fix: codeblock without "<" consumes extra char

README.md

@@ -7,48 +7,47 @@ well-formed; the _input_ (vimdoc) is secondary. The first step should always be
 to try to fix the input (within reason) rather than insist on a grammar that
 handles vimdoc's endless quirks.
 
-Notes
------
+Overview
+--------
 
 - vimdoc format "spec":
     - [:help help-writing](https://neovim.io/doc/user/helphelp.html#help-writing)
     - https://github.com/nanotee/vimdoc-notes
-- whitespace is intentionally captured in `(word)`, because it is often necessary to be
-  able to correctly layout vim help files (especially old/legacy).
-- `(codeblock)` is contained by `(line)` because `>` can start a code block at the end of a line.
-- `(column_heading)` is contained by `(line)` because `>` (to close
-  a `(codeblock)` can appear at the start of `(column_heading)`.
-- `h1` ("Heading 1"): `======` followed by text and optional `*tags*`.
-- `h2` ("Heading 2"): `------` followed by text and optional `*tags*`.
-- `h3` ("Heading 3"): only UPPERCASE WORDS, followed by optional `*tags*`.
+- whitespace is intentionally captured in all atoms, because it is often used
+  for "layout" and ascii art in legacy help files.
+- `block` is the main top-level node which contains `line` nodes.
+  - ends at blank line(s) or a line starting with `<`.
+- `line`:
+  - contains atoms (words, tags, taglinks, …)
+  - contains `codeblock` because `>` can start a codeblock at the end of a line.
+  - contains headings (`h1`, `h2`, `h3`) because `codeblock` terminated by
+    "implicit stop" (no terminating `<`) consumes blank lines, so `block` has
+    no way to end.
+  - contains `column_heading` because `<` (the `codeblock` terminating char)
+    can appear at the start of `column_heading`.
+- `codeblock`:
+  - contains `line` nodes which do not contain `word` nodes, it's just the full
+    raw text line including whitespace. This is somewhat dictated by its
+    "preformatted" nature; parsing the contents would require loading a "child"
+    language (injection). See [#2](https://github.com/neovim/tree-sitter-vimdoc/issues/2).
+  - the terminating `<` (and any following whitespace) is discarded (anonymous).
+- `h1` = "Heading 1": `======` followed by text and optional `*tags*`.
+- `h2` = "Heading 2": `------` followed by text and optional `*tags*`.
+- `h3` = "Heading 3": only UPPERCASE WORDS, followed by optional `*tags*`.
 
 Known issues
 ------------
 
-- `line_li` ("list item") is _experimental_. It doesn't support nesting yet and
-  it may not work well; you can treat it as a normal `line` for layout purposes.
-- `codeblock` ">" must not be preceded only by tabs, a space char is required (" >").
-  See `:help lcs-tab` for example. Currently the grammar doesn't enforce this.
-- `codeblock` terminated by an "implicit stop" (i.e. no terminating `<`)
-  consumes the first char of the terminating line, and continues the parent
-  `block`, preventing top-level forms like `h1`, `h2` from being recognized
-  until a blank line is encountered.
-- `line` in a `codeblock` does not contain `word` atoms, it's just the full
-  raw text line including whitespace. This is somewhat dictated by its
-  "preformatted" nature; parsing the contents would require loading a "child"
-  language (injection). See [#2](https://github.com/vigoux/tree-sitter-vimdoc/issues/2).
+- `line_li` ("list item") is experimental. It doesn't support nesting yet.
+- Spec requires that `codeblock` delimiter ">" must be preceded by a space
+  (" >"), not a tab. But currently the grammar doesn't enforce this. Example:
+  `:help lcs-tab`.
 - `url` doesn't handle _surrounding_ parens. E.g. `(https://example.com/#yay)` yields `word`
 - `url` doesn't handle _nested_ parens. E.g. `(https://example.com/(foo)#yay)`
-- Ideally `block_end` should consume the last block of the document _only_ if that
-  block is missing a trailing blank line or EOL ("\n").
-    - TODO: consider simply _not supporting_ docs without EOL?
-- Ideally `line_noeol` should consume the last line of the document _only_ if
-  that line is missing EOL ("\n").
-    - TODO: consider simply _not supporting_ docs without EOL?
 
 TODO
 ----
 
 - `line_noeol` is a special-case to support documents that don't end in EOL.
-  Grammar could be a bit simpler if we just require EOL at end of document.
-- `line_modeline` (only at EOF)
+  Grammar could be simpler if we require EOL at end of document.
+- `line_modeline` ?

corpus/arguments.txt

@@ -43,7 +43,10 @@ NOT an argument
     (line
       (argument
         (word)
-        (ERROR))
+        (MISSING "}"))
+      (word)
+      (argument
+        (word))
       (word)
       (codespan
         (word))

corpus/codeblock.txt

@@ -28,7 +28,8 @@ block3:
       (word))
     (line
       (codeblock
-        (line)))
+        (line))))
+  (block
     (line
       (word)))
   (block
@@ -92,17 +93,28 @@ text
       (word))))
 
 ================================================================================
-codeblock with implicit stop (FIXME)
+codeblock with implicit stop
 ================================================================================
 >
   line1
   line2
 
--------------------------------
+===============================
 h1-headline *foo*
+line1
+
+line2
+
+>
+  line1
 
 -------------------------------
-h1-headline *foo*
+h2-headline *foo*
+
+>
+  line1
+
+H3 HEADLINE *foo*
 
 --------------------------------------------------------------------------------
 
@@ -114,15 +126,35 @@ h1-headline *foo*
         (line)
         (line)))
     (line
-      (word))
+      (h1
+        (word)
+        (tag
+          (word))))
     (line
-      (word)
-      (tag
-        (word))))
-  (h2
-    (word)
-    (tag
-      (word))))
+      (word)))
+  (block
+    (line
+      (word)))
+  (block
+    (line
+      (codeblock
+        (line)
+        (line)))
+    (line
+      (h2
+        (word)
+        (tag
+          (word)))))
+  (block
+    (line
+      (codeblock
+        (line)
+        (line)))
+    (line
+      (h3
+        (uppercase_name)
+        (tag
+          (word))))))
 
 ================================================================================
 codeblock with empty lines
@@ -155,7 +187,9 @@ x
         (line)
         (line)
         (line)
-        (line)))))
+        (line)))
+    (line
+      (word))))
 
 ================================================================================
 tricky codeblock
@@ -166,7 +200,17 @@ tricky codeblock
   < line3
 <
 
+    Example: >
+
+        vim.spell.check()
+        -->
+        {
+            {'quik', 'bad', 4}
+        }
+<
+
 tricky
+
 --------------------------------------------------------------------------------
 
 (help_file
@@ -176,6 +220,16 @@ tricky
         (line)
         (line)
         (line))))
+  (block
+    (line
+      (word)
+      (codeblock
+        (line)
+        (line)
+        (line)
+        (line)
+        (line)
+        (line))))
   (block
     (line
       (word))))
@@ -243,3 +297,51 @@ To test for a non-empty string, use empty(): >
       (word)
       (codeblock
         (line)))))
+
+================================================================================
+codeblock stop and start on same line
+================================================================================
+    Examples: >
+        :lua vim.api.nvim_command('echo "Hello, Nvim!"')
+<    LuaJIT: >
+        :lua =jit.version
+<
+    *:lua-heredoc*
+:lua << [endmarker]
+{script}
+
+    Example: >
+        lua << EOF
+        EOF
+<
+
+--------------------------------------------------------------------------------
+
+(help_file
+  (block
+    (line
+      (word)
+      (codeblock
+        (line))))
+  (block
+    (line
+      (word)
+      (codeblock
+        (line))))
+  (block
+    (line
+      (tag
+        (word)))
+    (line
+      (word)
+      (word)
+      (word))
+    (line
+      (argument
+        (word))))
+  (block
+    (line
+      (word)
+      (codeblock
+        (line)
+        (line)))))

corpus/codespan.txt

@@ -46,11 +46,10 @@ an error`.
       (word))))
 
 ================================================================================
-NOT a codespan
+NOT codespan
 ================================================================================
-						*'* *'a* *`* *`a*
-'{a-z}  `{a-z}		Jump to the mark.
-						*g'* *g'a* *g`* *g`a*
+				*'* *'a* *`* *`a*
+				*g'* *g'a* *g`* *g`a*
 g'{mark}  g`{mark}
 
 --------------------------------------------------------------------------------
@@ -66,14 +65,6 @@ g'{mark}  g`{mark}
         (word))
       (tag
         (word)))
-    (ERROR)
-    (line
-      (argument
-        (word))
-      (word)
-      (word)
-      (word)
-      (word))
     (line
       (tag
         (word))

corpus/heading1_2.txt

@@ -15,20 +15,24 @@ Text2
 --------------------------------------------------------------------------------
 
 (help_file
-  (h1
-    (word)
-    (tag
-      (word)))
+  (block
+    (line
+      (h1
+        (word)
+        (tag
+          (word)))))
   (block
     (line
       (word)))
-  (h2
-    (word)
-    (word)
-    (tag
-      (word))
-    (tag
-      (word)))
+  (block
+    (line
+      (h2
+        (word)
+        (word)
+        (tag
+          (word))
+        (tag
+          (word)))))
   (block
     (line
       (word))))
@@ -50,19 +54,23 @@ Text
 --------------------------------------------------------------------------------
 
 (help_file
-  (h1
-    (tag
-      (word))
-    (word)
-    (word))
+  (block
+    (line
+      (h1
+        (tag
+          (word))
+        (word)
+        (word))))
   (block
     (line
       (word)))
-  (h2
-    (tag
-      (word))
-    (word)
-    (word))
+  (block
+    (line
+      (h2
+        (tag
+          (word))
+        (word)
+        (word))))
   (block
     (line
       (word))))