Leave caller-supplied separators alone in the HTML passthrough

Rewriting a bare <br> boundary reached into blockquotes: the paragraph
matching is regex-based, not nesting-aware, and a <br> between two
paragraphs inside a blockquote is inline content the editor keeps
untouched (verified against lexxy v0.9.29). Insert the durable separator
only where paragraphs are genuinely adjacent, and leave separators the
caller supplied as they came.

Author: jorgemanrubia

internal/richtext/richtext.go

@@ -398,14 +398,16 @@ func MarkdownToHTML(md string) string {
 // distinction), so contiguous paragraphs from HTML input are treated as
 // separate paragraphs.
 //
-// A boundary already carrying an empty separator paragraph (<p><br></p> or
-// <p></p>) on either side is left untouched, so running this on Basecamp editor
-// output — or on its own output — is a no-op. A boundary separated only by bare
-// <br> tags is rewritten to the separator paragraph: the editor drops those on
-// the first edit, so leaving them would keep the spacing they express fragile.
-// Only directly adjacent <p> blocks are separated; anything else between them,
-// such as a heading, list, or attachment, already provides its own break and is
-// left alone.
+// The transform is byte-preserving apart from the inserted separators and is
+// idempotent: a boundary that already carries a separator — a bare <br> between
+// the paragraphs, or an empty separator paragraph (<p><br></p> or <p></p>) on
+// either side — is left untouched, so running it on already-separated content
+// (including Basecamp editor output) is a no-op. Separators the caller supplied
+// are left as they came: this matching is not nesting-aware, and a <br> the
+// caller put between two paragraphs inside a blockquote is legal inline content
+// that survives editing. Only directly adjacent <p> blocks are separated;
+// anything else between them, such as a heading, list, or attachment, already
+// provides its own break and is left alone.
 func insertParagraphSeparators(s string) string {
 	locs := reP.FindAllStringIndex(s, -1)
 	if len(locs) < 2 {
@@ -431,8 +433,8 @@ func insertParagraphSeparators(s string) string {
 
 		nextStart := locs[i+1][0]
 		gap := s[end:nextStart]
-		if !empty[i] && !empty[i+1] && isSeparatorGap(gap) {
-			b.WriteString(reBR.ReplaceAllString(gap, ""))
+		if !empty[i] && !empty[i+1] && strings.TrimSpace(gap) == "" {
+			b.WriteString(gap)
 			b.WriteString(paragraphSeparator)
 			cursor = nextStart
 		}
@@ -441,13 +443,6 @@ func insertParagraphSeparators(s string) string {
 	return b.String()
 }
 
-// isSeparatorGap reports whether the markup between two paragraph blocks holds
-// nothing but whitespace and bare <br> tags — i.e. whatever spacing it expresses
-// can be replaced by a separator paragraph.
-func isSeparatorGap(gap string) bool {
-	return strings.TrimSpace(reBR.ReplaceAllString(gap, "")) == ""
-}
-
 // isEmptyParagraph reports whether a <p>...</p> block has no visible content —
 // i.e. it is empty or contains only <br> tags and whitespace, including
 // non-breaking-space entities (&nbsp;, &#160;, &#xa0;) that rich text editors

internal/richtext/richtext_test.go

@@ -2241,15 +2241,23 @@ func TestMarkdownToHTMLInsertsParagraphSeparators(t *testing.T) {
 			expected: "<p>A</p>\n<p><br></p><p>B</p>",
 		},
 		{
-			name:     "bare br separator becomes an editor-durable separator paragraph",
+			name:     "caller-supplied bare br separator is left untouched",
 			input:    "<p>A</p><br><p>B</p>",
-			expected: "<p>A</p><p><br></p><p>B</p>",
+			expected: "<p>A</p><br><p>B</p>",
 		},
 		{
 			name:     "separator paragraph is left untouched (editor canonical)",
 			input:    "<p>A</p><p><br></p><p>B</p>",
 			expected: "<p>A</p><p><br></p><p>B</p>",
 		},
+		{
+			// A <br> between paragraphs nested in a blockquote is inline content,
+			// which the editor keeps — matching is not nesting-aware, so leaving
+			// caller-supplied separators alone is what protects it.
+			name:     "bare br inside a blockquote is left untouched",
+			input:    "<blockquote><p>A</p><br><p>B</p></blockquote>",
+			expected: "<blockquote><p>A</p><br><p>B</p></blockquote>",
+		},
 		{
 			name:     "empty paragraph separator is left untouched",
 			input:    "<p>A</p><p></p><p>B</p>",
@@ -2301,9 +2309,9 @@ func TestMarkdownToHTMLInsertsParagraphSeparators(t *testing.T) {
 			expected: "<p></p><p>A</p>",
 		},
 		{
-			name:     "mix of contiguous and bare-br boundaries both get separator paragraphs",
+			name:     "mix of separated and contiguous only fills the gap that lacks a separator",
 			input:    "<p>A</p><p>B</p><br><p>C</p>",
-			expected: "<p>A</p><p><br></p><p>B</p><p><br></p><p>C</p>",
+			expected: "<p>A</p><p><br></p><p>B</p><br><p>C</p>",
 		},
 	}