Line wrapping inside Markdoc tags breaks parser when closing tag is on same line

[jlevy]

Summary

When Flowmark wraps long Markdoc/Markform tags across multiple lines, it can break Markdoc's parser if the closing tag ({% /tag %}) remains on the same line as the wrapped opening tag's closing delimiter (%}).

Environment

• Flowmark version: v0.6.1.dev13+0a5781e
• Markdoc version: @markdoc/markdoc (latest)

The Problem

Flowmark correctly identifies and preserves Markdoc tag delimiters ({% ... %}), but when wrapping long tags, the closing tag placement can trigger a Markdoc parser bug that corrupts the AST structure.

Input (single line - works correctly)

{% outer %}
{% tag1 attr1=1 attr2=2 %}{% /tag1 %}
{% tag2 %}{% /tag2 %}
{% /outer %}

AST result: tag1 and tag2 correctly at the same nesting level inside outer.

After Flowmark wrapping (broken)

{% outer %}
{% tag1 attr1=1
attr2=2 %}{% /tag1 %}
{% tag2 %}{% /tag2 %}
{% /outer %}

AST result: tag2 incorrectly nested deep inside tag1, phantom tags appear.

Workaround that works

{% outer %}
{% tag1 attr1=1
attr2=2 %}
{% /tag1 %}
{% tag2 %}{% /tag2 %}
{% /outer %}

When the closing tag is on its own line after a multi-line opening tag, parsing works correctly.

Root Cause

This is technically a Markdoc parser bug (the Markdoc spec explicitly allows multi-line tags), but it's triggered by a specific pattern that Flowmark produces.

The problematic pattern is:

{% tag attr=value
more_attrs %}{% /tag %}

Where %}{% /tag %} appears on the same line after a multi-line opening tag.

Suggested Fix

When Flowmark wraps a long Markdoc tag across multiple lines, it should place the closing tag on a new line:

Current behavior:

{% field kind="string" id="name" label="Name" required=true minLength=2
maxLength=50 placeholder="Enter your name" %}{% /field %}

Suggested behavior:

{% field kind="string" id="name" label="Name" required=true minLength=2
maxLength=50 placeholder="Enter your name" %}
{% /field %}

This maintains proper formatting while avoiding the Markdoc parser bug.

Reproduction

# Create test file
cat > test.md << 'TESTEOF'
{% form id="test" %}
{% group id="group1" %}
{% field kind="number" id="age" label="Age" role="user" required=true min=0 max=150 integer=true placeholder="25" %}{% /field %}
{% field kind="string" id="name" %}{% /field %}
{% /group %}
{% group id="group2" %}
{% field kind="url" id="website" %}{% /field %}
{% /group %}
{% /form %}
TESTEOF

# Format with flowmark
flowmark --auto test.md

# Parse with Markdoc and count groups (should be 2, but shows 1)
node --input-type=module -e "
import Markdoc from '@markdoc/markdoc';
import { readFileSync } from 'fs';
const ast = Markdoc.parse(readFileSync('test.md', 'utf-8'));
function countGroups(node) {
  let count = 0;
  if (node.type === 'tag' && node.tag === 'group') count++;
  for (const child of node.children || []) count += countGroups(child);
  return count;
}
console.log('Groups found:', countGroups(ast));
"

Impact

This affects any project using Markdoc-style tags (Markdoc, Markform, similar template systems) where tags have many attributes that exceed the line width limit.

[jlevy]

Update: More extensive Markdoc parser bug

After more testing, the issue is broader than initially reported. The Markdoc parser has problems with ANY newlines within or around tag structures:

Format	Result
{% tag %}{% /tag %} all on one line	OK
{% tag %}\n{% /tag %} closing on new line	BROKEN
{% tag attr=val\nattr2=val2 %}{% /tag %} wrapped opening	BROKEN
{% tag %}\n{% /tag %}\n\n{% tag2 %} with blank line	BROKEN

The only working format is everything on a single line.

Recommended fix for Flowmark: Treat {% ... %} as atomic/non-wrappable. The entire tag from {% to %} must stay on one line. For self-closing patterns like {% tag %}{% /tag %}, keep both on the same line regardless of length.

This means long Markdoc tags will exceed the line width limit, but that's preferable to broken parsing.