Add support for indented HTML blocks

api_test/main.c

@@ -1255,6 +1255,104 @@ static void test_front_matter(test_batch_runner *runner) {
 #undef PARSE
 }
 
+static void test_html_block_blank_lines(test_batch_runner *runner) {
+#define PARSE(str, opts) cmark_parse_document(str, sizeof(str) - 1, opts)
+
+  static const char nested_html[] =
+      "<div>\n"
+      "\t<p>one</p>\n"
+      "\n"
+      "    <p>two</p>\n"
+      "</div>\n";
+
+  cmark_node *doc = PARSE(nested_html, CMARK_OPT_DEFAULT);
+  cmark_node *html = doc->first_child;
+  INT_EQ(runner, cmark_node_get_type(html), CMARK_NODE_HTML_BLOCK,
+         "indented HTML: default first node is HTML");
+  OK(runner, html->next != NULL,
+     "indented HTML: default parsing ends at the blank line");
+  cmark_node_free(doc);
+
+  doc = PARSE(nested_html, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  INT_EQ(runner, cmark_node_get_type(html), CMARK_NODE_HTML_BLOCK,
+         "indented HTML: flagged node is HTML");
+  STR_EQ(runner, cmark_node_get_literal(html), nested_html,
+         "indented HTML: consistent indentation retains blank lines");
+  OK(runner, html->next == NULL,
+     "indented HTML: consistent content remains one block");
+  cmark_node_free(doc);
+
+  static const char initially_blank[] =
+      "<div>\n"
+      "\n"
+      "\t<p>content</p>\n"
+      "</div>\n";
+
+  doc = PARSE(initially_blank, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  STR_EQ(runner, cmark_node_get_literal(html), initially_blank,
+         "indented HTML: indentation may be established after a blank line");
+  OK(runner, html->next == NULL,
+     "indented HTML: initially blank content remains one block");
+  cmark_node_free(doc);
+
+  static const char dedented[] =
+      "<div>\n"
+      "\tcontent\n"
+      "\n"
+      "outside\n";
+
+  doc = PARSE(dedented, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  STR_EQ(runner, cmark_node_get_literal(html), "<div>\n\tcontent\n\n",
+         "indented HTML: retained blank line belongs to preceding HTML");
+  INT_EQ(runner, cmark_node_get_type(html->next), CMARK_NODE_PARAGRAPH,
+         "indented HTML: dedented content terminates the block");
+  cmark_node_free(doc);
+
+  static const char inconsistent[] =
+      "<div>\n"
+      "\tcontent\n"
+      "\n"
+      "  less indented\n";
+
+  doc = PARSE(inconsistent, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  INT_EQ(runner, cmark_node_get_type(html->next), CMARK_NODE_PARAGRAPH,
+         "indented HTML: shallower indentation terminates the block");
+  cmark_node_free(doc);
+
+  static const char unindented[] =
+      "<div>\n"
+      "content\n"
+      "\n"
+      "more\n";
+
+  doc = PARSE(unindented, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  OK(runner, html->next != NULL,
+     "indented HTML: unindented content still ends at a blank line");
+  cmark_node_free(doc);
+
+  static const char custom_element[] =
+      "<slide-diagram>\n"
+      "  first\n"
+      "\n"
+      "    nested\n"
+      "</slide-diagram>\n";
+
+  doc = PARSE(custom_element, CMARK_OPT_HTML_BLOCK_BLANK_LINES);
+  html = doc->first_child;
+  STR_EQ(runner, cmark_node_get_literal(html), custom_element,
+         "indented HTML: type 7 blocks and deeper indentation are supported");
+  OK(runner, html->next == NULL,
+     "indented HTML: custom element remains one block");
+  cmark_node_free(doc);
+
+#undef PARSE
+}
+
 static void test_feed_across_line_ending(test_batch_runner *runner) {
   // See #117
   cmark_parser *parser = cmark_parser_new(CMARK_OPT_DEFAULT);
@@ -1487,6 +1585,7 @@ int main() {
   test_cplusplus(runner);
   test_safe(runner);
   test_front_matter(runner);
+  test_html_block_blank_lines(runner);
   test_feed_across_line_ending(runner);
   test_pathological_regressions(runner);
   source_pos(runner);

man/man1/cmark-gfm.1

@@ -62,6 +62,10 @@ be rendered as curly quotes, depending on their position.
 Parse language prefixes on inline code spans, such as
 \f[C]ruby:`Object.new`\f[], and expose the language identifier as code info.
 .TP 12n
+.B \-\-html-block-blank-lines
+Allow indented content in type 6 and 7 HTML blocks to continue across blank
+lines while its established indentation is preserved.
+.TP 12n
 .B \-\-unsafe
 Render raw HTML and potentially dangerous URLs.
 (Raw HTML is not replaced by a placeholder comment; potentially

readme.md

@@ -25,6 +25,8 @@ Compared with GitHub's `cmark-gfm`, CMarkly:
   as `CMARK_NODE_FRONT_MATTER` nodes.
 - Supports opt-in language prefixes for inline code, such as `` ruby:`Object.new` ``,
   through `CMARK_OPT_INLINE_CODE_INFO` and the code-info accessors.
+- Supports opt-in blank lines within consistently indented HTML content through
+  `CMARK_OPT_HTML_BLOCK_BLANK_LINES`.
 - Provides `cmark_node_clone` for independently cloning complete node trees,
   including supported extension metadata and footnote relationships. User data
   is intentionally not copied.

src/blocks.c

@@ -91,8 +91,8 @@ static CMARK_INLINE bool S_ends_on_current_line(cmark_parser *parser, cmark_node
          // similar to fenced code blocks.
          // Types 6-7 end at a blank line, so their last content line is
          // the previous line and they should NOT match here.
-         (S_type(b) == CMARK_NODE_HTML_BLOCK && b->as.html_block_type >= 1 &&
-          b->as.html_block_type <= 5) ||
+         (S_type(b) == CMARK_NODE_HTML_BLOCK && b->as.html_block.type >= 1 &&
+          b->as.html_block.type <= 5) ||
          // Single-line blocks: finalized on same line they started
          b->start_line == parser->line_number;
 }
@@ -1037,7 +1037,7 @@ static bool parse_code_block_prefix(cmark_parser *parser, cmark_chunk *input,
 static bool parse_html_block_prefix(cmark_parser *parser,
                                     cmark_node *container) {
   bool res = false;
-  int html_block_type = container->as.html_block_type;
+  int html_block_type = container->as.html_block.type;
 
   assert(html_block_type >= 1 && html_block_type <= 7);
   switch (html_block_type) {
@@ -1051,7 +1051,32 @@ static bool parse_html_block_prefix(cmark_parser *parser,
     break;
   case 6:
   case 7:
-    res = !parser->blank;
+    if (!(parser->options & CMARK_OPT_HTML_BLOCK_BLANK_LINES)) {
+      res = !parser->blank;
+    } else if (parser->blank) {
+      // Tentatively retain blank lines. The next nonblank line determines
+      // whether the HTML block continues:
+      res = true;
+    } else if (S_last_line_blank(container)) {
+      // Establish the content indentation lazily so a blank line may follow
+      // the opening tag. A non-indented line still terminates the block:
+      if (container->as.html_block.indent == 0 && parser->indent > 0) {
+        container->as.html_block.indent = parser->indent;
+      }
+
+      res = container->as.html_block.indent > 0 &&
+            parser->indent >= container->as.html_block.indent;
+    } else {
+      // Record the shallowest positive content indentation before a blank
+      // line. Deeper nested HTML may then continue without changing it:
+      if (parser->indent > 0 &&
+          (container->as.html_block.indent == 0 ||
+           parser->indent < container->as.html_block.indent)) {
+        container->as.html_block.indent = parser->indent;
+      }
+
+      res = true;
+    }
     break;
   }
 
@@ -1224,7 +1249,7 @@ static void open_new_blocks(cmark_parser *parser, cmark_node **container,
                                    input, parser->first_nonspace))))) {
       *container = add_child(parser, *container, CMARK_NODE_HTML_BLOCK,
                              parser->first_nonspace + 1);
-      (*container)->as.html_block_type = matched;
+      (*container)->as.html_block.type = matched;
       // note, we don't adjust parser->offset because the tag is part of the
       // text
     } else if (!indented && cont_type == CMARK_NODE_PARAGRAPH &&
@@ -1421,7 +1446,7 @@ static void add_text_to_container(cmark_parser *parser, cmark_node *container,
       add_line(container, input, parser);
 
       int matches_end_condition;
-      switch (container->as.html_block_type) {
+      switch (container->as.html_block.type) {
       case 1:
         // </script>, </style>, </pre>
         matches_end_condition =

src/cmark-gfm.h

@@ -798,6 +798,12 @@ char *cmark_render_latex_with_mem(cmark_node *root, int options, int width, cmar
  */
 #define CMARK_OPT_INLINE_CODE_INFO (1 << 19)
 
+/** Allow indented content in type 6 and 7 HTML blocks to continue across
+ * blank lines. The indentation established by the HTML content must be
+ * preserved after each blank line.
+ */
+#define CMARK_OPT_HTML_BLOCK_BLANK_LINES (1 << 20)
+
 /**
  * ## Version information
  */

src/main.c

@@ -65,6 +65,7 @@ void print_usage() {
   printf("  --full-info-string              Include remainder of code block info\n"
          "                                  string in a separate attribute.\n");
   printf("  --inline-code-info              Parse inline code language prefixes\n");
+  printf("  --html-block-blank-lines        Continue indented HTML across blank lines\n");
   printf("  --help, -h       Print usage information\n");
   printf("  --version        Print version\n");
 }
@@ -168,6 +169,8 @@ int main(int argc, char *argv[]) {
       options |= CMARK_OPT_FULL_INFO_STRING;
     } else if (strcmp(argv[i], "--inline-code-info") == 0) {
       options |= CMARK_OPT_INLINE_CODE_INFO;
+    } else if (strcmp(argv[i], "--html-block-blank-lines") == 0) {
+      options |= CMARK_OPT_HTML_BLOCK_BLANK_LINES;
     } else if (strcmp(argv[i], "--table-prefer-style-attributes") == 0) {
       options |= CMARK_OPT_TABLE_PREFER_STYLE_ATTRIBUTES;
     } else if (strcmp(argv[i], "--strikethrough-double-tilde") == 0) {

src/node.h

@@ -33,6 +33,11 @@ typedef struct {
   int8_t fenced;
 } cmark_code;
 
+typedef struct {
+  int type;
+  int indent;
+} cmark_html_block;
+
 typedef struct {
   int level;
   bool setext;
@@ -104,7 +109,7 @@ struct cmark_node {
     cmark_heading heading;
     cmark_link link;
     cmark_custom custom;
-    int html_block_type;
+    cmark_html_block html_block;
     int cell_index; // For keeping track of TABLE_CELL table alignments
     void *opaque;
   } as;