feat(lsp): markdown diagnostic messages proposal

[MariaSolOs]

Implementation proposal for microsoft/language-server-protocol#1905.

[mfussenegger]

Do I get this right that this takes the kind from the first diagnostic and then formats all messages accordingly?

What if some are text and others markdown? That could cause to wrong highlighting

[MariaSolOs]

Duh you're right. These diagnostics might not all have the same kind. I should instead set message_kind to markdown if any one of them has Markdown content.

[mfussenegger]

I should instead set message_kind to markdown if any one of them has Markdown content.

I think the highlighting should be restricted to the messages where message_kind is markdown. Highlighting plaintext as markdown can lead to incorrect styling. E.g. unquoted variables like _foo with a trailing _ later in the text would highlight the whole section.

[clason]

Agree; that's the whole point of declaring the format: knowing when to parse/render and when not to.

[MariaSolOs]

But then what if some diagnostics have message_kind set to markdown and others set to plaintext? They would all be displayed in the same floating window, so how do we highlight that?

I might be missing something here though.

[mfussenegger]

I'm not sure what's the best way to go about it, but we have to correlate diagnostics with the line numbers where the messages of them were rendered.

Or maybe the highlight application can be integrated directly into the loop here:

neovim/runtime/lua/vim/diagnostic.lua

Lines 1798 to 1812 in 4459e0c

	for j = 1, #message_lines do
	local pre = j == 1 and prefix or string.rep(' ', #prefix)
	local suf = j == #message_lines and suffix or ''
	table.insert(lines, pre .. message_lines[j] .. suf)
	table.insert(highlights, {
	hlname = hiname,
	prefix = {
	length = j == 1 and #prefix or 0,
	hlname = prefix_hl_group,
	},
	suffix = {
	length = j == #message_lines and #suffix or 0,
	hlname = suffix_hl_group,
	},
	})

Otherwise it would need to track the number of lines used per diagnostic. For example if there are 3 diagnostics, where the second uses markdown you'd have a mapping of:

• 1-diagnostic: rendered in line 1-4,
• 2-diagnostic: rendered in line 5-6,
• 3-diagnostic: rendered in line 7

And then the api.nvim_buf_add_highlight would only affect line 5-6

[mfussenegger]

#28412 might be useful for this(?)

[MariaSolOs]

Oh yes, that does look handy 👀

[MariaSolOs]

Note to self: Remove hardcoded protocol changes and use the generator script to pick up the upstream definitions, but we need microsoft/language-server-protocol#1910 to be merged first.

Done