Fix lazy paragraph after definition

6 files changed, 93 insertions, 13 deletions

diff --git a/src/construct/document.rs b/src/construct/document.rs
index 4ef6acc..b438808 100644
--- a/src/construct/document.rs
+++ b/src/construct/document.rs
@@ -337,16 +337,65 @@ pub fn flow_end(tokenizer: &mut Tokenizer) -> State {
         state,
     );
 
-    let paragraph = matches!(state, State::Next(StateName::ParagraphInside))
-        || (!child.events.is_empty()
-            && child.events
-                [skip::opt_back(&child.events, child.events.len() - 1, &[Name::LineEnding])]
-            .name
-                == Name::Paragraph);
-
     tokenizer.tokenize_state.document_child_state = Some(state);
 
-    if child.lazy && paragraph && tokenizer.tokenize_state.document_paragraph_before {
+    // If we’re in a lazy line, and the previous (lazy or not) line is something
+    // that can be lazy, and this line is that too, allow it.
+    //
+    // Accept:
+    //
+    // ```markdown
+    //   | * a
+    // > | b
+    //     ^
+    //   | ```
+    // ```
+    //
+    // Do not accept:
+    //
+    // ```markdown
+    //   | * # a
+    // > | b
+    //     ^
+    //   | ```
+    // ```
+    //
+    // Do not accept:
+    //
+    // ```markdown
+    //   | * a
+    // > | # b
+    //     ^
+    //   | ```
+    // ```
+    let mut document_lazy_continuation_current = false;
+    let mut stack_index = child.stack.len();
+
+    // Use two algo’s: one for when we’re suspended or in multiline things
+    // like definitions, another (b) for when we fed the line ending and closed
+    // a)
+    while !document_lazy_continuation_current && stack_index > 0 {
+        stack_index -= 1;
+        let name = &child.stack[stack_index];
+        if name == &Name::Paragraph || name == &Name::Definition {
+            document_lazy_continuation_current = true;
+        }
+    }
+
+    // …another because we parse each “rest” line as a paragraph, and we passed
+    // a EOL already.
+    if !document_lazy_continuation_current && !child.events.is_empty() {
+        let before = skip::opt_back(&child.events, child.events.len() - 1, &[Name::LineEnding]);
+        let name = &child.events[before].name;
+        if name == &Name::Paragraph {
+            document_lazy_continuation_current = true;
+        }
+    }
+
+    if child.lazy
+        && tokenizer.tokenize_state.document_lazy_accepting_before
+        && document_lazy_continuation_current
+    {
         tokenizer.tokenize_state.document_continued =
             tokenizer.tokenize_state.document_container_stack.len();
     }
@@ -366,7 +415,8 @@ pub fn flow_end(tokenizer: &mut Tokenizer) -> State {
         }
         Some(_) => {
             tokenizer.tokenize_state.document_continued = 0;
-            tokenizer.tokenize_state.document_paragraph_before = paragraph;
+            tokenizer.tokenize_state.document_lazy_accepting_before =
+                document_lazy_continuation_current;
             // Containers would only be interrupting if we’ve continued.
             tokenizer.interrupt = false;
             State::Retry(StateName::DocumentContainerExistingBefore)
diff --git a/src/construct/gfm_task_list_item_check.rs b/src/construct/gfm_task_list_item_check.rs
index 62ff8aa..b30659a 100644
--- a/src/construct/gfm_task_list_item_check.rs
+++ b/src/construct/gfm_task_list_item_check.rs
@@ -81,7 +81,7 @@ pub fn start(tokenizer: &mut Tokenizer) -> State {
 /// ```
 pub fn inside(tokenizer: &mut Tokenizer) -> State {
     match tokenizer.current {
-        Some(b'\t' | b' ') => {
+        Some(b'\t' | b'\n' | b' ') => {
             tokenizer.enter(Name::GfmTaskListItemValueUnchecked);
             tokenizer.consume();
             tokenizer.exit(Name::GfmTaskListItemValueUnchecked);
diff --git a/src/tokenizer.rs b/src/tokenizer.rs
index 731b829..83514cb 100644
--- a/src/tokenizer.rs
+++ b/src/tokenizer.rs
@@ -142,8 +142,8 @@ pub struct TokenizeState<'a> {
     pub document_data_index: Option<usize>,
     /// Container exits by line number.
     pub document_exits: Vec<Option<Vec<Event>>>,
-    /// Whether the previous flow was a paragraph.
-    pub document_paragraph_before: bool,
+    /// Whether the previous flow was a paragraph or a definition.
+    pub document_lazy_accepting_before: bool,
     /// Whether this is the first paragraph (potentially after definitions) in
     /// a list item.
     /// Used for GFM task list items.
@@ -282,7 +282,7 @@ impl<'a> Tokenizer<'a> {
                 document_container_stack: vec![],
                 document_exits: vec![],
                 document_continued: 0,
-                document_paragraph_before: false,
+                document_lazy_accepting_before: false,
                 document_data_index: None,
                 document_child_state: None,
                 document_child: None,
diff --git a/tests/block_quote.rs b/tests/block_quote.rs
index 5c72fb9..be9da40 100644
--- a/tests/block_quote.rs
+++ b/tests/block_quote.rs
@@ -89,6 +89,18 @@ fn block_quote() {
     );
 
     assert_eq!(
+        micromark("> [\na"),
+        "<blockquote>\n<p>[\na</p>\n</blockquote>",
+        "should support lazy, definition-like lines"
+    );
+
+    assert_eq!(
+        micromark("> [a]: b\nc"),
+        "<blockquote>\n<p>c</p>\n</blockquote>",
+        "should support a definition, followed by a lazy paragraph"
+    );
+
+    assert_eq!(
         micromark(">"),
         "<blockquote>\n</blockquote>",
         "should support empty block quotes (1)"
diff --git a/tests/gfm_task_list_item.rs b/tests/gfm_task_list_item.rs
index 5db5d1b..66b646f 100644
--- a/tests/gfm_task_list_item.rs
+++ b/tests/gfm_task_list_item.rs
@@ -70,6 +70,9 @@ fn gfm_task_list_item() {
 
 + [X] With an upper case `x`
 
+* [
+] In a lazy line
+
 - [  ] With two spaces
 
 +  [x] Two spaces indent
@@ -163,6 +166,9 @@ EOL after:
 <li><input type="checkbox" disabled="" checked="" /> With an upper case <code>x</code></li>
 </ul>
 <ul>
+<li><input type="checkbox" disabled="" /> In a lazy line</li>
+</ul>
+<ul>
 <li>[  ] With two spaces</li>
 </ul>
 <ul>
diff --git a/tests/list.rs b/tests/list.rs
index fc416cc..bbba7cd 100644
--- a/tests/list.rs
+++ b/tests/list.rs
@@ -262,6 +262,18 @@ fn list() {
     );
 
     assert_eq!(
+        micromark("- [\na"),
+        "<ul>\n<li>[\na</li>\n</ul>",
+        "should support lazy, definition-like lines"
+    );
+
+    assert_eq!(
+        micromark("- [a]: b\nc"),
+        "<ul>\n<li>c</li>\n</ul>",
+        "should support a definition, followed by a lazy paragraph"
+    );
+
+    assert_eq!(
         micromark("- foo\n  - bar\n    - baz\n      - boo"),
         "<ul>\n<li>foo\n<ul>\n<li>bar\n<ul>\n<li>baz\n<ul>\n<li>boo</li>\n</ul>\n</li>\n</ul>\n</li>\n</ul>\n</li>\n</ul>",
         "should support sublists w/ enough spaces (1)"