describe('#351 flat generative round-trip — completeness contract', () => {
       ).toBe(true);
     }
   });
+
+  it('every MARK attribute is value-fuzzed OR allowlisted (no invisible hole)', () => {
+    // The node guard above covers node attrs; marks are fuzzed by the text
+    // generator, so their coverage is tracked separately. A new mark attr (or a
+    // newly-attributed mark) that lands in neither set turns this red.
+    const unaccounted: string[] = [];
+    for (const key of allSchemaMarkAttrKeys()) {
+      if (!MARK_ATTR_FUZZED.has(key) && !MARK_ATTR_ALLOWLIST.has(key)) {
+        unaccounted.push(key);
+      }
+    }
+    expect(
+      unaccounted,
+      `these mark attrs are neither in MARK_ATTR_FUZZED nor MARK_ATTR_ALLOWLIST:\n  ${unaccounted.join(
+        '\n  ',
+      )}`,
+    ).toEqual([]);
+  });
+
+  it('the MARK fuzz/allowlist sets have no stale rows (every entry is a real schema mark attr)', () => {
+    const all = new Set(allSchemaMarkAttrKeys());
+    for (const key of [...MARK_ATTR_FUZZED, ...MARK_ATTR_ALLOWLIST]) {
+      expect(all.has(key), `stale mark-attr registry row: ${key}`).toBe(true);
+    }
+  });
 });
 
 describe('#351 flat generative round-trip — properties', () => {