[rchl]

Discussion and context in #2928

Fixes #2928

[rchl]

Well, it would regress clangd so maybe we should instead make it more specific.

Before

After

Payload

{
  "contents": {
    "kind": "markdown",
    "value": "### instance-method `assign`\n\nprovided by `<vector>`\n\n---\n→ `void`\n\nParameters:\n\n- `_InputIterator __first`\n- `_InputIterator __last`\n\n---\n```cpp\n// In vector<char>\npublic: template <\n    class _InputIterator,\n    __enable_if_t<\n        __has_exactly_input_iterator_category<_InputIterator>::value &&\n            is_constructible<value_type, typename iterator_traits<\n                                             _InputIterator>::reference>::value,\n        int> = 0>\nvoid assign(_InputIterator __first, _InputIterator __last)\n```"
  },
  "range": {
    "end": {
      "character": 12,
      "line": 17
    },
    "start": {
      "character": 6,
      "line": 17
    }
  }
}

[rchl]

We can add a workaround inside LSP-clangd itself. Probably something like result = re.sub('[ ]{2,}\n-', '\n\n-', result)

[rwols]

Why does the code flow not use the CommonMark parser from mdpopups?

[jwortmann]

Why does the code flow not use the CommonMark parser from mdpopups?

It does:

LSP/plugin/core/views.py

Lines 601 to 602 in 5561bfa

	if MARKO_MD_PARSER_VERSION:
	frontmatter["markdown_parser"] = "marko"

But only if supported by the mdpopups version (i.e. on ST dev build).

[rwols]

So: it should be auto-fixed when MARKO_MD_PARSER_VERSION becomes true-ish for the majority of users. Then we can remove the workaround.

I will revert this because it breaks my hover info.

[rchl]

Wait. I said that I will add this workaround to LSP-clangd instead, where it belongs.

[rchl]

And please follow our convention and only push any meaningful changes to main through a PR.

[rwols]

Not all users use LSP-* helper packages... so that would only help for a subset of the users.

[rwols]

And please follow our convention and only push any meaningful changes to main through a PR.

OK, I'll keep it in mind.

[rchl]

Not all users use LSP-* helper packages... so that would only help for a subset of the users.

I think this is a pretty weak argument. We don't have numbers of course but why would one not uses dedicated package for it given that it provides all the extra convenience like documented settings and extra features? And also that's the documented way to set up clangd according to our instructions.

It doesn't even use managed server by default so those who have problem with that shouldn't complain either.

I think that if someone is in the minority that is not using that package should also not mind this minor rendering inconvenience.

I don't want to create a conflict about this now but I really think you should revert your revert. It's an issue in clangd so we shouldn't create incorrect and non-compliant workarounds that break rendering in other servers.