Merge pull request 'fix(prosemirror-markdown): hardBreak переживает round-trip (trailing/consecutive/table-cell) (#549)' (#553) from fix/549-hardbreak-md-canon into develop

Reviewed-on: #553

packages/prosemirror-markdown/src/lib/markdown-converter.ts

@@ -987,9 +987,14 @@ export function convertProseMirrorToMarkdown(
         // like a paragraph followed by a list don't collide into "line1- a".
         // Then collapse newlines and escape pipes so a cell containing "|" or a
         // line break cannot corrupt the surrounding GFM row.
+        // #549: a GFM pipe cell is single-line — the generic `  \n`/newline
+        // collapse below flattens a hardBreak into a space, dropping it. Convert
+        // the two-space hardBreak marker to `<br>` FIRST (GFM cells allow inline
+        // `<br>`, which re-imports to a hardBreak) so intra-cell breaks survive.
         return nodeContent
           .map(processNode)
           .join(" ")
+          .replace(/ {2}\n/g, "<br>")
           .replace(/\r?\n/g, " ")
           .replace(/\|/g, "\\|");
       }
@@ -1338,6 +1343,27 @@ export function convertProseMirrorToMarkdown(
         parts[i] = mathInlineHtml(nodes[i].attrs?.text || "");
       }
     }
+    // #549: position-aware hardBreak. The two-space `  \n` form (emitted by the
+    // processNode hardBreak case) only re-imports to a hardBreak when it is
+    // FOLLOWED by same-paragraph text. It is silently lost when the break is
+    // (a) the last inline child — the top-level `.trim()` strips the trailing
+    // `  \n`, and even a non-last paragraph's trailing break sits before the
+    // `\n\n` block gap — or (b) immediately followed by another hardBreak — the
+    // intervening whitespace-only line reads as a paragraph separator on
+    // re-parse, splitting the run. Emit `<br>` there instead: marked passes the
+    // inline HTML break through and generateJSON rebuilds a hardBreak in every
+    // position. A safe mid-paragraph break keeps the `  \n` form so the existing
+    // golden output stays byte-stable. Skipped inside footnote bodies, whose
+    // `^[…]` inline form collapses breaks to spaces (unchanged; already lossy).
+    if (!inFootnoteBody) {
+      for (let i = 0; i < nodes.length; i++) {
+        if (nodes[i]?.type !== "hardBreak") continue;
+        const next = nodes[i + 1];
+        if (!next || next.type === "hardBreak" || (parts[i + 1] ?? "") === "") {
+          parts[i] = "<br>";
+        }
+      }
+    }
     return parts.join("");
   };
 

packages/prosemirror-markdown/test/roundtrip-all-nodes.test.ts

@@ -295,3 +295,44 @@ describe('git-sync converter: lose-prone atoms keep their VALUES across a round
     expect(allText(src)).toContain('shared body');
   });
 });
+
+// ---------------------------------------------------------------------------
+// #549: hardBreak COUNT/position preservation across a pm -> md -> pm round
+// trip. The FIXTURES block above only proves a mid-paragraph hardBreak survives
+// as a node; these pin the three sub-cases whose `  \n` two-space form was
+// silently dropped on re-import (trailing / consecutive / table cell), plus a
+// mid-paragraph regression guard so the byte-stable `  \n` form is not lost.
+// ---------------------------------------------------------------------------
+describe('#549: hardBreak survives every round-trip position', () => {
+  const HB = { type: 'hardBreak' };
+  const countHardBreaks = (n: any): number => {
+    let c = n?.type === 'hardBreak' ? 1 : 0;
+    if (Array.isArray(n?.content)) for (const ch of n.content) c += countHardBreaks(ch);
+    return c;
+  };
+  const roundTripCount = async (d: any): Promise<number> =>
+    countHardBreaks(await markdownToProseMirror(convertProseMirrorToMarkdown(d)));
+
+  it('mid-paragraph break is still preserved (regression guard)', async () => {
+    expect(await roundTripCount(doc(P(T('a'), HB, T('b'))))).toBe(1);
+  });
+
+  it('(1) trailing break at the end of the last paragraph survives', async () => {
+    expect(await roundTripCount(doc(P(T('a'), HB)))).toBe(1);
+  });
+
+  it('(2) two consecutive breaks in one paragraph survive', async () => {
+    expect(await roundTripCount(doc(P(T('a'), HB, HB, T('b'))))).toBe(2);
+  });
+
+  it('(3) a break inside a table cell survives', async () => {
+    const table = doc({
+      type: 'table',
+      content: [
+        { type: 'tableRow', content: [{ type: 'tableHeader', content: [P(T('h'))] }] },
+        { type: 'tableRow', content: [{ type: 'tableCell', content: [P(T('a'), HB, T('b'))] }] },
+      ],
+    });
+    expect(await roundTripCount(table)).toBe(1);
+  });
+});