Strong text between strong+emphasis text

[chartinger]

I am currently doing experiments with Milkdown which uses remark in the background. When saving strong text inside a string and emphasis text, the resulting markdown cannot be parsed again the same way.

Using the output from Pandoc the resulting html does look different

Testcase:

  t.equal(
    micromark('***foo****bar****baz***'),
    '<p><strong><em>foo</em></strong><em>bar</em><strong><em>baz</em></strong></p>',
    'should support strong inside strong and emphasis'
  )

Output:

    expected: |-
      '<p><strong><em>foo</em></strong><em>bar</em><strong><em>baz</em></strong></p>'
    actual: |-
      '<p><em><strong>foo</strong>**bar</em><em><strong>baz</strong></em></p>'

So my question is: Is this a bug in the parser or is the markdown not valid according to the specs?

[wooorm]

We try to adhere to CommonMark. So pasting this input there: https://spec.commonmark.org/dingus/?text=***foo****bar****baz***. Seems that they yield something different then what micromark does (but it’s not your expected value either).

[wooorm]

See also: https://babelmark.github.io/?text=***foo****bar****baz***.

[chartinger]

Thanks, seems the correct test case for micromark according to commonmark should be:

  t.equal(
    micromark('***foo****bar****baz***'),
    '<p><em><strong>foo</strong></em><em>bar</em><em><strong>baz</strong></em></p>',
    'should support strong inside strong and emphasis'
  )

    expected: |-
      '<p><em><strong>foo</strong></em><em>bar</em><em><strong>baz</strong></em></p>'
    actual: |-
      '<p><em><strong>foo</strong>**bar</em><em><strong>baz</strong></em></p>'

[wooorm]

This might be related/the same as syntax-tree/mdast-util-from-markdown#15

[chartinger]

After a bit more investigation i wonder if equal parsing to commonmark is even possible with the current micromark parser strategy.

Extending the attention skip code with a special case for nesting parsed sections "solves" my testcase and the one from the linked issue

          const leftLength = events[open][1].end.offset - events[open][1].start.offset;
          const rightLength = events[index][1].end.offset - events[index][1].start.offset;
          const totalLength = leftLength + rightLength;

...

              (rightLength) % 3 && !(totalLength % 3) && (
                ((leftLength > rightLength) && ((events[open+1][1].type !== 'emphasis'))) ||
                ((leftLength < rightLength) && ((events[open+1][1].type !== 'strong')))
              )

but still fails with more crazy strings like *******foo****sdf*****var***** Without checking the code, it seems CommonMark prioritizes by *-count somehow compared to the location-based parsing from micromark.